KTZ is rumored to be planning a launch for its Hong Kong IPO as early as October, seeking to raise up to $5 billion to build new routes for western-bound Chinese freight

Key Takeaways:

• Rail operator Kazakhstan Temir Zholy is reportedly preparing to launch its Hong Kong IPO as early as October, after filing its initial prospectus in June
• Cross-border cargo accounted for only 10% of the company's freight volume in 2025, but contributed 33.6% of its freight revenue

By Lee Shih Ta

Freight trains traveling from China to the West often pass through Kazakhstan on their way to destinations across Eurasia, making the Central Asian nation an important link on this modern steel version of the Silk Road. Positioned at that crossroad, Kazakhstan Temir Zholy (KTZ), the country’s rail operator, is now looking eastward to finance its Eurasian overland transport dreams by tapping the Hong Kong capital market.

Specifically, the company is planning to raise up to $5 billion through a listing on the Hong Kong Stock Exchange, and could launch the IPO as early as October after filing its initial prospectus earlier this year, according to Kazakh media reports last week.

Saltanat Satzhan, a managing director at KTZ's parent, the Kazakh sovereign wealth fund Samruk-Kazyna, previously told the South China Morning Post that KTZ aims to complete the Hong Kong listing before the end of this year.

Wholly owned by Samruk-Kazyna, KTZ operates approximately 16,000 kilometers of railway in Kazakhstan, connecting China, Central Asia, Russia, and the Caspian Sea region. Its network facilitates the movement of a wide range of goods between China and Eurasia, ranging from finished products, to commodities like coal, mineral ores, grain and petroleum products. The company transported 320 million tons of cargo last year, with the freight business accounting for nearly 90% of its revenue.

KTZ’s transit freight operations are likely to emerge as a crown jewel as it prepares to go public, and thus merit particular scrutiny. While this category accounted for just 10% of its freight volume in 2025, it generated 33.6% of freight revenue. Longer shipping distances for transit cargo and exemption from domestic Kazakh tariffs is partly to credit for the segment’s relatively large revenue contribution compared to its volume share. The company's transit freight volume grew from 27.3 million tons in 2023 to 33 million tons in 2025. The figure could continue to climb at a similarly brisk pace if cross-border cargo flows continue to climb.

 ‘Middle Corridor’ prospects

As the world’s biggest manufacturer and one of its largest commodities consumers, China is pivotal to KZT’s expansion blueprint. KTZ operates the Khorgos Gateway dry port and is also has logistics operations in the Western Chinese city of Xi’an and the Eastern port city of Lianyungang.

In its prospectus filed in June, the company proposed funneling a portion of its IPO proceeds into constructing its planned 272-kilometer Bakhty-Ayagoz railway line, which would add another border crossing into China. For the company, the new line’s value hinges on whether the dry port, tracks, and transportation services can generate a surge in freight volume, since building extra capacity doesn’t guarantee it will get used.

Industry trends certainly provide a foundation for KTZ's expansion plan. A 2023 World Bank study noted that the Trans-Caspian International Transport Route, or “Middle Corridor,” connecting China and Europe via Central Asia and the Caucasus, has the potential to triple its freight volumes and halve travel times by 2030 if necessary policy improvements and investments are made. However, cross-border cargo volumes remain constrained by the efficiency of ports, customs and railways along the route.

Data from KTZ's prospectus also shows that not all Eurasian freight routes are expanding in tandem. The volume of transit containers traveling between China and the EU via Kazakhstan dropped by 31% from roughly 662,000 twenty-foot equivalent units (TEUs) in 2021 to about 458,000 in 2025, according to the document. While this doesn’t necessarily translate to a contraction in KTZ's overall transit business, it illustrates that freight flows shift with trade patterns and geopolitical dynamics. How much cargo the new crossing can capture will still depend on actual transit times, costs, and route stability.

The company's total revenue and other income climbed 27.4% last year to 2.76 trillion tenge ($6.26 billion). Its net profit for 2025 more than doubled to 343.6 billion tenge, as its gross margin climbed to 30.2% from 25.1% the previous year. The company said the growth was fueled by higher transit volumes, upward revisions in regulated tariffs, and favorable commodity market conditions. Notably, Kazakhstan's regulated freight tariffs were raised by about 24% and 28% in 2024 and 2025, respectively.

Funding pressures

Behind its strong growth, KTZ still faces fierce competition from rival Eurasian transport routes. Shippers can opt for a northern route through Russia, the Middle Corridor across the Caspian Sea, or simply default to maritime shipping. Their choices ultimately hinge on freight rates, delivery times and route reliability.

KTZ’s routes also present a unique logistical challenge because Kazakhstan and China operate on different railway gauges, meaning cargo must be transloaded at the border. Therefore, even after expanding its domestic capacity, KTZ must continue to rely on its neighbors for tracks outside its borders, as well as seamless coordination of border crossings to successfully capture a larger share of Chinese freight flows.

Meanwhile, its ongoing expansion is also testing KTZ's finances. By the end of 2025, the company's net current liabilities stood at 963 billion tenge, with total liabilities swelling to 5.21 trillion tenge. For the year, its operating activities generated 746.8 billion tenge in cash, whereas investing activities consumed 932.3 billion tenge.

KTZ projects that its capital expenditures from 2026 to 2030 will exceed those of the previous five-year period, with roughly 61% of planned investments slated for funding through external borrowing. Aside from constructing the new border line into China, the company plans to use its IPO proceeds for digitalization initiatives and paying down a portion of its interest-bearing debt.

Determining its valuation will be the final hurdle KTZ faces on its road to a Hong Kong listing. The company is reportedly angling for a valuation of up to $30 billion, though some analysts believe the actual figure will fall short of that mark. A $30 billion valuation, combined with the company’s net profit of roughly $700 million in 2025, translates to a price-to-earnings (P/E) ratio north of 40 times, a goal that may prove overly aggressive.

On the whole, KTZ boasts a relatively unique railway network that is difficult to replicate, firmly positioning it to capitalize on the growing flow of goods between China and Eurasia. Its outlook will ultimately depend on whether its newly added capacity can attract sufficient cargo and revenue to justify the high valuation it’s seeking, while paying for its construction investments, and servicing its debt load.

To subscribe to Bamboo Works weekly free newsletter, click here

The Shanghai-listed company has filed for a Hong Kong IPO, reporting it swung to a loss this year as it failed to fully pass on rising material costs to its customers

Key Takeaways:

• Dynamic Electronics has filed to list in Hong Kong, reporting it fell into the red in the first half of this year, even as its revenue jumped by 33%
• The PCB maker is getting pressured by steadily rising material costs, especially for copper, which caused its gross margin to plunge this year

By Bai Xin Rui

Within the hardware realm, printed circuit board (PCB) makers and their suppliers have been major beneficiaries of the AI boom, despite their traditional positioning as an anonymous but important player in the computing space. While many such hardware stocks viewed as AI beneficiaries pulled back sharply in July, overall sentiment toward the group remains relatively strong.

Seizing on that momentum, Dynamic Electronics Co. Ltd. (603175.SH), which is already listed in Shanghai, filed for a second listing in Hong Kong last week, even as rising material costs fueled by the AI explosion pushed it into the red this year.

Founded in 2015, Dynamic Electronics focuses on PCBs, which have wide applications across sectors including automotive electronics, data storage, consumer electronics and communications. The company’s emphasis is on automotive electronic PCBs, which accounted for 51.4% of its revenue in the first half of 2026.

Star-studded clientele

Dynamic Electronics currently ranks sixth globally in automotive PCBs and has captured the global top spot in automotive high-density interconnect (HDI) boards. Its clients include such marquee names as EV pioneer Tesla, auto parts giant Bosch, storage device maker Western Digital, and memory chip giant SK Hynix.

A PCB is a functional board constructed with predefined circuit patterns on a copper-clad laminate or insulating substrate. It is designed to connect electronic components and facilitate their signal transmissions. Serving as the fundamental interconnection for many electronic devices, a PCB's performance typically hinges on its layer count, circuit density, material properties, reliability and end-use applications.

The global PCB industry remains highly fragmented, with the top 10 manufacturers accounting for just 37.7% of the market last year. Competition has moved beyond a simple capacity arms race to a more comprehensive battle over technological thresholds, yield management, client certification and global supply chain capabilities.

According to third-party market data in Dynamic Electronics’ listing document, the global PCB market is expected to grow about 9.1% annually from 2026 to 2030, fueled by steady advances in technology and an explosion in downstream demand.

The PCB product spectrum encompasses HDI boards, multilayer boards, single- and double-sided boards, flexible printed circuits (FPCs) and packaging substrates. HDI boards use microvia, blind and buried vias and sequential build-up structures to achieve denser interconnection within constrained spaces. They are mostly used in compact designs and for high-speed signal transmission, as well as in high-end applications such as domain controllers for autonomous driving and smart cockpits, memory modules, GPU accelerator carrier boards, and AI computing boards.

Meanwhile, multilayer boards feature multiple conductive layers separated by insulating materials and compressed into a single circuit board. Such architecture provides extra wiring space, supporting more complex circuit layouts, power distribution, and signal transmission. Current product tiers include 4- to 6-layer, 8- to 14-layer, and 16-layer and higher PCBs, which are used in vehicle control systems, storage modules, communication equipment and AI computing boards.

The average selling price of multilayer boards is lower than that of HDI boards. According to Dynamic Electronics' listing application, the average price for one of its multilayer boards stood at 1,351 yuan in the first half of 2026, up 19.6% year-over-year. HDI boards fetched an average of 2,833 yuan, up 14.8% annually. Together, multilayer and HDI boards accounted for 45.4% and 43.8% of the company’s revenue, respectively, accounting for nearly 90% of the total in the first half of 2026.

Driven by strong demand from a wide range of applications across automotive electronics, data storage, communications and AI servers, Dynamic Electronics' overall average selling price rose by 27% to 1,684 yuan in the first half of 2026. This helped to lift its revenue by 33.3% to 2.91 billion yuan (326 million) during the six-month period.

But the company began to bleed red ink this year, swinging to a net loss of 191 million yuan in the first half from a profit a year earlier. It attributed the reversal to a net foreign exchange loss of 171 million yuan in the half-year period.

Rising copper prices

While foreign exchange losses played a part in the swing to the red, Dynamic Electronics also suffered a steep profitability squeeze as its gross margin plummeted 6.8 percentage points in the first half to 15.2% from 22% a year earlier. That drastic slide was primarily the result of surging raw material prices alongside rising freight, labor, and other production-related costs.

Raw materials are the biggest item in Dynamic Electronics' cost of sales, representing between 55% and 57% of the total in the last year and a half. Key materials include copper-clad laminates, prepregs, copper foils and copper balls or powder. The company indicated that its inability to promptly pass on these raw material price hikes to customers was the key factor pressuring its gross margin this year.

What’s more, the trend of spiking copper prices looks unlikely to ease anytime soon. London copper futures have touched multiple record highs this year alone, briefly nearing $15,000 per ton — up over 18% year-on-year. With global copper mining capacity on a broader downtrend and aging European and U.S. power grids driving demand for copper wire, demand for the metal is expected to keep growing. That means continued inability to promptly pass on those price hikes to its clients could continue to weigh on Dynamic Electronics’ profitability.

In a sign of shifting sentiment towards these companies, shares of Kinwong Electronic (3228.HK; 603228.SH), which specializes in automotive PCBs, initially fell in their Hong Kong trading debut on Tuesday, before recovering to trade up by more than 10% in the afternoon.

On the whole, Dynamic Electronics looks well-positioned to reap dividends on booming business from the AI and smart vehicle industries, giving it a solid growth narrative. Still, its massive margin compression and resulting near-term losses should serve as a cautionary signal for investors looking for the latest hot AI stocks. Instead, buyers might be well advised to wait for signs of a gross margin recovery, and not focus solely on the company’s strong top-line growth.

To subscribe to Bamboo Works weekly free newsletter, click here

The chip designer combines an established smart-device human-machine interaction business with newer fast-growing computing products and other technology segments

Key Takeaways

• ESWIN Computing’s computing-chip sales grew nearly 20% in the first quarter of 2026, adding a new growth engine as it prepares to list in Hong Kong
• The company’s RISC-V technology platform and expanding commercial portfolio are expected to drive growth across human-machine interaction, multimedia processing and computing

By Teri Yu

Selling the brains behind intelligent machines sounds far sexier in today’s capital markets than simply supplying the chips behind TV screens. That’s the story that Beijing ESWIN Computing Technology Co. Ltd. (1256.HK) hopes to sell to investors ahead of its Hong Kong IPO, showing how it’s turning its established smart-device human-machine interaction business into a foundation for expansion into connected machines and AI.

ESWIN launched its Hong Kong IPO on Sept 28, seeking to raise up to HK$2.5 billion ($319 million). The company is offering 1.57 billion shares at HK$1.59 apiece, valuing it at as much as HK$34.97 billion, with Citic Securities and China Securities International as co-underwriters. Subscriptions close on Oct 6, with trading scheduled to begin on Oct 9.

ESWIN describes itself as a provider of chips built on RISC-V, an open-standard instruction set architecture based on reduced instruction set computing (RISC) principles, with a strategic focus on combining RISC-V with AI. Using that foundation, the company is building a broader portfolio that covers human-machine interaction, multimedia processing, connectivity and computing, with uses in vehicles, robotics, industrial equipment and intelligent computing.

The company’s latest financials show growing momentum in those newer directions. Its revenue rose 20% to 2.43 billion yuan ($363 million) in 2025, accelerating from 15.6% growth a year earlier. The momentum continued to build this year, with first-quarter revenue up 18.4% to 494.2 million yuan.

Building a business around RISC-V+AI

The RISC-V technology at the heart of ESWIN’s bet on the future provides an open, flexible, modular, and extendable instruction set that companies can use to develop processors, including proprietary designs, according to RISC-V International. That flexibility allows chip developers to build differentiated products around a shared architectural foundation.

ESWIN combines that flexibility with its own intellectual property and integrated hardware and software capabilities. Its products bring together chips and chipsets or circuit boards with the software needed to perform specific tasks.

By March 2026, the company had accumulated more than 620 intellectual-property (IP) modules, over 20 serial RISC-V cores and more than 1,740 patent applications. It had also commercialized more than 150 hardware and software products and served 220 customers.

Those assets are helping ESWIN to pursue several application markets without starting from scratch each time. Its technology modules can be reused and adapted, allowing engineering work on one product to contribute to the development of others.

The same principle extends to customers and suppliers. Existing customer relationships create opportunities to introduce additional products, while shared supply-chain resources support development and production across a broader portfolio.

ESWIN’s RISAA (RISC-V+AI) ecosystem and technology platform, built around RISC-V and AI, brings another dimension to that model by supporting technology sharing, customization and collaboration. The commercial goal is to shorten development cycles and spread the value of accumulated engineering investment across a growing range of applications.

ESWIN resembles U.S. chip giant Qualcomm (QCOM.US) in its business model, designing chips while outsourcing their manufacture, rather than producing them in its own factories. The key difference lies in architectural focus: ESWIN is building its product ecosystem around the open RISC-V instruction set, while Qualcomm’s Snapdragon processors use architecture from Britain’s Arm Holdings (ARM.US).

Clear growth curves

ESWIN’s strategy spans multiple areas: its established smart-device human-machine interaction business, expanding applications in embodied intelligence, and opportunities in AI chips for edge computing and computing centers. Combined, they give the company both an existing revenue base as well as additional markets to develop.

Smart-device chip products provide the foundation. Its human-machine interaction chips support screen-related functions in televisions, monitors, laptops, phones and watches, while multimedia-processing products help process and enhance audio, video and images.

Human-machine interaction products generated 1.86 billion yuan in revenue in 2025, accounting for 76.3% of total sales. Multimedia-processing products contributed a further 150.6 million yuan, up from 100.7 million yuan in 2024.

According to third-party market data cited in the filing, ESWIN was China’s largest domestic provider of smart device human-machine interaction chip products by 2025 revenue. That position gives it a commercial base from which to deepen customer relationships and broaden its product offerings.

ESWIN traces its origins to 2019, founded by veteran semiconductor-display-industry executive Wang Dongsheng. With more than four decades of experience, Wang is often called the “father of China’s semiconductor display industry.”

The company is run by a team with extensive industry experience, including Chairman and CEO Mi Peng, President and COO Hu Weihao and CTO He Ning and others, who bring experience in management, fundraising and product development. Co-founder and Vice Chairman Wang Bo has more than 20 years in the semiconductor industry and is leading the company’s overseas expansion, a key part of its growth strategy.

Computing products are becoming an increasingly important second growth engine in the company’s revenue mix. Revenue from that segment surged 342.5% to 319.8 million yuan last year. The changing mix illustrates how quickly the company’s business is broadening beyond its original human-machine interaction-related strength.

Its computing and connectivity products address complementary needs in intelligent machines. Connectivity products transmit information, while computing products support processing and decision-making, from microcontroller operations to more demanding AI applications.

That combination creates opportunities in embodied intelligence, in areas where vehicles, robots and industrial equipment sense their surroundings and act on the information they collect. It also provides a basis for developing more integrated offerings rather than simply supplying isolated components.

ESWIN has built a range of AI inference chips for edge computing, a market set to benefit as large language models spread. RISC-V chips are well suited to edge inference because they use little power and are highly flexible. For data centers, the company is also developing RISC-V AI chips and CPUs, backed by intellectual property and chip-design capabilities it owns.

The Hong Kong listing is expected to help ESWIN widen its RISC-V ecosystem and speed up commercialization, supporting growth across several businesses.

The Bamboo Works offers a wide-ranging mix of coverage on U.S.- and Hong Kong-listed Chinese companies, including some sponsored content. For additional queries, including questions on individual articles, please contact us by clicking here

To subscribe to Bamboo Works free weekly newsletter, click here

The company plans to invest up to 5 billion yuan over the next five years on its Stars Distributed Intelligent Computing Center AI infrastructure project

By Doug Young

As a listed company, its history dates back to 2019 when it was engaged in financial services. But for Maase Inc. (MAAS.US), history really begins in May 2025, when it took on its current name and embarked on a brief acquisition spree that forms its current core assets. Among its three acquisitions, Huazhi Future has quickly become Maase’s primary focus. The other two came in the new energy and health and wellness sectors.

The Huazhi acquisition takes Maase directly into the hot area of AI infrastructure, which revolves around the data centers filled with servers, GPUs and other hardware needed to run complex AI applications. Following the acquisition, the company moved quickly to secure an additional $50 million in fresh funding, and laid out a 5 billion yuan ($747 million) roadmap for developing data centers running its own self-developed large model.   

Spearheading the new AI infrastructure push is Li Zhifeng, Huazhi Future’s previous chairman who was named as Maase’s chief technology officer after the Huazhi acquisition. Li sat down with Bamboo Works to discuss Maase’s AI infrastructure roadmap in more detail, including its development timeline and how it differs from other developers piling into the market.

Bamboo Works: Let’s start with the new $50 million investment. Can you tell us who was the investor, and what drew them to Maase?

Li Zhifeng: The investor is a financial investment institution with a focus on AI infrastructure. From their perspective, we sit at the intersection of two things institutional investors are watching closely right now — the global computing power shortage, and the growing premium being placed on AI safety and data sovereignty. Our pitch wasn’t: “We buy and resell GPU time.” It was that we’re building a closed loop — infrastructure, our own AI models and enterprise delivery — with signed, revenue-generating contracts already in place, not just a roadmap slide.

Q: Tell us a little bit about the Stars Distributed Intelligent Computing Center launched in April, which is the backbone of your AI infrastructure initiative. Where is it and what kind of services does it provide?

A: The Stars Project was announced by our subsidiary Huazhi Future, together with two partners — China Power Computing Technology Application (Beijing) Co. Ltd. and Sino-International Yuzhi (Chongqing) Co. Ltd. It’s designed to support China’s national “East-to-West” and “Xinjiang-to-Chongqing” computing-resource-transfer initiatives.

The architecture is “dual-core + multi-level edge + a unified dispatch platform” consisting of two centralized training-and-inference hubs — one in Yinchuan, Ningxia, with an initial phase already operational, targeting roughly 512 high-performance servers; and one in Xinjiang. Plus a planned network of 50 to 100 containerized data centers,  each carrying 10 to 16 containerized compute modules, for low-latency inference closer to end users.

Q: What’s the capacity for the center’s first edge node, and what plans are there to scale it up over time? What are the challenges involved in the replication and expansion of modular edge nodes?

A: The first node was deployed in April in the town of Lengshui of Shizhu county in Chongqing, with a designed compute capacity of 4,000 petaflops. On scaling: the modular, containerized design is meant to make replication faster than traditional data-center construction — no long civil-works cycle, it’s largely plug-and-play.

But real challenges remain: securing power-supply arrangements and local-government incentive agreements site-by-site, each of which currently requires its own negotiation; financing each additional node or cluster, since we haven’t yet finalized definitive financing arrangements for the broader network; and the normal execution risks of any distributed infrastructure build — construction timing, equipment delivery, and achieving utilization once nodes go live.

Q: At the time of the Stars Center announcement in April, the company said the project would have eventual investment of 5 billion yuan. What’s the time horizon for that investment, and where does the company expect to get that money?

A: The planned construction period is 60 months — five years — rolled out in five phases from pilot validation to large-scale implementation. So 5 billion yuan is the ceiling for the full build-out. As for funding: we expect financing for individual sites — starting with the Shizhu site — to be arranged at the project-company level, through a mix of equity investment and debt financing from external investors, potentially alongside local incentive arrangements like preferential land-use and tax policies that are still under discussion with the local government.

Q: Tell us about some of the customers for the Stars Center so far, including who they are, the size of their contracts, and what services are they using?

A: I want to draw a distinction here that matters: the signed commercial contracts we’ve announced to date sit within Huazhi Future’s broader computing-power and algorithm-solutions business rather than being revenue specifically generated by the Stars Center’s own infrastructure.

With that distinction clear, our recent disclosed wins include: a computing-services delivery to Guangzhou Benyun Artificial Intelligence Technology, valued at 1.65 million yuan, covering five AI computing nodes plus environment deployment, fully paid and accepted; and a 12-month, 76.8 million yuan agreement with Beijing VirtAI Technology for over 450-petaflop-class high-performance computing infrastructure.

These and several other deals collectively represent well over 100 million yuan in signed, disclosed contract value across the past few months, which we think demonstrates real, paying demand for our computing services — even as the flagship Stars infrastructure itself is still being built out.

Q: How does Maase plan to differentiate its AI services from all the many other companies out there that are setting up similar operations?

A: Most companies in this sector are competing on raw scale — more GPUs, bigger clusters, a general-purpose model with no real safeguards behind it. We think the differentiator that actually matters going forward is trust: can an enterprise or government customer put sensitive data and mission-critical workloads on your platform with confidence? Our approach is “safety by architecture,” not safety as an afterthought.

That starts with our own computing infrastructure layer, so we’re not just renting capacity from a third party we can’t control. It continues with Lingyanmiaoyu, or Lingyan, our large language model, which is designed from the ground up around secure, on-premises private deployment, data localization, and content risk-control, rather than being a public API we bolt safety filters onto later; and it extends to how we deliver — data governance, compliance support, and ongoing operations as part of the contract, not just a one-time model handoff.

Q: Tell us more about the Lingyan mixture-of-experts (MoE) large model that you’re using. What’s the future international roadmap for Lingyan?

A: Lingyan is our proprietary large language model, built on a mixture-of-experts architecture with a total parameter scale of 9 billion. We chose MoE first for efficiency — dynamically activating selected expert networks on demand instead of the full parameter set on every request, which lowers inference cost and improves concurrency. But the reason that efficiency matters strategically is that it’s what makes safety-hardened, privately-deployed AI commercially viable rather than a research showcase — enterprise and government customers that require on-premises, data-sovereign deployment need a model that’s both controllable and cost-effective enough to run that way at scale.

The Bamboo Works offers a wide-ranging mix of coverage on U.S.- and Hong Kong-listed Chinese companies, including some sponsored content. For additional queries, including questions on individual articles, please contact us by clicking here.

To subscribe to Bamboo Works’ free weekly newsletter, click here.

The world’s largest smartphone ODM is using acquisitions to cater to booming demand for AI infrastructure

Key Takeaways:

• Longcheer has revised terms for its acquisition of an AI infrastructure company, reducing the size of its stake and making performance targets stricter
• The world’s largest contract smartphone manufacturer is tapping demand for AI infrastructure with a recent string of acquisitions 

By Edith Terry

For years, its calling was making millions of smartphones that have become a fixture of everyday life for most people around the world. But these days, original design manufacturing (ODM) giant Shanghai Longcheer Technology Co. Ltd. (9611.HK, 603341.SH) is making a new bet on AI infrastructure that has become the flavor of the day among tech manufacturers.

But its roadmap, driven by acquisitions, hasn’t been completely smooth. With global demand for AI data centers growing fast, the company looked like it had a good plan last month when it announced it would purchase of 80% of Suzhou A-Rack Information Technology, a maker of data center equipment like server racks and power distribution units (PDUs).

It made the announcement just a few weeks before disclosing midyear results for 2026 that showed its core smartphone manufacturing business wasn’t going anywhere fast. Its revenue fell 6.3% to 18.7 billion yuan ($2.78 billion) in the first half of the year, while its profit fell by an even steeper 30.9% to 245.9 million yuan.

But then last week Longcheer abruptly announced it was reducing the size of the A-Rack stake it was buying to just 60%, in what looked like a hedging of its bet by investing less. It took the step after the Shanghai Stock Exchange, following the initial August announcement, issued a letter warning on integration risks, as well as a large gap between A-Rack’s financials and performance targets set by the two sides.

The lowering of its stake reduced Longcheer’s purchase price from an original 1.12 billion yuan to 840 million yuan, while leaving A-Rack’s valuation unchanged at 1.4 billion yuan. It also modified terms of the agreement by introducing penalties if A-Rack failed to meet performance targets, in a seeming acknowledgement that perhaps Longcheer had been too eager to sign the original agreement without taking all the risks into account.

The revised agreement also requires A-Rack founder Ding Zhiyong to buy at least 50 million yuan worth of Longcheer’s Shanghai-listed shares within 12 months, which would then be subject to a 12-month lockup period.

Investors in both Shanghai and Hong Kong applauded the revised terms that lowered Longcheer’s exposure and made A-Rack more accountable for its post-deal performance. The Shanghai stock rose by 10.8% over the next two trading days, while the Hong Kong stock rose by 5.2% over the same period.

A-Rack is one of three purchases Longcheer has made this year in a bid to buy its way into AI infrastructure manufacturing. In June it paid 540 million yuan for 60% of KC Precision Technology and Dongguan Geeia Metal Products. KC Precision makes high-precision metal etching products for consumer electronics, while Geeia makes thermal management devices for data centers.

According to Grand View Research, the global AI data center market that Longcheer is targeting is expected to grow from $147.3 billion in 2025 to a projected $810.6 billion in 2033, averaging about 25% growth annually over that time. In addition to powerful computing chips, AI data centers also require thermal management systems to dissipate the huge heat given off by those chips, as well as large amounts of electricity to power them.

Post-IPO weakness

In addition to its core smartphones, Longcheer’s current business also includes ODM services for other electronics like tablet PCs, AI of things (AIoT) devices and car electronics. In its prospectus ahead of its Hong Kong IPO in January, it cited third-party market data saying it was the world’s second largest ODM manufacturer of consumer electronics in 2024, based on shipments, and the largest smartphone ODM. Its A-list of smartphone customers includes the likes of Xiaomi, Samsung, Honor, Oppo and Vivo.

Hong Kong investors initially welcomed Longcheer, which raised around HK$1.52 billion ($194 million) in its January IPO whose retail portion was 15 times oversubscribed, and whose cornerstone investors included Xiaomi and Qualcomm. The stock briefly rose from its IPO price of HK$31 after its trading debut. But it quickly reversed course and has moved steadily downward since then. Its Wednesday close of HK$21.40 is about 30% below its IPO price, as investors flock to sexier stocks with greater growth potential.

While more favorable terms under the revised A-Rack transaction provided a brief respite for the stock, the reality is that Longcheer’s financials are nothing to brag about. Its revenue fell 7% in 2023, before rebounding strongly by 70% in 2024, only to fall again by 9.3% to 42.1 billion yuan last year. The company’s annual profit has also fluctuated, but has generally ranged between 500 million yuan and 600 million yuan every year since 2022.

The company’s main revenue driver in 2025 was a 41% increase in sales of AIoT products, which rose to 7.8 billion yuan to account for 18.5% of revenue, driven by growing demand for AI glasses and other smart eyeware. Smartphones still accounted for about two-thirds of its sales, or 68.6% of revenue at 28.9 billion yuan. But that figure was down by a sharp 20%, from 36.1 billion yuan in 2024, as sales by many major brands slumped after they were forced to raise prices due to soaring memory costs.

So, how does Longcheer compare to its peers? Its closest rival is Huaqin (3296.HK, 603296.SH), which also manufacturers smartphones but is pushing into data center infrastructure and robotics. Huaqin’s Hong Kong-listed shares carry a price to earnings (P/E) ratio of about 17, similar to Longcheer’s multiple of 18.4 for its Hong Kong shares.

But Huaqin’s revenue has been on a steadier upward track, including 55% growth last year to 172.4 billion yuan. Its net profit also increased by more than 50%, from 2.7 billion yuan in 2023 to 4.1 billion yuan in 2025. Like Longcheer, Huaqin’s Hong Kong shares haven’t fared too well since their April IPO, currently trading about 8% below their offer price.

Nonetheless, analysts are still relatively bullish on both companies, with those surveyed by Yahoo Finance giving them “buy” ratings. Perhaps they’re lured by the AI angle, which has fueled explosive gains for many related stocks over the last year.

To subscribe to Bamboo Works weekly free newsletter, click here

"If you want to ensure the success of a company that is still very much regional Southeast Asia when going to market, it's probably a bit easier in Hong Kong than in the U.S." — Explaining why a Southeast Asian consumer brand would prefer a Hong Kong listing over Wall Street.

Key Takeaways:

• A U.S.-developed AI tool's mock breach of WeChat highlights escalating cybersecurity risks and could trigger tighter oversight from Beijing
• Jollibee's decision to list in Hong Kong over the U.S. underscores the city's growing appeal for regional consumer brands seeking Asian capital

By Rene Vanguestaine and Doug Young

A couple of recent stories are reflecting a shift in the trans-Pacific business landscape. On one hand, a mock AI attack on China's premier WeChat messaging app has exposed the vulnerabilities of the country's digital ecosystem to U.S.-developed technology. On the other, a major Southeast Asian fast-food conglomerate has decided to bypass Wall Street in favor of Hong Kong for its international listing. Both events underscore the evolving regionalization of Asia's digital and financial infrastructure — and the intricate push-and-pull dynamics with the U.S.

We're learning that a group of benign researchers in California recently used U.S. AI to build a tool that could breach millions of accounts on WeChat, the ubiquitous app owned by Tencent (0700.HK), in just hours. Their tool, dubbed WeWorm, can hijack a user’s account, call their contacts, and spread from phone to phone without anyone ever answering a call.

For a platform boasting 1.4 billion users that essentially functions as a telecoms carrier in its own right, this is deeply alarming. We believe Beijing isn't reacting well to this development. Usually, the shoe is on the other foot, with the U.S. accusing Chinese hackers of infiltrating its digital spaces. WeChat is the nexus of communication for the Chinese population. The ability to control this network and reach such a massive audience carries severe risks of abuse by anyone seeking to incite social instability and spread disinformation.

While Beijing doesn't directly own the network, it maintains heavy indirect control, likely communicating with Tencent on a daily or even hourly basis. Beijing has consistently demonstrated its ability to control the internet in ways previously thought impossible. If the Chinese security apparatus believes Tencent isn't doing enough to contain this AI threat, we're sure the government will step in with heavier oversight to bring the situation under control.

The AI angle is equally compelling. Until just a few days ago, the major tech heads pushing AI were essentially telling the world not to worry — assuring us that while there might be a minuscule chance of disaster, everything would generally be okay. Then, they suddenly reversed course, warning that the technology might be getting out of control and require immediate slowing down.

We're sure WeChat is working to fix this specific vulnerability that was uncovered by the California team. But as we've seen with social media companies losing control or maliciously manipulating users, this certainly isn't the last time we're going to talk about this kind of risk.

In response to offshore threats, we expect Beijing to crack down and implement stronger oversight on domestic AI companies. While China's public focus has largely been on AI applications to boost manufacturing productivity and counter a shrinking working-age population, it's highly probable the government is heavily involved in behind-the-scenes military and defense applications.

Shifting tides in Asian IPOs

Shifting gears to the capital markets, we're seeing another interesting regional play. Jollibee, a Philippines-based fast-food giant that owns brands like The Coffee Bean & Tea Leaf, recently announced a change of direction for its international operations. The company scrapped plans to spin off and separately list the international operation in the U.S., choosing to move the listing to Hong Kong instead.

Wall Street has long held the upper hand in attracting major Asian listings, but Hong Kong is gaining momentum. Much of this is driven by Mainland Chinese companies that currently face a hard time getting Beijing's approval to list in the U.S. But Jollibee's case highlights a different trend. Fast food doesn't typically boast the high margins or name recognition required to excite U.S. retail investors, who already have no shortage of domestic food and beverage IPOs to choose from.

In contrast, Hong Kong hosts a sizable, dedicated pool of investment money — including funds from the U.S. and Europe — focused specifically on Asia and Southeast Asia. These investors have a much better understanding of local economies in the region and why this company is successful. There are some cases where consumers have tried the brand in Hong Kong and responded positively, reflecting this deeper regional familiarity. Going to market in Hong Kong is simply easier for a regional Southeast Asian business.

This presents a vital test for Hong Kong: can its stock market attract more Southeast Asian companies to diversify away from solely Mainland firms? We think Hong Kong is far better equipped for this than Singapore. Despite being the financial capital of Southeast Asia where Jollibee is strongest, the Singapore Stock Exchange just doesn't seem to have the liquidity depth and trading volumes to satisfy companies of a certain size. Hong Kong's advantage is further bolstered by having China as its massive financial benefactor. That doesn't mean Wall Street is losing its crown entirely. For unproven, early-stage high-tech companies, the U.S. remains the better market. U.S. investors better understand the tech sector and are more willing to risk capital, offering sustained valuations beyond just the current AI hype. But for the regional consumer sector, Hong Kong is proving to be a much more welcoming home.

The company’s stock soared after its controlling shareholder acquired a large stake in an AI company, but its financial reality is grounded in a deteriorating fireproofing business

Key Takeaways:

• Easy Smart warned that its net loss for its latest fiscal year ballooned, citing fierce competition for new tenders in its core fireproofing business
• The warning follows a huge surge in its shares from May to August, after its top shareholder acquired a substantial stake in an AI company

By Warren Yang

A pivot to AI seems to be a cheat code for attracting investors these days, with no shortage of Hong Kong-listed companies suddenly discovering a new high-tech angle to their businesses. But it’s also no secret that many AI narratives that suddenly awaken long-slumbering stocks are written around hype, often with little or no substance.

A case in point is Easy Smart Group Holdings Ltd. (2442.HK), a subcontractor that specializes in the design, supply and installation of passive fire protection materials for buildings and other construction. In just a few months, this otherwise mundane company delivered an eye-popping rally on Hong Kong’s stock market, fueled by an apparent attempt to hop on the AI bandwagon.

That story, however, is completely disconnected from Easy Smart’s financial reality that remains much closer to earth. Last Thursday, Easy Smart warned that its net loss for its fiscal year through June widened dramatically to as much as HK$32 million ($4.1 million) from HK$500,000 in the prior year. In its brief filing, Easy Smart attributed the bottom-line deterioration to fierce, price-sensitive competition for new tenders following the completion of major public sector projects.

Founded in 2021, Easy Smart’s core business is decidedly unsexy, installing fire-rated boards, sprays and coatings to slow the spread of flames when fires occur in commercial, residential and public works buildings.

Then, in May, the company started making interesting, if not perplexing, moves. That month, Easy Smart brought in Tang Tian-Shen, a tech veteran with experience at major global chip makers Intel and SMIC, as its executive chairman. The sudden addition probably left many speculating that the company was headed toward flashy advanced technology. And in anticipation of that kind of development, the company’s stock price began taking off around this time.

Last month, Easy Smart’s controlling shareholder, Talent International Group, transferred a quarter of the Hong Kong-listed company’s issued stock to a business named Guangte Yuanzhi. In exchange, Talent International received about 46% of a wholly-owned Guangte Yuanzhi subsidiary. Following announcement of the transaction, Easy Smart’s stock skyrocketed to a record HK$159.50, a more than 100-fold surge from the start of last year, lifting its market valuation to more than $7 billion.

Guangte Yuanzhi’s core business is providing software and hardware stacks that allow large language models to run efficiently across mixed chip configurations to optimize computing performance and energy efficiency. So casual observers can easily jump to the conclusion that through the equity tie-up, Easy Smart is positioning itself as a provider of intelligent computing services.

But here’s the catch. It’s Easy Smart’s parent, Talent International, that has moved into AI, not Easy Smart itself. That means Easy Smart won’t be able to consolidate any profit from the Guangte Yuanzhi AI business into its own earnings. Talent International may eventually transfer its stake in the Guangte Yuanzhi subsidiary into Easy Smart so that the fireproofing company has some direct exposure to AI to justify its sky-high valuation, though that remains to be seen.

No synergy

It’s also hard to see how Talent International can derive any operational synergy for Easy Smart from the Guangte Yuanzhi deal. Easy Smart doesn’t need any of Guangte Yuanzhi’s services to install fireproofing materials. At best, Easy Smart could set up a small operation to distribute Guangte Yuanzhi’s products using its corporate contacts, which are limited to the small Hong Kong market. But a full-on transformation into an AI company would be difficult, given its current resources and expertise.

On the other hand, being tied to a listed company can bring substantial benefits for Guangte Yuanzhi, whose AI subsidiary is presumably burning through cash to fund its high-cost operations like other AI computing-related peers. With a 25% stake in Easy Smart, Guangte Yuanzhi has secured a sort of backdoor onto the Hong Kong Stock Exchange without going through the costly and lengthy traditional IPO process.

That could allow Guangte Yuanzhi to use Easy Smart as a vehicle for offshore fundraising. For example, as a major shareholder, Guangzhou Yuanzhi could exert significant influence over Easy Smart’s decisions to issue new shares or convertible bonds, and route the proceeds back to the AI company via a joint venture, licensing agreements or related-party transactions. Furthermore, Easy Smart’s elevated market valuation directly boosts Guangte Yuanzhi’s own implied valuation since the two entities swapped equity in the August deal, strengthening its leverage for future private funding rounds. Guangte Yuanzhi can also pledge its Easy Smart shares, now worth around $1 billion, as collateral for debt financing.

Notably, Guangte Yuanzhi’s 25% stake sits just below the 30% threshold that would require it to make a mandatory buyout offer for Easy Smart under Hong Kong rules. This structural positioning indicates a preference to utilize Easy Smart as a back-end funding platform without incurring regulatory requirements or the capital required for a full takeover.

For Easy Smart’s controlling shareholder, Talent International, the deal looks like a strategic monetization and risk-management play. Easy Smart’s soaring stock gave Talent International some huge paper gains, while also diversifying it into a hot new area away from its relatively mature fireproofing business with limited growth potential.

All this turned Easy Smart into a bonanza for speculative traders, especially because the company’s free float is small, about 35% of its total issued shares, which makes it easy to pump up its valuation quickly. But that kind of thin liquidity means the stock can fall back to earth just as fast as it rose. And indeed, the shares have lost more than half of their value since hitting the record high last month.

Even after the pullback, Easy Smart shares still command a massive price-to-sales (P/S) ratio of nearly 100, compared to 4.3 for Man Shun Group (1746.HK), which installs heating, ventilation, and air-conditioning systems and provides electrical and mechanical engineering services, and 2.2 for Lumina Group Ltd. (1162.HK), which provides fire safety system installation, repair, and maintenance services.

Then again, this group of companies is probably no longer comparable as peers for Easy Smart, at least in the eyes of investors. Perhaps a more suitable group would be AI hotshots like Z.AI (2513.HK) and MiniMax (0100.HK), which command even higher P/S ratios of 198 and 141, respectively. For investors still hoping to profit from Easy Smart’s spectacular rally, deteriorating financials for its low-tech core business are a sobering reminder of the gap between market narrative and reality, especially when the new story revolves around AI.

To subscribe to Bamboo Works weekly free newsletter, click here

The IT and software services giant is stumping up $141 million to gain a bigger foothold for its AI solutions in the power-generation and new-energy industry 

Key Takeaways:

• Chinasoft plans to acquire a controlling stake in energy services provider Gisway, paying 41% over the market price
• Chinasoft’s AI-related revenue more than doubled in the first half as it folds advanced computing and agentic functions into its enterprise services

By Lee Shih Ta

From high-performance models to autonomous agents, Chinasoft International Ltd. (0354.HK) has been gradually expanding its AI services over the past few years, targeting a range of business sectors.

Now the provider of IT and software services to large enterprises is making a deeper dive into the energy industry, via an engineering consultancy that helps to build and manage electrical power networks.

On Sept. 16 Chinasoft announced plans to spend 944 million yuan ($141 million) to gain a controlling stake in Hangzhou Gisway Information Technology Co. Ltd. (301390.SZ) as its looks to embed its evolving AI functions across a host of client industries, including power generation.

Under the proposed deal, the IT services giant would pay a 41% premium to acquire a 29.68% stake in Gisway at 53 yuan per share. Once the deal closes, Gisway would be consolidated into Chinasoft’s accounts but would retain the right to nominate its own chairman and most of its board members.

The move fits with Chinasoft's AI strategy over the past two years, built around an intelligent operating system designed to allow AI agents to handle complex processing tasks for large enterprise clients. 

Going vertical 

Chinasoft is not the only software and IT company to take this AI route. China’s Kingsoft Office (688111.SH) has embedded AI into WPS Office software package and enterprise workflows, while Digital China (000034.SZ) is expanding into AI computing power, enterprise AI agents and industry solutions. Chinasoft, however, places a stronger emphasis on deploying AI for its existing clients, in industries such as banking, telecoms, manufacturing and utilities.

Chinasoft's revenue from AI products and services more than doubled to 1.53 billion yuan in the first half of this year from the same period of 2025. The company is gradually infusing AI into the energy and power sector. In June, it won a bid for a large model project related to the Yalong River, and during a collaboration with Moonshot AI in July, energy was highlighted as a priority industry. The Gisway acquisition extends Chinasoft's footprint within the power industry.

Gisway’s expertise spans planning, design, construction, operation and maintenance of power networks, including new-energy solutions such as solar energy systems and charging stations for electric vehicles. Beyond a new batch of energy clients, the new subsidiary offers Chinasoft engineering credentials, an on-site project presence and hands-on experience.

These capabilities stretch Chinasoft's existing AI, software and digitization services into areas such as forecasting power generation, inspecting equipment and maintaining grids. In its announcement, Chinasoft noted that the two parties could share market channels, supply chains and client resources, driving the integration of software and hardware within the power sector.

Although AI-related income is rising, Chinasoft earnings are feeling the pressure of the transition. Revenue surpassed 20 billion yuan in 2022 but has hovered around 17 billion yuan for the last two years.

Its acquisition target has also been feeling the pinch. Gisway's profits fell sharply last year and sliding revenues tipped the company 23.12 million yuan into the red in the first half. Part of the pressure stems from delays in breaking ground on new energy projects and a slowing pace of upgrades to existing networks for power distribution. On the bright side, its balance sheet remains relatively clean. By mid-year the company held around 314 million yuan in cash and cash equivalents, with zero bank loans.

A fair price to pay?

Chinasoft agreed to acquire its Gisway stake at around 41% over the pre-announcement closing price of 37.5 yuan, valuing the company at about 3.18 billion yuan. Gisway's net assets were roughly 921 million yuan at the end of June, putting the overall transaction at about 2.5 times net book value. Whether the premium is justified or not will depend on Gisway’s earnings trajectory.

Once the deal closes, Gisway will become the first Chinasoft subsidiary with a mainland listing, giving the group an equity presence in Hong Kong and Shenzhen. However, Chinasoft has pledged not to pursue a backdoor listing through Gisway or inject related-party assets for 36 months, meaning the short-term impact will depend on integrating the businesses.

The purchase price is a significant investment. According to the filing, at least half the capital will come from the buyer's shareholders, while the remainder may include bank M&A loans. The acquired shares are subject to a 60-month lock-up period for transfers and a 36-month lock-up for pledges, implying the success of the integration will take years to verify.

So far, Chinasoft has not enjoyed a clear premium from the rapid growth of its AI business. Its trailing price-to-earnings ratio is about 24 times, in the same range of 20 to 24 times seen between 2021 and 2024, and below the multiple of around 34.7 times at the end of 2025. In other words, after a pullback in its stock price, Chinasoft's valuation has returned to its historical norm.

To break above the ceiling, Chinasoft would need to halt the earnings slide at Gisway and win new AI services orders from the energy sector. But the cost of acquiring and integrating its new power unit could constrain the upward momentum.

To subscribe to Bamboo Works weekly free newsletter, click here

The company is aiming to list in Hong Kong, feasting on a Chinese direct drive actuator module market that soared from 200 million yuan in 2023 to 1.9 billion yuan last year

Key Takeaways:

• Direct Drive Tech has filed for a Hong Kong IPO, reporting its sales rose 40% in the first half of this year after tripling in 2025
• The company controls more than 60% of China’s market for consumer-use direct drive actuator modules that are a key component of affordable household robots  

By Doug Young

Humanoid robots look set to become key helpers in homes of the future, assisting with everything from daily chores like cleaning, cooking and laundry, to taking care of the elderly and children. But in the present, a much simpler generation of consumer robots is rapidly gaining traction by performing many similar tasks on much smaller budgets.

One company underpinning this quieter robot revolution is Direct Drive Tech Ltd., a robotics technology company with two business segments, robotic actuator modules and robots, which hopes to impress investors with its rapid growth and fast-improving margins, as it marches towards a planned Hong Kong IPO. The company filed its IPO prospectus with the Hong Kong Stock Exchange last month and added an update on Sept. 17, with Citic Securities (Hong Kong) as its sole sponsor.

The company was last valued at 3.2 billion yuan in its Series C financing at the end of last year, in which it raised 185 million yuan ($27.6 million). Now, it aims to raise another HK$982.5 million ($125 million) in its Hong Kong IPO by selling 50 million H-shares for HK$21.60 apiece, according to its latest listing document filed on Sept. 21.

The company’s core product, the actuator modules that give robots their ability to move, only sell for a few dollars each, reflecting the relative maturity and growing affordability of a fast-growing industry cranking out products for daily activities like household cleaning, lawn care and fitness.

But when you multiply that by the millions, the opportunity is quite large and growing quickly. China’s direct drive actuator module market grew by more than a factor of eight from just 200 million yuan in 2023 to an estimated 1.9 billion yuan last year, according to third-party research in Direct Drive Tech’s prospectus.

The company is hardly the only one chasing that space, competing with global players like Japan’s Harmonic Drive Systems (6324.T), as well as domestic rivals Leaderdrive (688017.SH) and ZhongDa Leader (002896.SZ). But it’s carved out a very comfortable space in the market for consumer-use direct drive actuator models, controlling 61.1% of that market in China last year, according to the prospectus.

Its listing would make Direct Drive Tech Hong Kong’s first “direct drive actuator module” stock. And unlike higher-profile humanoid robot makers and their suppliers, most of which are deeply in the red, Direct Drive Tech, despite its relatively short history, is inching its way towards profitability on an adjusted basis, which excludes share-based compensation and changes in financial instruments.

The company is also distinguished as the only one in its segment to ship more than 5 million consumer-use direct drive actuator modules. It was on track to ship more than double that figure this year, as it sold 5.7 million modules for consumer-use robots in the first half of this year.

That sales ramp-up is a big part of the company’s story. Annual capacity at its two production bases in the city of Dongguan more than quadrupled to about 15.7 million modules last year from about 4.1 million in 2024. It continued to grow this year, on track to reach more than 20 million units annually.

Young company

Direct Drive Tech has made significant strides in short order to get to its current phase, founded just six years ago in Dongguan before moving its headquarters to Beijing last year. Its founder and Chairman, Zhang Di, started his company at age 26 after earning a bachelor’s degree in mechanical engineering from the Beijing Institute of Technology, and then studying robotics systems and control engineering at the Hong Kong University of Science and Technology.

The company’s story is one of rapid scaling and improving business metrics since it rolled out its first actuator modules shortly after its founding in 2020.

Its revenue more than tripled to 282 million yuan last year from 79.8 million yuan in 2024, and grew another 40% to 200 million yuan in the first half of this year from 143 million yuan a year earlier. Consumer-use robotic direct-drive actuator modules are its biggest revenue source, accounting for 82% of its revenue this year.

The company’s industrial and commercial actuator modules business is a much smaller but fast-growing segment, roughly tripling to 21 million yuan in the first half of this year from 6.58 million yuan a year earlier, rising to 10.5% of revenue from 4.6% over that period.

Beyond modules, the company’s second business segment, robots, showcases its direct drive technology at the complete-machine level, despite the segment’s modest share of total revenue. With wheel-legged robot revenue of 7.3 million yuan in 2025, the company ranked fourth in China’s wheel-legged robot market with a 4.7% market share, according to research in its prospectus.

Advances in motor power density and efficiency could also broaden the use of its higher-margin joint modules in humanoid robots and other embodied-intelligence systems, potentially lifting overall gross margins. Joint modules generated a gross margin of 29.2% in the first half, compared with 17.2% for consumer direct-drive modules, although they accounted for just 1.7% of revenue.

Here, we should point out that one of the company’s vulnerabilities is its relatively high customer concentration. Its top five customers currently account for more than 80% of its sales, with its largest accounting for more than half in the first half of this year.

As its scale improves, the company’s gross margin has been improving rapidly, rising from 13.5% in 2023 to 21.5% last year.  Its expenses as a percentage of revenue have also been coming down steadily, which is exemplified by its R&D costs, which dropped from 223% of revenue in 2023 to just 19.7% last year. That figure rose to 26.8% of revenue in the first half of this year, as the company cited spending related to its rapid business expansion.

Direct Drive Tech’s adjusted non-IFRS net loss also dropped from 61.2 million yuan in 2023 to 43.2 million last year.

Despite its progress, the company’s 21.5% gross margin last year still trailed peers like Harmonic Drive and ZhongDa Leader, which recorded margins of 30.5% and 26.1% in their latest fiscal years, respectively. But steady improvement with its growing scale could soon help Direct Drive Tech to catch and even surpass those rivals, as it continues to feast on a fast-growing and proven market for simple home-use consumer robots.

The Bamboo Works offers a wide-ranging mix of coverage on U.S.- and Hong Kong-listed Chinese companies, including some sponsored content. For additional queries, including questions on individual articles, please contact us by clicking here

To subscribe to Bamboo Works free weekly newsletter, click here

The precision-tool maker founded by a former toy factory worker is planning another factory, even as its Hong Kong shares trade well  below their July IPO price

Key Takeaways:

• Dtech is planning to build a 700 million yuan materials factory, while net cash outflow and growing competition test the durability of its growth
• The company’s revenue more than doubled in the first half of 2026 as increasingly complex circuit boards needed for AI lifted demand for its precision tools

By Hu Minghe

Wang Xin entered China’s sprawling manufacturing sector as a worker on a toy factory assembly line back in the late 1980s. Fast forward nearly four decades, when her career trajectory has taken her up the value chain to producing the tiny drills used to make printed circuit boards (PCBs). Her somewhat anonymous product is enjoying a renaissance these days, thanks to demand from the AI boom. But Hong Kong investors barely seem to notice.

The drills made by her company, Guangdong Dtech Technology Co. Ltd. (1377.HK; 301377.SZ), create the tiny holes for electrical connections between the layers of circuit boards, which link components inside devices such as servers. AI servers require complex boards made with harder-to-machine materials, putting greater demands on the tools used to make them.

Based on the city of Dongguan, Dtech listed in Hong Kong in July this year, complementing its domestic listing in Shenzhen in 2022. Now, it’s taking a slight step back to the past with new plans for a 700 million yuan ($105 million) factory to expand its smaller businesses in industrial brushes, films and glass coatings, according to an announcement last week, adding the project is still awaiting investment and land agreements.

The businesses Dtech plans to expand have had mixed fortunes. Sales of brushes and other materials used mainly to clean and smooth PCBs rose about 50% in the first half of 2026 from a year earlier. But sales of films for displays and cars, including layers that make screens harder to read by anyone from the side, fell about 25%.

The company’s main growth driver remains its core cutting-tool business, which makes drills, cutters for trimming boards and tools for cutting other parts. That business accounted for most of the company’s revenue growth in the first half of the year, when the overall figure more than doubled to 1.83 billion yuan from 894 million yuan a year earlier, according to its midyear results published last month.

Early arrival to Dongguan

Wang’s history in Dongguan dates back to 1989, when she moved to the city, then an emerging manufacturing hub near Hong Kong, from Central China’s Henan province. She worked at several toy factories, rising from assembly-line worker to supervisor before moving into PCB sales. In 1997, she used 20,000 yuan to 30,000 yuan of savings to start her own business, buying used drills from large factories, repairing them and reselling them to smaller ones, according to Xinhua.

She then assembled a team to develop machines for making drills, initially taking no salary alongside several fellow shareholders. Workers were paid first, and the remaining money went back into improving the equipment and production methods.

The AI connection grew through the company’s existing customer base. Dtech says it has supplied a Chinese customer now making PCBs for AI servers since 2016, working together to develop tools to help manufacture more complex boards, according to the prospectus for its Hong Kong listing. The company ranked first worldwide by PCB drills sold in 2025, with 29.2% of the market, according to the prospectus.

Small tools, bigger returns

Drills wear out, so PCB manufacturers need new ones as old drills get retired, and also for capacity expansion. That double-pronged demand lifted Dtech’s tool revenue by 114.5% to 1.6 billion yuan in the first half of 2026, accounting for nearly 90% of total revenue.

Coated drills increased from 36.2% of drills sold in the first half of 2025 to 49.5% in the same period of 2026. Coated products carry higher price tags for Dtech, lowering customers costs per hole by making drills last longer. But longer life means fewer replacements, which ultimately translates to less sales for Dtech.

As its revenue more than doubled in the first half, the company’s profit more than quadrupled to 679 million yuan, thanks to a sharply improving gross margin that rose from 38% to 53.1%. Management attributed the big improvement to more sales of higher-value products and improving efficiency with economies of scale.

The latest results also include the German tool business Dtech bought from MPK Kemmer, which was insolvent at the time of the purchase in August 2025. Its latest first-half results do not separately identify that business’s contribution to growth in the first half of this year. But the modest purchase price of about 3 million euros ($3.4 million) suggests Dtech bought the company more for its intellectual property and German presence than its actual sales.

Dtech isn’t the only company getting a lift from AI demand. Japanese drill maker Union Tool (6278.T) reported 48.3% year-on-year revenue growth in the first half of 2026, while Taiwan’s Topoint (8021.TW) grew 66.6%. As demand remains strong, Topoint plans to double its April 2026 monthly capacity to 70 million drills by the end of 2027. Dtech is already well above that, with monthly capacity of more than 160 million drills in August.

Lukewarm shareholder reception

Despite its rapid growth, Dtech’s first-half operations again used more cash than they brought in. Its net cash outflow rose to 45.5 million yuan in 2026, against 29.4 million yuan in the same period of 2025, though much of that rise was due to a big jump in income tax.

The company’s Hong Kong IPO in July raised HK$4.67 billion ($595 million), providing money to expand. Despite its strong growth, however, Dtech’s Hong Kong shares have gotten a lukewarm investor reception in the more than two months since their debut. The stock climbed to an intraday high of HK$455.60 a week after its trading debut, up 20% from its IPO price of HK$380. But the shares have fallen since then and at their Monday close of HK$287 were 24.5% below their IPO price.

UBS attributed the retreat to waning investor enthusiasm and delays in the expected rollout of PCBs for Nvidia’s future Kyber AI computing design, according to a report this month in Zhitong. The report said UBS rates Dtech a “buy,” with a HK$439 price target, and forecasts its net profit will grow about 78% annually between 2026 and 2028.

The average forecast of five analysts tracked by Yahoo Finance puts Dtech’s revenue at 9.13 billion yuan in 2027, nearly double their estimate for this year. Those forecasts show that expectations for the company are high. Wang built her business by repairing used drills. Investors now need to see whether Dtech can keep its current more cutting-edge tools business profitable and growing in the face of stiff competition, as it also adds to its spending with the new materials factory expansion.

To subscribe to Bamboo Works weekly free newsletter, click here

Industrial robot maker Changzhou Microintelligence Co. Ltd. has completed its filing with the China Securities Regulatory Commission (CSRC) for an overseas listing and the full circulation of its domestic unlisted shares, a required regulatory step for Mainland Chinese companies seeking to list overseas.

According to a notice published by the CSRC last Friday, Microintelligence plans to issue up to 42.87 million overseas-listed ordinary shares on the Hong Kong Stock Exchange. Another 335.48 million domestic unlisted shares held by 37 shareholders are set to be converted into overseas-listed shares for trading in Hong Kong. The notice was issued on Aug. 24, and the company must complete the listing within 12 months.

Founded in 2018, Microintelligence provides embodied intelligent industrial robot (EIIR) products for industrial applications including inspection and assembly. The company first filed for a Hong Kong listing in September 2025, and subsequently updated its listing application in October that year and again in March 2026.

According to its latest listing application, the company’s revenue rose 32.5% in 2025 to 795.5 million yuan, while its net profit fell to 5.07 million yuan from 15.74 million yuan a year earlier. Revenue from its core EIIR products jumped 66.7% to 453.3 million yuan last year, accounting for 57% of total revenue, with a gross margin of 53.5%.

By Lee Shih Ta

To subscribe to Bamboo Works weekly free newsletter, click here

Following its earlier transition from construction to environmental work, this ‘chameleon company’ has now set its eye on becoming an AI wunderkind

Key Takeaways:

• Envision Greenwise has spent nearly 1.7 billion yuan on servers as it makes a new move into AI infrastructure
• The company, traditionally engaged in construction services and recycling, aims to boost its computing power capacity to 16 MW in the short term

By Lau Chi Hang

A recent announcement of plans to invest 380 million yuan ($57 million) for over 100 servers may have left people less familiar with Envision Greenwise Holdings Ltd. (1783.HK) scratching their heads. The company, whose main business until recently was in battery recycling, said the investment was aimed at bolstering its cloud and intelligent computing service capabilities.

Many might not be very familiar with Envision Greenwise. But knowing that high-flying large AI model developer MiniMax is among its investors may help to explain the new purchase.

The old adage that “change is the only constant” quite fittingly describes Envision Greenwise these days. The company was historically engaged in traditional building construction, and only ventured into the recycling business a little over three years ago. But that apparently wasn’t enough, and this year it abruptly pushed into the red-hot AI sector, sensing a new opportunity.

Two cousins

The company’s story began with two Hong Kong entrepreneurs, Chan Kam Tong and his cousin Chan Kam Ming. In 1985, the pair acquired Head Fame, a building contractor, with dreams of taking on superstructure construction projects. After three decades of hard work, they successfully took their company, renamed Golden Ponder, public in 2018.

But the construction industry is hardly a capital market darling these days. Devoid of any glamorous narrative and plagued by chronically low valuations, Golden Ponder also faced its own major operational struggles as a result of the pandemic and a slumping Hong Kong property market. In 2022, it brought in Kwok Chun Sing, a graduate of Fujian Agricultural University with over two decades of experience in environmental technology and new energy. Kwok founded the Hong Kong Recycling Chamber of Commerce in 2015 and has served as a director at the applied technology research arm of China Resources Environmental Protection since 2021.

Shortly after coming aboard, Kwok launched a reverse takeover of the Chan brothers' equity interests to become the company’s controlling shareholder. He then immediately set out to transform the company into a green energy and circular economy platform. Anticipating a coming boom for new energy vehicles (NEVs) at the time, he zeroed in on the battery sector, branching out into battery recycling, battery energy storage systems, and electric vehicle (EV) charging facilities.

Kwok's most striking asset was his extensive network, which allowed his company to form a steady string of tie-ups with top-tier enterprises. As recently as earlier this year, the company announced a collaboration with lithium behemoth Tianqi Lithium to jointly develop lithium-ion battery recycling technologies.

His efforts weren’t wasted. Fueled by revenue from reverse supply chain management and environmental-related services, the company's top line nearly tripled to HK$2.46 billion ($316 million) in its latest fiscal year. It also posted a profit of HK$61.03 million for the year. That said, the limited profitability of environmental projects caused Greenwise’s gross margin to tumble by 3.8 percentage points to 8% last year.

Switching lanes again

Dissatisfied with such razor-thin margins, Kwok turned his sights to the AI frenzy of the past few years, determining that infrastructure providers were set to benefit from the technology’s rise. As a result, in late June he suddenly announced a HK$2.2 billion acquisition of Shanghai Yovole Cloud Calculation — a firm specializing in cloud computing, data centers, and AI computing services, firmly staking a claim in the AI computing infrastructure realm.

Yovole Cloud's most valuable assets are its licenses for cloud computing services, along with the credentials required to procure high-performance GPUs that form the backbone of high-powered computing needed for AI.

Currently, Envision Greenwise can supply about 1.5 MW of computing power to its clients. It aims to ramp that up to 16 MW in the near term, with a goal of getting 40% of its revenue from intelligent computing services by March next year. To achieve that, the company spent 1.29 billion yuan ($193 million) on Aug. 6 to purchase around 400 servers.

But such purchases require financial firepower. To address that, the company launched a placement of 118 million new shares, accounting for 3.92% of its expanded share capital, for HK$4.66 apiece on Aug. 1, raising roughly HK$550 million. Concurrently, it issued nearly $70 million worth of convertible bonds. All told, this fundraising campaign pulled in nearly HK$1.1 billion.

Uncertain horizon

Envision Greenwise's cross-sector metamorphosis looks fraught with risk. The company lacks expertise in AI, and Kwok's professional background hardly lies in AI technology. His ability to manage and steer the company's development post-acquisition remains a major question mark.

The company's frequent pivots also raise the question of how long it will remain focused on AI, after only entering the environmental sector for three years before making this major shift. The reality is that Envision Greenwise is itself quite green in the AI computing market. Even if it can scale its capacity to its targeted 16 MW, challenging more established players will require substantially more legwork.

Complicating matters, the company’s latest direction will demand heavy capital investment. Envision Greenwise held a mere HK$230 million in cash at the end of March, though it’s shown it’s capable of raising new funds with the August share placement and bond issue. But it’s hardly clear if investors will want to continue showering this AI newbie with new funds.

Despite all that, Envision Greenwise's stock initially went on a tear in May as investors gobbled up the new AI theme, more than doubling from the HK$3 mark to HK$6.42 just a month later. But the shares plummeted after the acquisition announcement and have remained volatile since, now hovering above HK$4. That shows that shares of this type of “chameleon company,” whose business is perpetually in flux, have probably become the plaything of short-term traders looking for fast profits on each new headline. That’s hardly consolation for anyone who may be thinking of long-term investment.

To subscribe to Bamboo Works weekly free newsletter, click here

The cross-border logistics company has quadrupled the size of its IPO in response to new rules seeking to stamp out suspicious new Chinese listings, seeking to raise $33 million

Key Takeaways:

• World Road Inc.’s updated Nasdaq listing application reveals its business began to deteriorate in its latest fiscal year as a result of U.S. policy changes
• The cross-border e-commerce logistics company is seeking a valuation multiple that’s many times higher than far larger peers like S.F. Holding and UPS

By Doug Young

Sometimes delivery is all about timing. That’s certainly the case with World Road Inc., a Chinese supplier of cross-border e-commerce logistics services, which filed an updated prospectus this week for a Nasdaq IPO. The company first filed for its listing in May last year, back when shipping e-commerce items from China to the rest of the world was all the rage. But much has changed since then, which is painfully apparent in World Road’s latest financials.

The company is making the IPO bid into an environment fraught with obstacles and uncertainties. The biggest of those is coming on the trade front, as the U.S. and Europe take steps to tamp down the flood of Chinese imports pouring into their markets. But there are also political obstacles, as the U.S. and China both crack down on suspicious “pump and dump” Chinese IPOs on Wall Street. 

That crackdown is reflected in World Road’s notable quadrupling of its fundraising target in its latest IPO application. The company says it now plans to raise about $33 million in a Nasdaq listing by selling 6 million shares for between $5 and $6 each. In its original filings last year, the target was much lower, aiming to raise around $7.5 million by selling 1.5 million shares for between $4 and $6 each.

The big fundraising increase looks directly related to a new Nasdaq rule that took effect this year, requiring all Chinese companies making new listings on the exchange to raise at least $25 million. World Roald’s new fundraising target meets that threshold, though it’s far from clear this listing will make it to market.

The company still needs to get clearance from the Chinese securities regulator, which it said is still pending. That regulator, the Chinese Securities Regulatory Commission, has also emerged as an important gatekeeper trying to screen out potential “pump and dump” IPOs that were giving China a bad reputation on Wall Street.

A typical case came from a company called Pomdoctor (POM.US), which sold IPO shares last October for $4 each. The stock initially rose above $5, until one day in December, when it suddenly tanked to $0.50 from its $5.42 close the previous day. The stock has never recovered since then, last closing at about $0.83 on Thursday.

A big factor behind such spectacular crashes is inflated valuations at the time of the IPOs, which could also be the case with World Road. A pricing at the middle of its range would value the company at about $190 million, which, when combined with sales from its latest fiscal year, gives it a price-to-sales (P/S) ratio of about 3.3. While that’s not huge, it’s quite a bit higher than the 0.39 for S.F. Holding (6936.HK; 002352.SZ), one of China’s top logistics providers, and 0.93 for global giant UPS (UPS.US).

That immediately puts World Road’s shares in danger of collapse if and when the company completes its IPO, since there’s no apparent reason why it should be valued so much higher than these industry leaders.

Deteriorating financials

Making matters worse, World Road’s latest financials hardly look too encouraging. Things were quite different when the company, based in the Central Chinese city of Wuhan, first filed its IPO prospectus in May last year. Back then, it boasted booming revenue that more than tripled to 464 million yuan ($69 million) in its fiscal year through March 2025 from 130 million yuan the previous year.

At that time, cross-border e-commerce of products flowing from China to the rest of the world was booming, much of that conducted over major platforms like Shein, Temu, TikTok and AliExpress. Another factor driving the boom was the rise of Chinese brands that were finding success in Western markets, such as Anker, Aukey and Ugreen.

But in February last year the U.S. eliminated a loophole that had previously allowed packages worth less than $800 coming from China to enter the country duty-free. Europe took a similar step by imposing a temporary 3 euro customs duty on packages containing goods entering the bloc worth 150 euros ($174) or less from February this year. Such goods had also previously been allowed to enter duty free.

As those policies took effect, World Road’s revenue fell 18% to 381 million yuan year-on-year during its fiscal year through March 2026. It predicted things would continue to worsen before they get better.

“We anticipate further revenue decline over the next 12 months due to competitive pressures and uncertainties within the economic environment,” it said in its prospectus. “Unstable tariffs on Chinese imports introduced in April 2025 are expected to disrupt cross-border trade.”

Logistics has never been a very profitable business to start with, which is reflected in World Road’s low gross margins. What’s more, its gross margin has been dropping steadily amid all the cross-border trade frictions, falling to 6.2% in its latest fiscal year from 6.4% the previous year and 7.1% the year before that.

On the bottom line, World Road’s profit fell 3.7% in its latest fiscal year to just 8.6 million yuan from 8.93 million yuan a year earlier. The smaller rate of profit decline compared with the company’s revenue decline owed mostly to aggressive cost cutting, which is commendable but hardly a reason to be positive about this company.

It’s also slightly notable that the name of World Road’s IPO underwriter is redacted in the latest prospectus, unlike earlier versions that named Craft Capital Management and R.F. Lafferty. That’s significant because in March a U.S. congressional committee sent letters to three other small investment banks looking into their potential role in underwriting suspicious IPOs by small Chinese companies. So it’s not surprising that these small underwriters want to stay as low-profile as possible, and would quite likely flee from any Chinese listing at the first sign of trouble.

It’s obviously too early to say if that kind of trouble is on the horizon for World Road. But the high valuation it’s seeking, combined with its deteriorating finances, certainly don’t bode well for its IPO.

To subscribe to Bamboo Works weekly free newsletter, click here

The company’s bitcoin mining business held steady in the second quarter, as its first high-performance computing center was ready to receive customers in July

Key Takeaways:

• Cango completed its overhaul of a high-performance computing center in Georgia in July, as it signed the first customer for its new AI services
• The company continues to refine its core bitcoin mining business by phasing out older machines and experimenting with leased capacity

By Doug Young

First there was the storm. Then there was the post-storm clean-up. Now the rebuilding begins.

That sums up the recent turbulent history for Cango Inc. (CANG.US), which, in the space of just nine months has gone from the brink of a liquidity crisis to a herculean effort to right its corporate ship. Its latest financial results show the company’s finances returned to stable footing in the second quarter, as it forged ahead with a new AI-related business model.

The company’s latest report, released Aug. 31, contained some important new developments on the AI front, including the signing of the first customer for the high-performance computing (HPC) centers Cango is setting up within one of its existing bitcoin mining facilities. The company also competed an overhaul of a portion of an existing mining center in the U.S. state of Georgia, converting roughly 3 MW of the site's 50 MW of capacity, positioning it to become its first such operational HPC facility.  

It should come as no surprise that Cango’s recent turbulence is directly tied to its embrace of bitcoin mining as a business model in late 2024. The company was a car trader in China before that, but ditched that sputtering business in favor of bitcoin mining as the cryptocurrency traded at record highs. But then bitcoin prices crashed starting late last year.

The company responded by selling off a big portion of its bitcoin reserves starting in January. It used that cash to pay down its long-term debt, which dropped to $30.6 million by the end of March from $557.6 million just three months earlier. Its latest report showed that long-term debt level remained low, at $31.2 million at the end of June.

Going forward, CEO Paul Yu said Cango will continue to run its bitcoin and AI operations as “parallel businesses.” But the focus seems to be shifting to HPC centers as bitcoin prices remain stubbornly low, despite a recent rally of more than 20% for the cryptocurrency since the start of July. 

Cango said the average cost for each bitcoin it mined stood at $73,313 in the second quarter, down 5% sequentially, as the company shifted its mining strategy, which we’ll discuss in more detail shortly. Significantly, that cost is below the latest bitcoin market price of about $77,500, meaning Cango is spending less to mine each coin than the currency’s actual value. But we should also point out the company’s all-in mining cost of $98,405 per bitcoin during the quarter was still well above the market price.

“In our bitcoin mining business, we continue to focus on unit economics rather than scale,” said Yu. “At the same time, we continued to deliver on our AI modular build at our LN mining site,” he added, referring to the company’s first HPC center in Georgia.

First customer onboard

The Georgia facility is one of dozens of sites on three continents that Cango currently operates, mostly as bitcoin mining facilities. It’s using the Georgia facility as a proof of concept, aiming to show it can host the heavy-duty computing power needed to run AI applications. The network of centers Cango envisions today are well suited for an emerging field of more company- and industry-specific agentic applications that require less power, but it doesn’t not rule out expanding into larger-scale AI inference workloads as its infrastructure and GPU fleet evolves over time.

The HPC centers are being operated under Cango’s new EcoHash unit, which it set up last year in the U.S. state of Texas. Cango said the Georgia site’s conversion was completed in early July, and now has infrastructure that can support up to 3 MW of computing power, with room for future expansion. Necessary hardware has been procured and is being installed in batches to support a phased ramp-up of the site.

And perhaps most significantly, Cango said the site has signed its first customer, while the company continues discussions with several others. As a result, it expects to start generating its first revenue from the AI business in the third quarter. Outside Georgia, Cango added it has already begun operating other AI test nodes in Texas and on the U.S. West Coast to serve customers with proximity-based deployment needs in those regions. It said it also continues to evaluate other potential new sites, as well as the possibility of building its own new HPC facilities.

New business from the AI operation would be a welcome addition for Cango, whose revenue from its core bitcoin mining operation has been rapidly shrinking as it focuses on more efficient mining. The company generated $50.8 million in revenue during the second quarter, most of that from the bitcoin operation, which was down roughly by half from $102 million in the first quarter. The company attributed the drop to phasing out some of its older, less efficient mining machines, and shifting a portion of its capacity from self-mining to a hosted leasing model.

As it shifted its focus to greater efficiency, Cango mined 656 bitcoins during the quarter, averaging about 219 per month, down from its monthly average of 422 bitcoins in the first quarter. It operated 27.58 EH/s of mining capacity as of June 30, also down from 37 EH/s as of March 31. The company held 1,056 bitcoins in its treasury at the end of June, similar to the 1,026 it had at the end of March, but down dramatically from nearly 7,500 in January this year before it started selling its holdings.

The company also disclosed that it has started to execute a hedging strategy as a buffer against bitcoin price volatility, though CFO Simon Tang stressed the move was purely for risk management and not for speculative purposes.

The focus on greater efficiency led to dramatic improvements in the company’s profitability metrics. Its loss from operations narrowed sharply to $80.6 million in the second quarter from $254.4 million in the first, while its adjusted EBITDA loss fell to $10.7 million from $154.1 million over that period. And on the bottom line, Cango’s net loss of $81.6 million from continuing operations also marked a big improvement from a $261.1 million loss on that basis in the first quarter.

“Looking into the second half, our priorities are: managing the mix of self-mining and leased hashrate prudently; executing our first AI deployments and continuing to sign new customers; and building on the operating experience from Georgia as we evaluate further site expansion,” said CEO Yu. “Capital discipline and operating efficiency remain our priorities.”

The Bamboo Works offers a wide-ranging mix of coverage on U.S.- and Hong Kong-listed Chinese companies, including some sponsored content. For additional queries, including questions on individual articles, please contact us by clicking here

To subscribe to Bamboo Works free weekly newsletter, click here

The global leader in cellular connectivity chips has filed for a Hong Kong listing as it accelerates its push into the higher-margin business of specialized ASIC chips

Key Takeaways:

• ASR Microelectronics logged a 29% rise in first-half revenue and turned a small profit for the period, helped by rising demand for application-specific chips
• The company still makes most of its money from general-purpose chips, but custom ASIC solutions rose to 13% of turnover, with a big order backlog

By Lee Shih Ta

China’s biggest provider of cellular chips for smart devices, ASR Microelectronics Co. Ltd. (688220.SH), is aiming to tap Hong Kong’s equity market as it shifts its focus towards premium products for AI-enabled connectivity and computing.

Nearly five years after joining Shanghai’s STAR Market, the company filed for a secondary listing on the main board of the Hong Kong Stock Exchange on Sept. 11, citing plans to invest in next-generation chip design.

The company has come a long way since its Shanghai equity debut in 2022, becoming the world’s biggest supplier of cellular connectivity chips by shipment volume last year, according to industry data cited in its prospectus. ASR also passed a profit milestone in the first half of this year, while expanding from general-purpose processors into higher-margin ASIC chips that are customized for specific tasks.

ASR founder Vincent Tai has a long pedigree in the chip industry. He led an earlier firm, RDA Microelectronics, to aNasdaq listing before the company was acquired by tech conglomerate Tsinghua Unigroup. Tai went on to establish ASR in 2015 and remains in the roles of company chairman and chief strategy officer.

ASR acquired Marvell Technology’s mobile business unit in 2017 to strengthen its capability in cellular baseband, which connects a device to networks, and processor applications. A year later it secured a cloud AI chip project from a leading AI company. In 2024 ASR became the world’s largest supplier of medium-speed Category 1 chips, going on to capture 37.8% of the global cellular connectivity chip market by volume in 2025.

Earnings lag

Still, the scale of its shipments has yet to translate into plentiful profits. Revenue rose from 2.6 billion yuan ($388 million) in 2023 to 3.82 billion yuan in 2025, while the company posted net losses of 506 million yuan, 693 million yuan and 390 million yuan over the three-year period. Heavy R&D spending and low prices per unit have weighed on profitability. The average selling price of ASR’s wireless connectivity chips fell from 12.6 yuan in 2023 to 10 yuan in 2025, while annual R&D expenses exceeded 1.1 billion yuan throughout the period.

But the picture began to shift this year, as ASR gained greater traction with higher-end products. Revenue rose 29.1% to 2.45 billion yuan in the first half of 2026, while gross profit jumped 72.3% to 700 million yuan and gross margin increased to 28.5% from 21.4%. The bottom line swung to a profit of 84.24 million yuan from a loss of 245 million yuan a year earlier, although the company remained 58.67 million yuan in the red excluding non-recurring gains and losses. Mass production of 5G products pushed the average selling price of wireless connectivity chips to 11.3 yuan from 9.4 yuan. Meanwhile, revenue from custom ASIC solutions increased to 322 million yuan from 126 million yuan, taking the segment’s overall revenue share to 13.1% from 6.7%, with a gross margin of 36.9% surpassing the average for the whole company.

The customized chip business has given ASR another way to monetize the design expertise it has built over the years. Unlike standard chips, which rely on high shipment volumes to spread R&D costs, ASICs can be designed to specific requirements and generate extra revenue once they enter volume production. As of the end of June, ASR’s custom ASIC order backlog exceeded 1.5 billion yuan, a year-on-year rise of 416.5%. The firm had also delivered dozens of projects, including 4nm node designs.

Industry trend

The strategy also reflects a broader shift in the global chip industry towards purpose-built AI chips, as technology companies try to cut costs and become less reliant on general-purpose GPUs. Broadcom generated $16.7 billion in AI semiconductor revenue in its latest fiscal quarter, up 221% from the year-earlier period. MediaTek, whose core business is built around smartphone chips, is similarly extending its technology into AI data centers, with its first custom AI chip scheduled to enter production in the fourth quarter of this year.

ASR benefits from experience in cellular baseband, complex System on a Chip (SoC) design and volume production, but the project-based ASIC business comes with inherent uncertainties. ASIC revenue fell to 217 million yuan in 2025 from 336 million yuan in 2024, only rebounding in the first half of this year. The current order backlog can only translate into sustainable profits if customers come back with further orders after receiving their first round of products.

As ASR targets premium solutions, some of its early shareholders have begun to cash out. Alibaba (9988.HK; BABA.US), which invested in 2017 through a tech subsidiary, sold a 3% ASR stake last year and another 1.86% this year, leaving it with a holding of around 10.58%. The sales are part of a broader wave of divestment from semiconductor companies with mainland listings. Seven companies, including AMEC, Montage Technologyand ASR, disclosed shareholder sell-down plans on May 22 covering an estimated 12.69 billion yuan based on share prices at the time. Meanwhile, despite its first-half profit, ASR drew on about 378 million of net cash for operating activities in the period, rising from 272 million yuan a year earlier, mainly due to increased inventories, prepayments and receivables.

Valuations among Hong Kong-listed AI chip stocks vary widely. Pure-play AI GPU company Iluvatar CoreX (9903.HK) trades at a price-to-sales ratio of about 44 times, compared with about 12.6 times for intelligent-driving chip company Horizon Robotics (9660.HK) and about 5.9 times for Black Sesame (2533.HK). By comparison, ASR has a market capitalization of about 37.2 billion yuan and trades at about 8.5 times its trailing 12-month revenue.

The market still regards ASR as a maker of cellular connectivity chips, which accounted for more than 70% of its first-half revenue. Rising ASIC income and higher gross margins could lift the valuation over time, but the question remains whether the company’s track record in chip design can be turned into a higher-value, next-generation business.

To subscribe to Bamboo Works weekly free newsletter, click here

The company is planning a Hong Kong listing, just nine months after its Shanghai trading debut, aiming to expand its financing capabilities during a period of rapid growth

Key Takeaways:

• Moore Threads has reportedly submitted a confidential application for a Hong Kong listing, planning to raise at least $1 billion
• The GPU maker’s revenue surged by 147% in the first half of the year, but it still requires major funds to sustain its cash-intensive operation

By Lee Shih Ta

China's leading GPU makers, producing the “brains” that power AI, are transitioning from a phase of technological verification to large-scale commercialization, consuming billions of dollars in that process. In that race for funds to keep their business humming, Moore Threads Technology Co. Ltd. (688795.SH), one of the country’s major players, has reportedly submitted a confidential application for a Hong Kong IPO, with plans to raise a minimum of $1 billion.

The step is noteworthy because Moore Threads’ stock just debuted on the STAR Market in Shanghai last December, where it raised nearly 8 billion yuan ($1.19 billion). But apparently that’s just a quick fix for such a heavily cash-burning company, prompting it to eye raising a similar amount from Hong Kong’s more internationally focused investor pool.

Moore Threads is just one among a group of Chinese GPU makers hitting up the capital markets for funds lately. In the past year alone, MetaX (688802.SH), Biren Technology (6082.HK), Iluvatar CoreX (9903.HK) and Enflame Technology (688801.SH) have all gone public in Shanghai and Hong Kong. All are hoping to become the “Nvidia (NVDA.US) of China,” tapping booming demand for AI training and inference, coupled with domestic substitution by Chinese AI developers looking to wean themselves from foreign products.

Growing capital requirements

At the end of June, Moore Threads had already allocated nearly 5 billion yuan of the 7.58 billion yuan it raised from its Shanghai IPO, leaving about 2.73 billion yuan in a special account. A Hong Kong listing would help it tap the global capital market, providing greater financial flexibility for its subsequent R&D and commercialization efforts, as well as for potential M&A.

Capital requirements for GPU development typically ramp up when products enter large-scale commercialization. By that time, development of the next generation of chips has already begun, while the transition from single-card setups to clusters containing thousands of cards and inventory stocking all further add to the need for capital. Stock market investors seem more than happy to provide such funds, as reflected by an oversubscription rate of more than 6,100 times for Enflame's STAR Market IPO last week that raised about 6.1 billion yuan for its fifth- and sixth-generation AI chips.

Moore Threads recorded revenue of 1.74 billion yuan in the first half of this year, up 147% year-on-year, exceeding the figure for all of 2025. Its gross profit rose 104% to 989 million yuan, while its net loss narrowed sharply to just 11.56 million yuan from 271 million yuan a year earlier. But after deducting non-recurring gains and losses, such as government subsidies and returns on financial assets, the company still posted a loss of 151 million yuan in the latest period. Commercialization of its MTT KUAE intelligent computing cluster product is accelerating, and its flagship product, the MTT S5000, has also achieved large-scale sales.

While Moore Threads’ revenue is growing quickly, so are its product development costs. Its R&D spending in the first half of the year reached 769 million yuan, equal to 44.3% of its revenue, bringing cumulative R&D spending since 2022 to nearly 5.9 billion yuan. Meantime, the company’s net cash outflow from operating activities stood at 2.17 billion yuan in the first half of the year, nearly double its 1.16 billion yuan outflow a year earlier, as it expanded its production scale and incurred growing procurement costs.

In such an environment, financing capabilities have become an integral part of what makes each GPU company competitive.

The industry has plenty of room for growth, as China has proposed increasing the supply of high-performance intelligent computing resources, building ultra-large-scale intelligent computing clusters, and cultivating an independently controllable software and hardware ecosystem under its “15th Five-Year Plan” that launched this year. The inclusion of such targets in the national development roadmap means the government often makes copious funds available to meet those goals through various channels.

From opportunity window to competition

As the market rapidly develops, the landscape is also rapidly changing. Reuters Breakingviews estimates that Nvidia’s share of China's $90 billion AI semiconductor market has dropped from a near monopoly a few years ago to roughly 55% now. Domestic manufacturers such as Moore Threads, MetaX, Biren Tech, and Enflame are increasingly picking up more of the market.

The competition has also extended from chips into software, where Nvidia's CUDA developer framework retains a significant advantage. By comparison, Moore Threads utilizes its Meta-computing Unified System Architecture (MUSA) software stack and related tools to reduce migration costs from CUDA programs. Its system supported more than 800,000 developers at the end of June, while achieving cluster deployments numbering 10,000 cards and more.

Investors are willing to pay high premiums for domestic GPU makers, evidenced by Enflame's high 61.8 price-to-sales (P/S) ratio for its 2025 sales. By comparison, Nvidia was valued at a “lowly” 25.4 times for the same period. But those sky-high valuations are almost certain to come down as more domestic GPU manufacturers enter the public market, giving investors more choice.

Moore Threads’ stock surged for a period after its Shanghai listing, though the shares have corrected recently in tandem with the end of a lock-up period for the first batch of offline placement restricted shares. While the shares unlocked on Sept. 7 only accounted for about 5.5% of the company’s total share capital, they increased the tradeable float by roughly 85%, causing a major selloff over the next two trading days.

The growing supply of tradeable shares has also compelled investors to reevaluate the value of Moore Threads’ shares. A Hong Kong IPO will expand its investor base and add a new capital source for future funding and expansion, while also further testing appetite for the stock.

Moore Threads is currently in expansionary mode. But as more domestic GPU manufacturers enter the market, valuation support provided by the scarcity factor will quickly diminish. That means that factors like revenue growth, cash flow, product iteration, customer deployment, and software ecosystems will become increasingly important as investors seek to gauge each company’s true worth.

To subscribe to Bamboo Works weekly free newsletter, click here

The vision sensor maker is exploring a Hong Kong listing as weaker sales to core client Ant Group and rising development costs weigh on its revenue and profits

Key Takeaways:

• Orbbec’s revenue growth slowed from 104.1% in the first half of 2025 to just 0.5% in the same period this year, even as its robot-vision sales more than doubled
• Higher spending on research, sales and administration has tested the company’s recent return to profitability

By Hu Minghe

Artificial brains that tell them where to go, what to do and how to do it may be the most important component in an emerging generation of humanoid and industrial robots. But also important are the sensors that help them execute those actions.

An emerging maker of those components is Orbbec Inc. (688322.SH), whose vision sensors help robots “see” where they are going and what they need to pick up. Now, the company is hoping to see its way to a Hong Kong IPO, complementing its existing Shanghai listing, as it searches for new investors to feed its research-heavy operation.

The company announced it was exploring a Hong Kong listing as early as July, following a recent trend by companies already listed on China’s domestic markets in Shanghai and Shenzhen seeking to tap Hong Kong’s more international investor pool.

Shenzhen-based Orbbec looks like a potential beneficiary of the movement by smarter machines into factories, warehouses and everyday life. But while its robotics business is growing rapidly, its latest financial report tells a less exciting story.

The company’s revenue barely rose in the first half of 2026, up just 0.5% to 437.6 million yuan ($65.2 million). That represents a huge slowdown from the 104.1% growth to 435.5 million yuan in the first half of 2025. Orbbec’s profit in the first half of this year also fell 30.8% to 41.6 million yuan.

The gap between booming enthusiasm for its robotics and Orbbec’s overall lackluster results reflects the company’s own transition from its current business mix and the one it is trying to build. Its cameras and other vision products traditionally performed more mundane functions like scanning and payment applications, a relatively mature business. Gains in newer, high-growth areas like robotics have yet to replace lost sales in those more mature areas, while the cost of developing and selling new products keeps rising.

That puts pressure on a turnaround that has only recently taken hold. Orbbec earned a profit of 127.9 million yuan on revenue of 940.7 million yuan in 2025, its first annual profit since listing on Shanghai’s STAR Market in 2022.

Robotics reboot

There are good reasons for its recent move into robotics. Advances in AI are helping robots learn tasks and respond to spoken instructions, raising the prospect of machines that can handle more varied work. Businesses also want automation to fill labor gaps and take over repetitive jobs, according to the International Federation of Robotics. Cameras such as Orbbec’s, which measure distance and identify objects’ positions, supply information those machines need to act on their surroundings.

Its latest financial report names Chinese humanoid robot maker Leju among Orbbec’s camera customers. Another customer for its sensors is reportedly the high-profile Unitree, whose humanoids and robot dogs have capture public imagination with their acrobatics and other performances, according to Chinese media.

But such high-profile demonstrations are largely for show and don’t often result in large orders. Instead, Orbbec’s management points to robots performing more mundane tasks like cleaning, delivery and materials handling as nearer-term opportunities because their work is relatively standardized.

Meanwhile, the company’s older applications, while less sexy, still matter. Orbbec attributed much of its first-half growth last year to products used for scanning and payments. Meantime, management has said uneven demand for traditional biometric applications like face and fingerprint recognition were holding back overall revenue growth.

Ant Group, the Alibaba affiliate behind Alipay, is a key customer and major cause behind this year’s sales slowdown. Orbbec’s sales to Ant and its affiliates fell to just 53 million yuan in the first half of 2026 from 130.9 million yuan a year earlier. Sales to other customers rose 26.3% over that time, adding roughly 80 million yuan. But that barely replaced the 78 million yuan in lost business from Ant.

Orbbec says its robot-vision revenue more than doubled in the first half of this year, without disclosing the amount. The strong growth is encouraging, though the failure to give a sales figure means its contribution is probably still relatively small.

Sales slowdown

Orbbec’s slowdown actually dates back to last year, when revenue growth eased to about 6% in the final quarter of 2025. It grew by a similar amount in the first quarter of this year, before slipping into contraction in the second quarter. As that happened, the company’s profit also swung from growing by 27.5% year-on-year to 31 million yuan in the first quarter of 2026, to plunging 70.3% to 10.6 million yuan in the second.

On a more positive note, the company’s gross margin improved from 41.3% in the first half of 2025 to 48.5% in the first half of 2026 as higher-margin products made up more of its business. But expenses, whose growth is outpacing revenue growth, ultimately undermined the company’s profits. In particular, R&D spending rose 22.2% to 111.4 million yuan, consuming about a quarter of revenue.

Management is asking investors to be patient, arguing that years of development are beginning to pay off. The company is also expanding into equipment that records demonstrations used to train robots. Last month, it announced a partnership with robotics startup Lingchu Intelligence, also known as PsiBot, to combine its recording hardware with the startup’s data software. The aim is to make collecting and managing robot-training data easier, opening another market for Orbbec’s hardware.

Funding that expansion is probably a factor behind Orbbec’s potential Hong Kong listing. The company raised 980 million yuan via a share placement through its Shanghai listing in June, mainly for research and manufacturing projects with a combined budget of nearly 2 billion yuan. It is also building a Vietnam factory expected to begin production in 2027.

Bocom International still expects the company to bounce back from its recent revenue and profit contractions. The bank cut its 2026 revenue forecast for Orbbec by 24% in August, but its revised estimate of 1.23 billion yuan would still represent growth of about 30% from 2025. It expects new robot-vision and training-data products to help sales in the second half of the year.

Investors already place considerable value on that potential. Orbbec currently trades at about 39 times 2025 sales, even after its shares fell about 40% over the last two months. That’s well ahead of the roughly 26 times for Mech-Mind (9615.HK), the industrial robot-vision company that listed in Hong Kong this month. Orbbec had more than twice its rival’s revenue in 2025, although their business mixes differ.

A Hong Kong listing could give Orbbec more money to pursue the robot opportunity that looks like the best bet to jumpstart its growth. Whether its robotics business can grow fast enough to replace lost sales elsewhere and translate to stronger profits remains the open question for Hong Kong investors.

To subscribe to Bamboo Works weekly free newsletter, click here

The solar panel maker will drop the “solar” from its English name, as it builds up a second business pillar investing in frontier industries

Key Takeaways:

• JinkoSolar is adding a second business pillar by investing in emerging high-tech industries like AI, complementing its struggling legacy solar business
• The company’s pivot includes early investments of typically 100 million yuan or less in AI startups Moonshot, StepFun and SiliconFlow

By Doug Young

The old adage “what’s in a name” is taking on new significance in what at first glance appears to be a somewhat cosmetic name change by leading solar panel maker JinkoSolar Holding Co. Ltd. (JKS.US). But a closer look at the announcement, which says the company will drop the words “solar” and “energy” from its English and Chinese names, respectively, turns out to be quite a game changer for one of China’s oldest solar manufacturers.

The company revealed that its main solar business will become one of its two main pillars, alongside a new second pillar that invests in emerging high-tech industries like AI, advanced materials and other “frontier technologies,” according to its announcement last Wednesday. Reflecting that shift, its board has voted to change the company’s English name to Jinko Holdings Ltd., losing the word “solar.” Similarly, its new Chinese name will lose the word “nengyuan,” or “energy.”

The big backstory is that China’s solar sector is suffering from massive overcapacity due to a huge state-led buildup over the last decade. That’s caused prices for solar panels and materials used to make them to plunge, sending companies throughout the supply chain deeply into the red.

China is taking steps to try fix the situation by closing or consolidating many of the smaller players, and forcing everyone to shut down capacity used to make older, less efficient panels and solar components. But the process has been quite slow, and investors lost their patience with this group of companies long ago. What’s more, it’s not at all clear that the government will continue to support all the current manufacturers through this difficult transition, meaning one or more major players may ultimately be forced to leave the sector.

All that has weighed on solar stocks, many of which have lost half or more of their value over the last year. JinkoSolar is typical of the group, with its stock down 57% this year alone, now trading at an eight-year low. The company also passed a dubious milestone in mid-June, when its market value dropped below the $1 billion mark that defines “unicorn” companies in emerging sectors. That said, some might argue solar is already quite mature, and companies from the sector lost their eligibility to be called “unicorns” long ago, regardless of their market value.

Investors were unimpressed with JinkoSolar’s latest pivot, as the stock sagged by a further 8% in the three trading days after last week’s announcement. Even analysts, normally a pretty positive group, are quite bearish on the company and its peers. Of the seven polled by Yahoo Finance, just one rates the company a “buy,” while four rate it a “hold” and one rates it an outright “sell.”

The company’s core solar business, which will continue to be held under its Shanghai-listed Jinko Solar Co. Ltd. (688223.SH) subsidiary, saw its revenue decline 45% between a peak in 2023 and last year. The figure fell another 31.3% year-on-year to 12.4 billion yuan ($1.85 billion) in the second quarter of this year. Jinko also fell into the red last year, and reported a net loss of another 697 million yuan in the second quarter of 2026.

Venture investor

Despite its woes, Jinko remains relatively cash rich, which is apparently what it will use for its new second pillar as a venture capital investor. The company had about 17 billion yuan in cash at the end of June, though that was down from 22.8 billion yuan just three months earlier. Part of the decline inevitably owes to the challenges the company is facing for its solar business, but part probably also owes to its new role as a venture investor.

"The proposed new name reflects what the company has become, a holding company anchored in a world-class solar and energy storage business and building a second engine of value creation through strategic investment," said CEO Du Dimi Du. 

Jinko quietly alluded to the change in its second-quarter earnings announcement in late August, which contained a section on “strategic investment highlights” with information on investments in two companies, LaPlace Renewable Energy and Hangzhou Gold Electronic Equipment. The latest announcement reveals that Jinko has been actively making minority investments in startups since the beginning of this year, with about 400 million yuan placed in eight companies so far, including LaPlace and Hangzhou Gold.

Among its other six investments, the three largest were about 100 million yuan invested in StepFun, 70 million yuan in SiliconFlow and 69 million yuan in Moonshot. Not surprisingly, all three of those are startups from the red-hot AI sector where company valuations are soaring as they consume huge amounts of cash to develop their products. Jinko’s other investments so far include 15 million yuan in Noetix Robotics, and 8.6 million yuan in Jabon Metallic Materials.

Jinko points out that it has already exited the LaPlace investment, netting 250 million yuan in realized gains, and that it recorded a 400 million yuan fair value gain for its investment in Hangzhou Gold, which recently completed its listing on Shenzhen’s ChiNext board.

The big issue with this type of investment is its huge level of risk, which can be difficult for less experienced investors like Jinko to manage. It’s quite easy to make money on such investments in the current market when valuations are soaring. But once the situation reverses, which many believe is inevitable, valuations are likely to come crashing back to earth. That could leave Jinko with huge losses if its investments fall below their valuations at the time it purchased its stakes.

Jinko isn’t the only solar company looking for relief from the struggling sector. In June, solar materials maker Daqo New Energy (DQ.US; 688303.SH) announced a similar pivot to AI data center electrical equipment, drawing on part of its past before it moved into polysilicon manufacturing for solar panels.  Investors haven’t been too impressed with that pivot either, with Daqo’s U.S.-listed shares losing about 30% of their value in the three months since that announcement.

At the end of the day, we can’t get too excited about these new pivots, even though we understand why the solar companies are taking these steps. In effect, both Jinko’s and Daqo’s moves look aimed at jumping from one burst bubble into another inflating bubble that’s almost certain to burst as well. Any bursting of the second AI bubble will only add to the growing losses from their solar business.

To subscribe to Bamboo Works weekly free newsletter, click here

Following back-to-back listings by Z.AI and MiniMax, the startup behind the Kimi chatbot is reportedly seeking a $50 billion valuation through a Hong Kong IPO

Key Takeaways:

• Moonshot has reportedly submitted a confidential filing for a Hong Kong IPO, seeking to raise $3 billion at an astronomical $50 billion valuation
• The move aims to tap into Hong Kong’s deep capital pool as the company behind the Kimi chatbot needs large funding to run its costly operations

By Warren Yang

When it comes to corporate naming, Chinese AI pioneer Moonshot AI certainly doesn’t lack ambition. Now, the Beijing-based startup behind the wildly popular Kimi chatbot is attempting a meteoric leap of its own.

Last week, Moonshot, founded just three years ago, reportedly submitted a confidential filing for a Hong Kong IPO, seeking to raise $3 billion at an astronomical $50 billion valuation. If successful, Moonshot will stand alongside peers Z.AI Co. Ltd. (2513.HK), also known as Zhipu, and MiniMax Group Inc. (0100.HK) as the leading-edge of Chinese large language model (LLM) startups going public.

The companies rely on private equity and capital markets to fund their hugely expensive operations, set apart from other major large model developers like Qwen and Doubao, which have the backing of internet giants like Alibaba and ByteDance.

Moonshot’s choice of Hong Kong over Shanghai or Shenzhen signals its desire to go for deep offshore institutional capital, similar to MiniMax and Z.AI. For an AI developer that needs billions of dollars to bankroll high-end computing hardware, cloud infrastructure and top-tier engineering talent, the vast international funding pool that Hong Kong offers should be an irresistible draw.

In addition to its deep funding pool, Hong Kong’s Chapter 18C listing regime, designed specifically for specialist technology companies, provides a clear regulatory pathway for IPOs by commercial-stage high-tech firms that are still pouring heavy capital into R&D.

Under traditional rules, companies must demonstrate a track record of net profits or substantial revenues before they can list in Hong Kong. Chapter 18C removes these hurdles by creating tailored entry criteria for high-tech sectors like AI. That allows fast-growing AI pioneers like Moonshot to raise public capital to fuel their costly expansion without waiting years to show profits. Listing in Hong Kong can also help Chinese startups raise their international profiles as they look to market their products abroad.

Surging valuation

Moonshot has certainly lived up to its name in terms of its surging valuation. The company was worth $4.3 billion during a funding round late last year, and the figure more than quadrupled to $20 billion by May. Now, it’s aiming to more than double that just months later with a target of $50 billion.

The driver behind its explosive ascent is the monetization of its Kimi series of large language models, with its annualized recurring revenue tripling to $300 million in June from just three months earlier, according to media reports.

A key appeal of Moonshot’s products is that they dramatically cut computing expenses by running an efficient design that activates only the necessary parts of its AI system for any given job. Because the model uses much less computing power and memory, Moonshot can deliver premium-quality products to corporate clients at a fraction of the prices charged by Western competitors. Kimi’s ability to process ultra-long text at low costs has made it an attractive proposition for anything from financial analysis to legal reviews and software code generation.

Despite their impressive growth, Moonshot and its Chinese peers face greater difficulty raising funds than their Western competitors. As a result, while Silicon Valley leaders regularly secure multibillion-dollar funding rounds, top Chinese startups have to operate on leaner balance sheets.

Expanding restrictions on Western institutions investing in Chinese tech firms amid rising U.S.-China tensions are widening this funding gap. Lacking easy access to foreign mega-funding, Chinese AI developers are forced to rely on a much smaller pool of domestic capital from local funds, state-backed guidance vehicles, and tech giants like Alibaba and Tencent.

So Moonshot is following in the footsteps of Z.AI and MiniMax, which made back-to-back debuts in Hong Kong’s equity market at the start of this year to become the first pure LLM makers to go public in the financial hub.

AI hype

Z.AI raised HK$4.35 billion ($559 million) in early January, and MiniMax took in HK$4.80 billion the next day. Both offerings were oversubscribed more than 1,000 times by retail investors, underscoring overwhelming demand for pure AI plays. The duo didn’t stop there. They launched much larger follow-on share placements immediately after their six-month lockup periods expired in July, with Z.AI securing HK$33.6 billion, the largest amount ever for a secondary equity offering in Hong Kong, and MiniMax raising a fresh HK$16 billion.

Moonshot isn’t alone in trying to capitalize on the AI hype. Among others, 01.AI has undertaken a corporate restructuring as it works towards a Hong Kong listing next year. Baichuan Intelligent Technology, founded by Sogou search engine creator Wang Xiaochuan and backed by Alibaba and Tencent, is exploring a path to a dual listing on China’s domestic A-share markets and in Hong Kong, while StepFun, led by former Microsoft executive Jiang Daxin, is targeting a Hong Kong IPO this year or next.

This rush, however, may end up creating a glut of AI stocks, making investors more selective. That means financial and technical realities will become an increasingly important differentiator.

The annualized recurring revenue metric thrown around by AI startups often blurs the line between traditional software-as-a-service (SaaS) subscription revenue and simple annualized run rates derived from monthly usage.

Because LLM consumption can fluctuate significantly based on short-term developer experimentation or promotional credits, an annualized run rate may overstate long-term revenue durability.

Furthermore, cloud infrastructure costs are a significant drag on developers’ profitability. Unlike traditional software vendors that enjoy fat gross margins, LLM makers face enormous, ongoing server and power expenses to run workloads at scale. At the moment, Moonshot isn’t profitable because of these heavy costs, with the low prices it charges for its products further undercutting its margins.  

In scrutinizing Moonshot’s prospects, investors will want clear visibility into metrics like customer retention rates and gross margins after server and bandwidth costs.

Moonshot’s ambitious IPO plan may materialize, but to satisfy public shareholders over the long haul, it will need to prove that it will eventually be able to turn durable bottom-line profits. Both Z.AI and MiniMax are also loss-making, but their shares command sky-high valuations. Z.AI shares trade at a price-to-sales (P/S) ratio of a whopping 577, while the figure for MiniMax is about 175.

Achieving a $50 billion valuation would require a similarly high multiple, and Moonshot may well succeed in that regard. But if it fails to generate significant cash on its own, and keep the figure growing at triple-digit rates over the near-term, its valuation could come back down to earth as dramatically as the moonshot behind its name.

To subscribe to Bamboo Works weekly free newsletter, click here

"With things like this in China, there's always an element of surprise to some extent simply because the system is so opaque that nobody knows what's going on." – on the sudden legal actions in the Evergrande case

Key Takeaways:

• Chinese state subsidies driving the humanoid robot sector are likely to prompt protectionist steps in the West
• The coordinated liquidation of Evergrande and sentencing of its founder reflect Beijing's priority of maintaining social stability

By Rene Vanguestaine and Doug Young

We're currently witnessing two very different, yet equally telling, chapters of China's state-managed economy playing out simultaneously. On one hand, Beijing is heavily subsidizing a futuristic humanoid robot industry, utilizing a familiar playbook that has previously upended global markets. On the other, the government is finally closing the book on fallen real estate titan Evergrande (3333.HK), liquidating it after years of careful, behind-the-scenes management. These two developments perfectly illustrate how the Chinese government pulls the levers of its economy to manufacture technological dominance while engineering social stability.

A recent eye-opening report from the Financial Times revealed exactly who is buying up many of the dancing and leaping humanoid robots suddenly coming out of China. Unsurprisingly, it turns out the answer is the government. A sizable portion of these companies' revenues comes from government-owned training centers. These centers buy the robots, generate vast amounts of operational data using them, and send that data back to the manufacturers to rapidly improve their performance. In one case, a company called Leju received 45% of the revenue for its flagship humanoid model last year directly from these government centers.

In China's humanoid robot industry, this kind of comprehensive state support undeniably works. Chinese companies receive substantial subsidies at the central, provincial, and local government levels. Throwing such substantial amounts of money directly and indirectly at these efforts helps to build a highly efficient supply chain and ecosystem. Inevitably, some companies thrive in this environment and decide to conquer the world, armed with highly competitive prices.

But this common refrain isn't going to make robot makers outside of China very happy. We've seen this exact Chinese playbook unfold in the solar and EV industries. Western governments repeatedly complain about unfair state support, even if it isn't always the most efficient use of capital. In response to this new wave of robots, we expect to see inevitable pushback. Governments in the U.S., Europe, and Japan will likely take belated measures to protect their domestic manufacturers, such as Tesla (TSLA.US) with its Optimus humanoid, from total obliteration.

Despite the glitzy videos of kung fu robots and machines breaking human running records, the reality is less glamorous. Away from the hype, these robots may still be very prone to making mistakes. There's not a whole lot of use for a kung fu robot in daily life. We believe the industry will split into two paths: industrial and personal use. On the industrial side, there are plenty of applications that can tolerate occasional, or even multiple, failures because the risk of harming humans is incredibly low.

However, on the personal usage front, we're far from mass adoption. All it will take is one spectacular incident — we wouldn't call it an accident, but an incident — where a robot beats up an elderly person in a hospital bed, for the industry to face massive trouble. We shouldn't forget how the Segway at the turn of the century was supposed to revolutionize the world. Similarly, the low-altitude economy of flying machines lost its momentum the moment a small plane crashed into a building in Beijing. Visionaries will always promise that technology will make life entirely peachy, but the safeguards simply aren't there yet for mass adoption.

A carefully orchestrated end for a real estate giant

Moving from the industries of tomorrow to the problems of yesterday, we're also watching the final act of a long-running saga in China's suffering property sector. In late August, Chinese courts suddenly found Hui Ka Yan, the founder of Evergrande, guilty of massive fraud and financial mismanagement, sentencing him to life in prison. Almost immediately after, a Chinese court accepted a bankruptcy petition to liquidate Evergrande — a move the company's creditors have been requesting forever.

Many are speculating that the close timing of these two events isn't a coincidence. With things like this in China, there's always an element of surprise because the system is so opaque that nobody truly knows what's going on behind the scenes. But in reality, everyone should have expected Hui to end up in serious trouble eventually.

Evergrande defaulted back in 2021 after years of growing skepticism in Western financial markets regarding the health of its finances. The real estate sector's troubles actually began before Covid, when the central government tightened borrowing rules to rein in companies that were building endlessly with cheap money in a market where everyone believed prices would always rise. When order comes to a bubble, it typically crashes.

Over the last five years, the government has slowly worked to stabilize the sector. Because the overriding concern in China is always potential social impact, Beijing never moves aggressively. Instead, the government took its time progressively transferring domestic assets to domestic creditors, local governments, and unfinished housing projects. Now, it appears the government feels the fallout has been contained enough to let the system work. This means forcing shareholders, certain creditors, and banks to absorb the remaining losses. Government-owned banks at various levels, of course, have less visible back-door channels to absorb these hits.

From a political and public relations standpoint, orchestrating the liquidation alongside Hui's sentencing makes perfect sense. It ensures that any remaining victims, such as those still waiting for their apartments, see that this isn't the government's fault. They can pin it all on the "bad guy." As detailed in previous reports on Evergrande's liquidation, this marks the first time Chinese courts have accepted a liquidation petition for such a major company. Will this trigger a wave of liquidations for other big companies like Country Garden (2007.HK) or Vanke (2202.HK; 000002.SZ)? We think there will have to be some more, but it's going to be a very slow process. The government's primary concerns remain employment and local tax revenues. Some companies will be cleaned up if the social impact is deemed nil, but others, we're afraid, will be kept on life support to maintain stability.

The recruitment app operator’s revenue has slowed steadily over the last three years as it solidifies its dominant position in China

Key Takeaways:

• Kanzhun’s revenue grew 14.1% in the second quarter, similar to last year’s growth rate but slower than its 24% growth in 2024
• The Boss Zhipin recruitment app operator is using AI and deeper penetration of China’s smaller markets to expand at home, while also looking for opportunities abroad

By Doug Young

What do you do to keep growing when you’re already the big “boss” in your home market? If you’re Kanzhun Ltd. (BZ.US; 2076.HK), more commonly known in China for its leading Boss Zhipin recruitment app, you start looking overseas for new opportunities.

That was a highlight in the company’s second-quarter earnings report, released on Tuesday, where it dangled the possibility of expanding its OfferToday international service beyond its current Hong Kong base to other developing markets like Brazil, Vietnam and Argentina.

The reality is that Kanzhun is huge in China, estimating it has served around 300 million of the country’s 500 million urban workers over the course of its 12-year history. The company also estimates it has served 22 million enterprise customers over that time, compared with around 40 million active businesses nationwide right now.

Kanzhun detailed many of the things it’s doing to keep growing its business in China, including getting more of users to pay for its services and moving beyond its stronghold in major cities into smaller markets. It’s also improving its efficiency through greater economies of scale and use of AI to match employers with employees.

But there’s no escaping the fact that Kanzhun is no longer the high-growth company it once was. Its revenue has continued to grow every year since its 2021 IPO, which is no easy feat in the current challenging economic environment. But that growth rate has slowed steadily from 32% in 2023 to 24% in 2024 and just 12.4% last year.

The growth rate slowed further to just 7.6% in the first quarter of 2026. So, against that backdrop, its latest growth rate of 14.1% in the second quarter looks like a nice rebound, as the figure climbed to 2.4 billion yuan ($357 million) in the three-month period from 2.1 billion yuan a year earlier, according to its latest report.

The roughly two dozen analysts who follow the company expect Kanzhun to stabilize at its current growth rate, with most forecasting around 12.5% growth this year and next. Most are also broadly positive on the company, with 21 of the 23 polled by Yahoo Finance rating Kanzhun a “buy” or “strong buy,” while the other two rate it a “hold.”

It’s not difficult to see why they are broadly positive, since China is probably the world’s largest employment market for traditional urban-based jobs, and Kanzhun is such a clear leader in that market. The company is the world’s second biggest publicly listed recruitment specialist, with a market cap of about $8 billion, behind only Japanese giant Recruit Holdings (6098.T), according to the CompaniesMarketCap website.

But in terms of valuation ratios, Kanzhun looks relatively weak compared to its peers, possibly due to the weak state of China’s economy and its slowing growth. Kanzhun currently trades at a price-to-earnings (P/E) ratio of just 12.4, a fraction of Recruit Holdings’ 50, and also behind the 15 for Switzerland’s Addeco (ADEN.SW) and 39 for U.S. giant Robert Half (RHI.US).

Long road abroad

Against that backdrop, we’ll take a deeper dive into Kanzhun’s latest results, starting with its international expansion that could put the company back on a stronger growth track. That expansion is coming from its OfferToday subsidiary, which launched in Hong Kong in 2024 and was a leading force in the city by the end of last year based on its mobile daily active user count, company officials said on their earnings call for the fourth quarter of 2025.

Hong Kong is a popular first stop for Mainland Chinese companies expanding abroad due to the city’s many cultural similarities, and thus they can expand relatively quickly there. But such expansion into less familiar markets will probably require more time, Kanzhun founder and Chairman Zhao Peng said on the latest earnings call. So, anyone expecting fast results from this global expansion might want to think again.

“The lessons we learned from OfferToday is that it took around two to three years for our new business like OfferToday to enter into a market,” said Zhao. “Then (it took) the next additional five years to grow to achieve $100 million to $115 million of revenue. We consider this kind of place or this kind of city worth investing.”

He described cities like Hong Kong as “a middle dish – not too fast, but not too slow.” But the bigger prize will come through expanding to entire countries with populations closer to 100 million, markets he described as “slow dishes.” He said development of such markets will take longer, perhaps 10 to 15 years, to achieve annual revenue of around $100 million.

There’s no evidence that the company has officially entered any markets outside Hong Kong just yet. But Zhao said markets the company is targeting include places like Vietnam, Argentina and Brazil, suggesting it’s laying the groundwork to enter those countries. While there’s no guarantee it will succeed, the company’s strong track record in China, which has many similar qualities to other developing markets, should give it a decent chance.

Back in China, most of the company’s latest business metrics told a similar story, namely of healthy numbers and slowing growth. Its total paying enterprise customers, which provide the vast majority of its revenue, rose 11% year-on-year to 7.2 million for the 12 months through June, while its average revenue per paying user (ARPPU) during the second quarter rose 7%. The company forecast more of the same in the third quarter, predicting its revenue would rise 11.4% to 15.6% during the period.

Its cost of revenue grew by just 1.6% in the second quarter, far slower than its revenue growth rate, helping the company to improve its gross and operating margins. But heavy marketing spending related to the World Cup weighed on its bottom line, as its adjusted net income rose just 9.4% to 1.03 billion yuan from 941 million yuan a year earlier.

The company has quite a lot of cash, and gave some of that back to investors through its declaration of a $230 million dividend, in addition to $300 million it has spent to buy back its shares so far this year.

Investors were quite positive on the report, bidding up Kanzhun’s U.S.-listed shares by 22% in the two trading days after the announcement. But even after that rally, the stock is still down 12% this year, and its Thursday close of $18 is still slightly below the $19 it sold shares for in its 2021 IPO. The stock could enjoy some potential upside in the near-term if investors decide it’s undervalued, which looks quite possible. But any upside from a return to strong double-digit revenue growth looks to be at least a few years away.

To subscribe to Bamboo Works weekly free newsletter, click here

The producer of battery materials posted higher first-half earnings, helped by a rebound in nickel prices, as it expands into new technologies for energy storage

Key Takeaways:

• To keep pace with changing demand, CNGR has been investing in producing lithium iron phosphate batteries for premium EVs and renewable energy systems
• It announced plans to reallocate the remaining proceeds from its Hong Kong IPO from a South Korean nickel project into a Chinese phosphorus facility

By Lee Shih Ta

One of the hardest challenges for the new energy industry is predicting which technology will prevail in the next product cycle. Batteries are a case in point.

Various battery types using different compounds have emerged in recent years to power electric vehicles, electricity grids, solar storage systems and data centers. Automotive ternary batteries, with a cathode mix typically including nickel and cobalt, are now being supplanted by cheaper lithium iron phosphate batteries that can also be used as energy storage units. Meanwhile, next-generation alternatives such as solid-state and sodium-ion batteries are gaining momentum.

In a rapidly evolving market, going all in on a single technology would be a risky move, which is why CNGR Advanced Material Co. Ltd. (2579.HK; 300919.SZ) has decided to hedge its bets, and the strategy appears to be paying off.

Starting out as a leading supplier of materials for ternary batteries, the company has expanded to cover elements for nickel, cobalt, phosphorus, sodium and solid-state batteries. Its operations have also spread upstream to encompass nickel, lithium and phosphorus resources, as well as processing, smelting and recycling

With this diversified approach, the company has just delivered higher half-year profits, helped by rising nickel prices. Earnings released on Aug. 24 show CNGR revenues jumped 57.5% to 33.58 billion yuan ($4.71 billion), while profit rose 78.4% to 1.31 billion yuan. Overall gross profit margin also increased to 12.8% from 11.9%.

Revenue from new energy metal products came to 10.11 billion yuan, around 30% of the total, with the profit margin rising to 12.3% from 7.4% in the year-earlier period, a rise attributed to a recovery in nickel product prices. The gross profit margin of nickel-based materials edged down to 17.1% from 17.8% but gross profit per ton increased to 15,100 yuan from 13,600 yuan, mainly due to higher benchmark prices.

Supply of nickel ore, which is 60% controlled by Indonesia, tightened early in the year, driving prices up more than 30%, according to analysis by Goldman Sachs. The investment bank later raised its forecast for the average nickel price in 2026 by 16% to $17,200 per ton. Indonesia’s annual production quota is also projected to be lower than last year, leaving supply dynamics as a key price driver.

Despite the positive factors for nickel, CNGR’s investment focus is shifting towards the raw materials for lithium iron phosphate batteries, tracking changes in demand.

The company announced plans to redirect the remaining HK$1.20 billion ($153 million) in proceeds from its 2025 Hong Kong listing, which were originally earmarked for a South Korean nickel project, into its phosphate mining and processing facility in Kaiyang, Guizhou.

China’s battery market is in a state of technological flux. In the first half of this year, lithium iron phosphate batteries accounted for 81% of China’s installed power battery capacity, while ternary batteries achieved just 18.9%. More importantly, growth in battery demand is moving from EVs to energy storage systems, where phosphorous-based products can offer cost and performance advantages.

China’s production of power and energy storage batteries rose 53.3% year on year in the first half, outpacing the 12% increase in domestic power battery capacity, indicating that new output is going towards energy storage and exports, Fastmarkets has reported, citing data from the automotive battery industry.

Global battery giants are also adjusting their strategies. LG Energy Solution (373220.KS) is shifting some of its North American EV battery capacity to energy storage. By the end of this year, five of its eight North American plants will produce energy storage batteries, while its technology mix is also moving from more nickel-dependent chemistries toward lithium iron phosphate, which is better suited to static energy storage.

New business starts to pay off

For CNGR, phosphorus has turned from a long-term bet into an income stream. Revenue from phosphorus-based materials rose about 55% to 1.04 billion yuan in the first half, while gross profit margin swung to 7.8% from negative 10.5% a year earlier. The upturn was attributed to economies of scale from the gradual release of production capacity, coupled with a recovery in downstream demand.

But ternary materials could still have room for growth, being widely used in high-end EVs for energy density and extended driving range, particularly in Europe and the United States. In the first half of this year, China’s installed capacity of ternary batteries still rose 14.2%, while global output of ternary precursors increased 26.1%. Meanwhile, many solid-state batteries also use high concentrations of nickel in their cathodes. CNGR holds a 26% share of the Chinese market in ternary precursors, while shipments of solid-state battery precursors reached 100 tons in the first half.

CNGR has been diversifying its battery materials on multiple fronts to stay ahead of technology trends and gain market share. Capital expenditures fell 35.1% to 1.11 billion yuan in the first half, indicating that investment in property, plant and equipment had slowed after a period of rapid expansion.

Nickel prices remain an important factor in near-term earnings. Over the longer term, as the battery industry enters an era of multiple coexisting technologies, the challenge will be to deploy capital where it can generate the highest returns.

To subscribe to Bamboo Works weekly free newsletter, click here

China’s ‘first smart parking stock’ reported double-digit revenue and profit declines in the first half of 2026, blaming project delays

Key Takeaways:

• Keytop Parking reported its revenue fell 11.9% in the first half of this year, with its core smart parking systems and management services down by nearly 18%
• The smart parking lot builder and manager’s net profit fell by 12% during the six-month period, and was down 40% on an adjusted basis

By Edith Terry

Did it hit the skids, or was it merely slowing down for a turn in the road? That’s what some investors might have been asking last week after Keytop Parking Inc. (2272.HK) issued results that hardly looked encouraging in its first post-IPO financial report since its Hong Kong IPO two months ago.

The company’s profit sank 12.5% year-on-year to 25.2 million yuan ($3.75 million) in the first half of 2026, and was down by an even steeper 40% on an adjusted basis to 28.7 million yuan. Its revenue wasn’t much better, down by 11.9% to 333.9 million yuan.

Keytop’s main business segments were equally uninspiring. Revenue from its core smart parking systems business was down by 17.4% to 161.5 million yuan, while its smart parking management services segment was down by 11.8% to 80.9 million yuan. Its parking facility and platform operations unit was roughly flat at 90.8 million yuan.

The uninspiring results weren’t completely unexpected, since Keytop warned late last month that its profit fell during the period, blaming project delays as it increased the size of its sales and marketing team to manage its new parking space rental platforms. Keytop doesn’t own or build parking garages, but instead helps third parties outfit their garages and manages existing parking spaces.

The declines in revenue and profit put the brakes on three years of growth, which saw Keytop’s revenue rise from 738 million yuan in 2023 to 830.6 million yuan in 2025, and its profit grow from 87 million yuan to profit of 107.9 million yuan over that time. Keytop described the recent reversal for its smart parking system and smart parking management services as “minor fluctuations,” adding that it believes both businesses remain well-positioned for long-term development.

Keytop’s shares fell about 10% in the week after its original profit warning in late July. Investors had a change of heart later, sending the shares up 4.4% on the first trading day after the release of the midyear results last week. By the end of Wednesday this week, the stock was roughly where it was before the original profit warning in July.

Tied to car ownership

So, what’s going on? Since its founding in 2006, Keytop’s growth has been tied both to car ownership and China’s property sector, which both boomed for the first two decades of the 21^st century, only to hit the skids in the 2020s. Parking failed to keep up with the new cars driving onto the road during the boom years, and there were 80% more cars on the road than parking spaces at the end of 2024 – specifically, 345.7 million cars to 190 million spaces.

That logic helped Keytop’s shares to more than triple in their trading debut on June 26. The stock has given back some of the gains since then, but at Wednesday’s close was still about 180% above its listing price.

On the positive side for investors, Keytop is unique as one of the only listed companies globally with a large-scale language model dedicated to parking lots, using data from its over 30,000 facilities to manage lots and spaces remotely. It doesn’t hurt that Tencent is a major backer and that Yu Minhong, founder of education giant New Oriental, was an early investor.

More importantly, under the hood, Keytop has been building the part of its business that doesn’t depend heavily on new property construction, specifically the business of providing parking facility management and operation services on behalf of other owners.

Based in South China’s Fujian province, the company earns nearly 50% of its revenue by providing smart parking systems for parking lot owners, with another 24% coming from parking management services. Both segments depend on sales to third-party customers, with hardware sales making up over half of smart parking systems revenue in the first half of this year. This is the company’s original legacy business, which began with sales of parking space availability indicators and parking guidance systems.

A third, newer business, providing 27% of revenue in latest report period, is parking facility and platform operations, which includes Keytop’s AI apps business. Revenue from the segment grew by 80% between 2023 and 2025, with much of that coming through contract operations. These are parking facilities managed under contract for a fixed annual fee, with Keytop either sharing or retaining revenue based on negotiated terms. Many of these facilities are existing garages, which Keytop refits with its own technology and software.

AI apps

Keytop’s apps are built around an AI-driven automatic number plate recognition (ANPR) system and include an AI kiosk that replaces on-site personnel, an AI parking manager and AI customer service agent. The company has drawn on its Tencent connection to partner with WeChat’s payment app for cardless payments since 2017.

Keytop said it plans to increase its investment in its parking facility and platform operations business, which currently looks like its biggest growth engine. Among other things, it plans to develop customized services on top of existing application scenarios for public parking, and to introduce its parking facility and platform operation services to large commercial complexes, scenic areas and industrial parks. It also plans to expand its international footprint, with a focus on developing markets including Southeast Asia and the Middle East.

Keytop was founded by Sun Longxi, a graduate of Changchun University of Science and Technology in Northeast China, who moved to the Southern city of Xiamen with 300,000 yuan in capital to build a startup selling communications products. He started with ultrasonic testing equipment for a city parking guidance system, only to have his first client cancel the deal after just three months, leaving Sun with a warehouse full of inventory.

Then he had a stroke of luck when a Singaporean customer found him through his website. It turned out that only one other company in the world made a similar product, a Spanish producer that charged much more. Sun pivoted to the parking industry as car ownership was exploding but parking management was still primitive. He launched his first ultrasonic car-finding systems, and Keytop took off.

Sun has proven that he is able to rebound from an apparently bad situation, and China still sorely needs better management of its limited volume of parking facilities to make them more efficient. Now, Sun must just show investors that he can shift his company back into growth mode, proving Keytop’s weak first-half results were just a minor bump in the road.

To subscribe to Bamboo Works weekly free newsletter, click here

The automotive cockpit maker has agreed to buy a controlling stake in Guangzhou Seagull to expand into China's fast-growing smart home market

Key Takeaways:

• Pateo Connect will buy 20% of Guangzhou Seagull for 800 million yuan in a deal that will give it control over the manufacturer of smart home hardware
• The company is looking to leverage its core automative cockpit software to add smart home-based hardware to its portfolio

By Warren Yang

Observers might call it a major shifting of gears.That’s a key takeaway on the latest move by Pateo Connect Technology (Shanghai) Corp. (2889.HK), which last week announced it’s broadening beyond its core focus on high-tech car cockpits with a foray into kitchen sinks and bathroom plumbing. The shift looks like some slightly head-spinning corporate road rage until you peek under the dashboard.

According to its announcement last Friday, the automotive smart cockpit maker has signed a deal to buy 20% of Guangzhou Seagull Residential Industry Co. (002084.SZ) from its controlling shareholder for 800 million yuan ($119 million). Although Pateo is acquiring a minority stake, the transaction will hand it control over the sanitary ware manufacturer, giving it the right to nominate seven out of Seagull’s nine board members. Pateo is paying 2.6 times Seagull’s book value, which includes a premium for the operational control.

At first glance, the move looks like a distraction from Pateo’s main business. The company is a maker of smart cockpits for cars. Guangzhou Seagull, on the other hand, manufactures plumbing fixtures and prefabricated bathroom modules.

The apparent disconnect between the two business lines makes this deal unusual. Pateo seems to be betting that at the end of the day, some of the same technologies that power vehicle digitalization can also drive home automation. Essentially, it hopes to leverage its technological capabilities for automotive environments to better digitalize home equipment. For example, Pateo in theory can inject its key products like AI voice assistants and ambient interaction systems into Guangzhou Seagull’s kitchen and toilet hardware.

“The transaction will further extend the group’s capabilities in intelligent vehicles to smart home and other related terminal scenarios, establishing a strategic layout of ‘vehicle-home integration’ and promoting the long-term development of the group,” Pateo said in its filing.

China’s smart home sector is set for fast growth, driven by rapid deployment of 5G telecommunications networks, government initiatives and unified ecosystem integration by giants like Huawei, Xiaomi and Haier. The domestic market is expected to grow at a compound annual rate of about 24% from 2026 to 2033 to $117 billion in annual sales by the end of that period, according to Grand View Research. That’s significantly faster than Global Market Insights’ projection of about 9% annual growth for the global market over the nine years through 2035, positioning China as the central engine behind the rise of the AI of things (AIoT).

By comparison, China’s smart automotive cockpit market, valued at roughly $12 billion to $15 billion, is strictly bound by limited vehicle production capacity and severely pinched by an ongoing electric vehicle (EV) price war that is forcing part suppliers like Pateo to absorb relentless margin-eroding price cuts for its products.

Crucially, smart home hardware makers have a broad range of potential customers, from property developers to home improvement chains and individual consumers. By comparison, the customer base is far more limited for auto suppliers like Pateo, leaving them at the mercy of a handful of carmakers. So for Pateo, bridging into home automation is not just a high-concept move into homes from its vehicle base, but also an attempt to jump from a brutal, margin-challenged auto supply chain into a higher-margin, more promising domestic market that is already much larger in terms of sales potential.

Smart toilets

Guangzhou Seagull makes a range of high-tech plumbing devices like smart toilets and bidets, infrared-activated motion-sensor faucets for kitchens and bathrooms and modular prefabricated bath units integrated with lighting, ventilation, and water-control sensors.

But the company lacks in-house software development capabilities, and that’s where it could leverage the new relationship with Pateo. Guangzhou Seagull makes physical equipment, relying on third parties for the software that makes them “smart.” Now, it can turn to Pateo for the necessary software to perform that function.

In addition, Pateo can throw AI chips into the mix. Early this month, the company struck another deal to purchase 70% of Chengdu Mingyi Electronics Technology for up to 1.4 billion yuan, a transaction that the company says completes a “software, hardware and cloud” integration model.

Chengdu Mingyi is a fabless chipmaker specializing in high-performance analog, radio-frequency and high-speed optoelectronic communication chips. While those chips were designed for use in AI data centers and high-speed in-vehicle networks, they are transferable for a wide range of home applications, including voice recognition for toilets and heating, ventilation, and air conditioning systems.

All that said, Pateo is heading into a market directly tied to a Chinese property sector that is grappling with a prolonged slump – which shows up in Guangzhou Seagull’s sputtering business. Guangzhou Seagull’s annual revenue shrank 8% to 2.62 billion yuan last year, marking its second consecutive annual decline, as its business bore the brunt of declining home sales in China.

Guangzhou Seagull still generates positive operational cash flow, but accounting factors like depreciation costs and impairment charges for unsold inventory are keeping it in the red. Last year, it lost 118.1 million yuan, which means it won’t provide any immediate boost to Pateo’s bottom line.

Pateo isn’t profitable either because of heavy R&D spending and pricing pressure across the EV supply chain in China. Its net loss more than doubled to about 1 billion yuan last year, even though its revenue jumped 37% to 3.5 billion yuan. 

Pateo’s stock barely moved on Monday, the first trading day following the announcement of the Guangzhou Seagull deal. Things were better for Guangzhou Seagull’s shares, which jumped by their 10% daily limit in Shenzhen. Such a reaction to an acquisition plan isn’t so surprising, especially if the buyer should issue new shares to fund the deal like Pateo may do.

The 800 million yuan price tag for the Guangzhou Seagull acquisition amounts to more than half of the 1.4 billion yuan in cash and cash equivalents that Pateo held at the end of last year. So Pateo should resort to other resources to pay for the deal, saying it plans to use “self-raised funds” to cover part of the cost.

It could issue new shares, taking advantage of its relatively high price-to-sales (P/S) ratio of 5.9, which is far higher than the 1.8 for Huizhou Desay SV Automotive (002920.SZ) and 0.4 for Ecarx (ECX).

Pateo’s diversion to the smart home sector is strategically reasonable and could lift its valuation further still. But any investors hoping for quick results may be in for disappointment, especially as things stand in China’s anemic property market. 

To subscribe to Bamboo Works weekly free newsletter, click here

The silicon carbide substrate maker has built a leading position in a strategic semiconductor market, but now faces the harder task of turning the technology into a sustainable business

Key Takeaways

• SICC has returned to revenue growth after a difficult period, but a return to profitability remains uncertain as price pressure continues across the silicon carbide industry
• China has made significant progress in building a domestic SiC supply chain, but companies now face a tougher challenge of surviving an industry shakeout

By Hu Minghe

For much of the past few years, silicon carbide (SiC) was one of China’s hottest semiconductor stories. It represented two ambitions at once: supplying the next generation of electric vehicles (EV), while also helping China reduce its reliance on foreign suppliers for a critical semiconductor material.

For years, high-end SiC substrates were largely controlled by overseas suppliers, creating challenges for China’s semiconductor industry as it sought to build its own supply chain. To address that, Beijing began promoting the development of SiC substrate companies.

One of those was SICC Co. Ltd. (2631.HK; 688234.SH), better known in China as Tianyue Advanced, which has become one of China’s industry leaders. But its latest financial results, released last week, highlight a bigger challenge: China has built up a large industry, yet turning its capacity into profitable businesses remains difficult.

SICC reported revenue of 914 million yuan ($127 million) in the first half of 2026, up 15.1% year-on-year. But it swung to a net loss of 58.6 million yuan from a profit of 10.9 million yuan a year earlier. The results showed big sequential improvement, as stronger sales returned the company to the black in the second quarter after it lost 60.5 million yuan in the first.

Despite the improvement, investors remained cautious. SICC’s Hong Kong-listed shares fell about 5% last Thursday after the earnings announcement, and have lost nearly a quarter of their value over the last five trading days. That suggests concerns remain about profitability, as pricing pressure remains despite stabilizing sales.

The latest results mark a reversal from the slowdown that emerged before SICC’s Hong Kong listing last August, when oversupply in its two major end markets, EVs and solar equipment, began weighing on growth and pricing. The company is now selling more products again, but the industry environment remains challenging.

SICC’s advantage has been its relatively long history and ability to move up the technology ladder. Founded in 2010, the company started with smaller, 2-inch substrates before moving into higher-yield 4-inch, 6-inch and 8-inch products. It has also developed the most cutting-edge 12-inch SiC substrates, although those products have yet to enter mass production.

That progress has helped SICC become one of the leading players in the global SiC substrate market. According to a March report by Japan’s Fuji Keizai, SICC ranked first globally in conductive SiC substrates in 2025 with 27.6% of the market, while its share of the 8-inch segment reached 51.3%. That position gives SICC an advantage over smaller competitors, but hasn’t insulated it from industry cycles.

From breakthrough to industry shakeout

The rise of SiC was closely linked to the growth of EVs. Tesla’s early adoption of SiC technology helped accelerate interest in the material, encouraging automakers and semiconductor companies worldwide to explore its use in power systems. Chinese companies followed quickly, hoping to build domestic supply chains for a strategically important part of the EV industry.

That enthusiasm triggered a wave of investment in China, as manufacturers rushed to build new capacity. But demand didn’t grow quickly enough to absorb all the new supply, resulting in overcapacity that pushed prices lower and squeezed profits. SICC was typical of the group, reporting increased sales volume last year but declining overall revenue due to falling prices.

The industry shakeout has also affected global players like Wolfspeed (WOLF.US), once regarded as a leading SiC supplier, which filed for Chapter 11 bankruptcy protection last year, after years of heavy investment in new capacity. The company later emerged from restructuring in September after cutting billions of dollars of debt.

The same challenge applies to the newer Chinese companies. TianKe Heda Semiconductor, another major producer, is pursuing its third IPO attempt on the Shanghai Stock Exchange. Like SICC, TianKe Heda has strong technology capabilities but is operating in a market where falling prices have made profitability difficult. The company’s gross margin fell to negative 20.06% in 2025, showing it was spending far more to make each substrate than it could sell them for.

By comparison, SICC’s gross margin in the first half of this year stood at a far healthier 22.86%, showing its production costs were still well below the prices it was charging for its substrates.

TianKe Heda’s IPO journey highlights a broader challenge facing China’s SiC industry: technological progress has moved faster than the industry’s ability to generate sustainable profits.

Market expectations have also become more demanding. SICC’s Shanghai-listed shares have experienced sharp swings, falling to 61.79 yuan last September before rising to a record 188.88 yuan in June this year, only to later retreat to around 110 yuan in late August. The volatility reflects a broader shift in investor attitudes towards the growing number of semiconductor companies available to them. They are increasingly looking beyond technology breakthroughs and expectation of strong state support, and asking whether companies can turn those advances into sustainable earnings.

Search for the next growth engine

To better utilize its capacity, the industry is now looking beyond EVs for new sources of demand. AI infrastructure has emerged as one potential opportunity. As AI data centers become more power-intensive, the industry is looking at SiC as a possible solution for improving electricity conversion efficiency inside increasingly demanding computing infrastructure.

SiC could play a role in future applications such as high-voltage direct current systems and other advanced power conversion technologies.

Major semiconductor companies are already positioning themselves for that trend. Infineon (IFX.DE), one of the world’s largest power semiconductor manufacturers, has highlighted AI infrastructure as a growth area for its power business, while other industry players are exploring SiC applications beyond EVs.

However, AI is unlikely to immediately absorb the excess capacity created during the EV boom. Electric vehicles remain the largest source of SiC demand, and higher-voltage EV platforms will continue to be the most important driver for the market in the near term.

The next stage of competition will not simply be about who can produce the most wafers. It will be about who can deliver reliable products, win customer certifications and control costs. SICC has already shown that Chinese companies can compete in a semiconductor market once dominated by overseas suppliers. The next test will be whether China’s recently minted field of SiC substrate makers can survive the boom-bust cycles the semiconductor industry is famous for to become sustainable businesses.

To subscribe to Bamboo Works weekly free newsletter, click here

The company’s revenue grew strongly and its loss narrowed sharply in the first half of the year, as it aggressive its develops AI cloud services

Key Takeaways:

• Kingsoft Cloud reported its revenue increased by 33.7% in the first half of the year, as its loss narrowed by 43%
• The cloud services company’s depreciation and amortization costs during the period nearly doubled

By Lau Chi Hang

When Kingsoft Cloud Holdings Ltd. (3896.HK; KC.US) released its 2025 annual results in March, the charismatic Lei Jun, who currently leads smartphone giant Xiaomi, announced his resignation as non-executive chairman, succeeded by vice chairman and CEO Zou Tao. Kingsoft Cloud seems to be doing well under its new leadership, judging from the solid results in its first financial report since that changing of the guard.

According to the report, released last week, the company’s revenue rose 33.7% year-over-year to 5.78 billion yuan ($858 million) in the first half of 2026. And while it’s still losing money, its loss for the period narrowed by a considerable 43.4% to 437 million yuan.

On a sequential basis, the company’s revenue rose by 13.7% in the second quarter compared to the first, while its loss narrowed by an even larger 72% over that time. That improvement dangles the tantalizing possibility that Kingsoft Cloud could achieve breakeven in the second half of the year.

Zou Tao deserves much of the credit for the improvement. A company veteran who joined Kingsoft Corp. (3888.HK) nearly three decades ago in 1998, Zou was on board as the company transitioned from office software to gaming, and most recently to AI cloud services. He became a director of the company’s separately listed Kingsoft Cloud in 2016 and assumed the role of acting CEO in 2022, before becoming the chairman this year.

Since taking over the CEO duties four years ago, Zou has moved aggressively to slash the company’s low-margin content delivery network (CDN) business. In its place, he has focused on AI cloud, constructing a suite of offerings including infrastructure as a service (IaaS), platform as a service (PaaS), and mobility as a service (MaaS).

As the transformation moved ahead, the company’s loss narrowed steadily from 2.66 billion yuan in 2022 to 936 million yuan last year, and fell further to slightly over 400 million yuan in the first half of this year. Intelligent computing cloud services have emerged as its new star performer during that time.

Revenue for the company’s AI business reached 926 million yuan in last year’s fourth quarter, nearly double the year-ago period. The AI business grew by another 90% in this year’s first quarter, rising to more than half of the company’s total public cloud revenue for the first time.

Revenue from public cloud services reached 4.35 billion yuan in the first half of this year, up 46.2% year-on-year, as the company credited strong demand for AI cloud services.

Investors sober up

Zou Tao has clearly placed the right bet, as AI-related industries have become hot among investors lately. Businesses engaged in large models, GPUs, memory chips, PCBs and optical transmission fiber have seen explosive growth, exciting investors. Such euphoria is nicely captured by Kingsoft Cloud’s former Chairman Lei Jun, who once famously said: “Even a pig can fly if it stands in the center of a whirlwind.”

Kingsoft Cloud was one company caught up in that whirlwind, which briefly lifted its Hong Kong stock roughly tenfold from around HK$1.10 two years ago to a high of HK$11.32 last year.

But investors have begun to have second thoughts about such high valuations, and are gradually sobering up from the buying binge. Kingsoft Cloud was caught up in the resulting AI-related stock correction dating back to June, and has seen its price subsequently pull back to the HK$6 level.

High investments and massive depreciation

While Kingsoft Cloud's AI cloud services are growing rapidly, there’s also a hidden darker side to its story. Most importantly, investments needed to operate AI cloud services are exceptionally high. Goldman Sachs previously estimated that Alibaba Cloud plans to invest 380 billion yuan over the next three years, and Tencent Cloud's capital expenditure for a single quarter in 2025 alone reached 17.9 billion yuan.

Alibaba and Tencent are both quite large and possess the financial resources to handle such expenses. But for companies like Kingsoft Cloud, whose market capitalization stands at less than HK$30 billion ($3.83 billion) and which holds less than $1 billion in cash, the slightest misstep could cost it dearly.

Kingsoft Cloud also disclosed in its midyear report that its capital expenditures and leased assets totaled 6.24 billion yuan in the first half of this year, up 26% from 4.95 billion yuan last year.

In addition to its big capital spending, the company also incurs massive depreciation and amortization costs for related facilities such as leased servers and networking equipment. That figure amounted to 930 million yuan in the first half of last year, and nearly doubled to 1.78 billion yuan this year.

While Kingsoft Cloud's revenue is rising steadily, particularly within its AI cloud business, the company has also come under fire for relying on its affiliated entities, Xiaomi and Kingsoft Corp. In the past three quarters, those two companies contributed roughly 30% of Kingsoft Cloud's total revenue. That means Kingsoft Cloud could take a hit if either of those two companies scales back its business.

Top five outsider

What’s more, competition within China’s AI cloud sector has become quite fierce. A report by Omdia ranked Alibaba Cloud as China’s largest AI cloud services provider last year with 38.1% of the market, while Volcano Engine ranked second with 20.4%. They were followed by Baidu Cloud, Tencent Cloud, and Tianyi Cloud. Kingsoft Cloud failed to make the top five, putting it squarely in the second tier of major players. Such status leaves the company at the risk of being marginalized, and presents greater difficulties in competing for new business. Its smaller size also means its costs are higher than the big players as a percentage of sales, which makes it harder to operate profitably.

Zou Tao once remarked that “With the arrival of the AI era, the ceiling of the cloud has not merely been elevated; rather, it has been completely torn off.” But for Kingsoft Cloud, the more fundamental problem is its relatively low market share. That means that even if the market ceiling has been blown wide open, how much of the sky can it actually claim?

To subscribe to Bamboo Works weekly free newsletter, click here

The optical transceiver maker’s revenue rose 182% in the first half of this year, sharply accelerating from 70% growth for all of 2025

Key Takeaways:

• Innolight’s revenue nearly tripled in the first half of the year, while its profit rose by an even stronger 243%
• The maker of optical transceivers said capital spending by China’s internet giants was notably lower than their global peers in the first quarter of this year

By Doug Young

The maiden financial report from Zhongji InnoLight Co. Ltd. (3308.HK; 300308.SZ), released on Friday, is notable for several things, most notably its strong triple-digit growth on the company’s top and bottom lines. That growth isn’t a huge surprise, since InnoLight makes optical transceivers that are a key component in the infrastructure used for AI computing.

More notable is the dichotomy between AI spending in China and the rest of the world detailed in the company’s report. While the report shows frenzied spending in the West by companies like Google and Microsoft, spending is far more muted in China, despite reports about huge spending by local names like Alibaba and ByteDance. That dichotomy is working to InnoLight’s advantage, since the company gets the vast majority of its sales from outside China.

Finally, the report is also notable for the mushrooming size of InnoLight’s indebtedness as it rushes to expand to meet booming demand, which could be a red flag for investors if the current AI spending frenzy slows.

Investors didn’t seem to like the report, with InnoLight’s Hong Kong-listed shares falling more than 10% on Monday morning, the first trading day after the announcement. The company raised a sizable HK$53 billion ($6.76 billion) in its IPO in July – cash that it will need as it rapidly scales up to meet booming demand. The company’s optical transceivers are a key component for data transmission, serving as the hardware that converts electrical signals to light signals that can then be sent over fiber optic.

The Hong Kong stock has notched some modest gains since the IPO, though nothing like some other AI and robotics plays that we’ve seen this year. After the early decline on Monday morning, the shares were trading at HK$1,016, about 3.7% above their IPO price of HK$980. Even at that level they still command a relatively premium forward price-to-earnings (P/E) ratio of 39, ahead of a 28 for U.S. rival Coherent Corp. (COHR.US) and a 22 for Chinese peer Eoptolink Technology (300502.SZ), its two largest competitors.

InnoLight is the leader for optical interconnect solutions with 21.2% of the global market, according to third-party data in its IPO prospectus. In its inaugural earnings report, it describes 2022 as a pivotal year that kicked off a new phase of development for the cloud computing sector, thanks to exploding demand for large-scale computing clusters to support AI training and inference workloads.

That demand has accelerated sharply in the last two years, as reflected by InnoLight’s own explosive growth. The company’s revenue nearly tripled in the first half of the year, rising 182% to 41.8 billion yuan ($6.22 billion) from 14.8 billion yuan a year earlier, according to the latest report. That marked a sharp acceleration from 2025, when the company’s revenue rose 60% to 38.2 billion yuan, a figure that looked impressive at the time but pales compared with the latest growth rate. On a quarterly basis, the company’s year-on-year growth rate slowed to 175% in the second quarter from 192% in the first.

Uneven demand

That takes us to the next topic we raised earlier, namely, the lopsided situation between the Chinese and international markets. Much is written about AI spending in general, and China is typically characterized as spending just as heavily as the West on the technology. But the trio of Alibaba, Tencent and Baidu logged a combined 64.7 billion yuan in capital expenditure in the first quarter of this year, up just 18% year-on-year, according to InnoLight, citing previously published financial results. By comparison, Microsoft, Amazon, Meta and Google reported combined first-quarter capex of $164.9 billion, up about 86% year-on-year.

That big difference was also present in InnoLight’s own numbers. The company’s international revenue rose 210% in the first half of the year to 39.6 billion yuan. But its China revenue rose by a far smaller 7.7% during the period to 2.16 billion yuan. That difference boosted international revenue to 95% of the company’s total from 86% a year earlier.

In its IPO prospectus, InnoLight said the global market for optical interconnect products is expected to grow at an average rate of 31.6% annually between 2025 and 2030 to reach $111 billion by the end of that period. InnoLight’s growth rate is well above that, suggesting the first-half revenue growth rate may be an anomaly rather than a new norm. And if companies rein in their AI spending, which could easily happen, InnoLight could quickly find itself with too much capacity from its own recent buildup.

That leads us to the topic of InnoLight’s growing indebtedness as it aggressively ramps up its capacity. The company’s net cash from operating activities dropped to 1.79 billion yuan in the first half of this year from 3.22 billion yuan a year earlier, as it blamed increasing payments for raw materials. Its net cash used in investment activities also rose sharply to 6.8 billion yuan from 696 million yuan over that time. As a result, its net cash decreased by 4.2 billion yuan over the six months to June 30, though that doesn’t include the HK$53 billion that it raised from the IPO in July. As it spent heavily on expansion, the company’s gearing ratio also more than doubled to 17.1% at the end of June from 8.26% at the end of last year.

The company is clearly expanding very quickly, and should be able to use the money from the July IPO to fund that expansion. But if the global AI buildout loses momentum, InnoLight could easily find itself stuck with huge amounts of idle capacity.

For now, at least, things look relatively strong for the company. Its expenses rose roughly in line with its strong revenue growth. And its growing economies of scale allowed InnoLight to boost its gross margin by more than 7 percentage points to 45.8% in the latest period from 38.7% a year earlier. That helped lift the company’s first-half profit by 243% to 13.7 billion yuan from 4 billion yuan a year earlier, outpacing its revenue growth rate by a large margin.

On the whole, the report looks quite positive. The slower AI spending in China comes as a slight surprise, though InnoLight is insulated from that by its heavy reliance on the international market. The lone red flag is the company’s rapid expansion, which could come back to haunt it if global AI infrastructure spending begins to slow.

To subscribe to Bamboo Works weekly free newsletter, click here

The optical module maker is heavily reliant on the U.S. tech giant, which accounted for 99% of its revenue in the first half of the year

Key Takeaways:

• Cowell e Holdings reported its revenue rose 18% in the first half of the year, while its profit rose at nearly twice that rate, up 33.4%
• The optical module maker is trying to diversify beyond its heavy reliance on Apple, but with little progress so far

By Cheng Shui Tong

If finding your way into Apple’s supply chain is the holy grail for electronic component makers, then Cowell e Holdings Inc. (1415.HK) has found a special place in that hallowed realm.

The optical module maker’s financial results for the six months to June, announced earlier this month, show its profit rose 33.4% year-over-year to $89.87 million during the period. Its revenue increased by 18% to $1.61 billion, driven by rising shipments of its high-end products as it works to improve its product mix.

But the most notable thing about the report is the share of Cowell’s sales going to its largest customer, widely known to be Apple, which rose from an already high 98.1% in the first half of 2025 to an even higher 99.3% in the latest period. That’s not really a surprise, since Apple has consistently accounted for more than 90% of Cowell's revenue since 2018.

Investors didn’t seem to mind the near-total reliance on Apple, perhaps even applauding it, as the company's shares rallied 8% the day after the release of the latest report.

Hooked on Apple

While Cowell’s reliance on Apple has grown, other major companies in the tech giant’s supply chain have moved in the opposite direction. A case in point is iPhone glass supplier Lens Technology (6613.HK) whose revenue from Apple fell from over 70% of its total in 2022 to 45% last year. Similarly, Apple's contribution to Luxshare’s (2475.HK; 002475.SZ) revenue dropped from 75% in 2023 to 57% in 2025.

Originally listed in South Korea, Cowell entered Apple’s supply chain more than a decade ago in 2009. It was later taken private by its controlling shareholder before pivoting to a Hong Kong listing in 2015. It remained a rather unremarkable industrial stock for years, with its shares languishing in penny territory.

But 2020 marked a watershed for the company. As many Japanese and South Korean peers curtailed or suspended production when the pandemic began, Cowell scooped up a massive influx of Apple orders, catapulting its earnings for the first half of that year by a staggering 33 times. In December, Luxvisions Innovation, an affiliate of Luxshare, acquired nearly 45% of Cowell to become its controlling shareholder, sending the stock soaring.

Luxshare has long been a fixture in Apple's supplier network. The Cowell acquisition generated substantial synergies, fueling explosive growth in Apple orders and propelling Cowell's stock from just over HK$1 in early 2020 to a peak of more than HK$40 last year.

Relying on a single client for years is undeniably a double-edged sword. With Apple's backing, massive order volumes are guaranteed, and Cowell's performance and share price reliably get a boost during each new product launch and upgrade cycle. The company's technology and interests are deeply intertwined with Apple's. Its sophisticated production lines are tailor-made to meet Apple's stringent standards, creating a formidable economic moat in the optical module industry that it won’t easily lose in the near term.

Downside potential

Then there are the negatives. Any sneeze from Apple in response to slowing product sales, antitrust fines or other headwinds could result in a major cold for Cowell. Moreover, Apple tends to sign up multiple suppliers for its individual components to mitigate risk. And when doing business with a tech juggernaut wielding such strong pricing power, suppliers often find their profit margins getting squeezed.

In the past, management has rarely discussed looking for new clients to reduce its reliance on Apple. But the recent AI boom has made cameras indispensable portals for receiving data input, making highly sophisticated optical modules important in a wide range of applications, including robotics and autonomous driving.

In the outlook section of its report, Cowell’s management noted that optical modules are not merely essential functional components for end products, but serve as a critical foundational capability enabling environmental perception and data collection for an array of smart devices. As a result, the company is closely watching industry trends in emerging forms of smart terminals for potential new opportunities.

A prime example is the company's recent push into light detection and ranging (LiDAR) products, which could represent an important diversification step. LiDAR has wide-ranging applications, spanning autonomous driving, meteorological observation and facial recognition. Cowell previously partnered with leading automotive LiDAR provider RoboSense to establish a joint venture, Luxsense, signaling a joint foray into the automotive LiDAR market. While that move has been seen as the company’s attempt to develop a second growth curve outside Apple, it has yet to yield any material revenue contribution so far.

Lowly valued

While Cowell's business is closely tied to Apple's fortunes, shares of the two companies have hardly moved in lockstep. Apple's stock has surged past the $300 mark to an all-time high in recent months. But since reaching a new peak above HK$40 last September, Cowell's shares have retreated by roughly half to hover near HK$20, only recently rebounding on the back of its strong midyear report.

In terms of its valuation compared with other Hong Kong-listed Apple suppliers, Cowell trades at a trailing price-to-earnings (P/E) ratio of just 11, far below Lens Technology's 30 and Luxshare’s 23. That discount likely reflects Cowell's smaller operational scale and its heavy reliance on a single customer.

Nevertheless, the brokerage community is broadly positive on the company after the latest earnings release. JPMorgan pointed out that Cowell's recent share underperformance reflects investor concerns over weak demand and product pricing pressures. Emphasizing that its current valuation is 40% below historical averages, the investment bank expects the company’s strong earnings growth to provide some upside for the stock, maintaining a “buy” rating.

To subscribe to Bamboo Works weekly free newsletter, click here

The producer of GPUs for AI applications has flagged up a more than 18-fold leap in revenue and a sharply reduced loss, as it starts to capture economies of scale

Key Takeaways:

• The company projected its six-month revenue would reach at least $170 million, already exceeding its full-year turnover for 2025
• But as China steps up its bid for self-reliance in AI hardware, fast product cycles mean R&D spending is likely to stay high

By Lee Shih Ta

As China’s demand for AI computing keeps rising, chipmakers are facing a new test. Can they capitalize on the soaring demand to turn product lines into profits?

Until now, makers of the specialized AI chips have focused on developing the technology to allow China to close in on self-sufficiency in advanced computing. But increasingly they will be judged by their ability to boost their shipments of graphics processing units (GPUs) and staunch the flow of red ink.

One of China’s rising stars in the AI microchip sector, newly listed Shanghai Biren Technology Co. Ltd. (6082.HK), has just laid down a marker that it is speeding along the path towards profitability. 

The maker of general-purpose GPUs and AI computing hardware issued an earnings alert on Aug. 16 that predicted a leap in half-year revenues and forecast its net loss would shrink by at least two thirds.

Citing rapid growth in the AI industry, the company said it expects to post revenue in a range from 1.15 billion yuan to 1.30 billion yuan ($170 million to $193 million) for the six months to the end of June, from 58.9 million yuan a year earlier. That translates into a colossal jump of 1,852% to 2,107%, taking six-month revenue past the company’s full-year total for 2025. The net loss for the half year was projected to shrink to between 320 million yuan and 400 million yuan, 75% to 80% less than the 1.60 billion yuan deficit in the same period a year earlier.

China’s push for AI resilience could help the sector move beyond its seemingly endless phase of burning through cash, judging from recent earnings reports. Moore Threads (688795.SH), sometimes dubbed the Nvidia of China, narrowed its first-half net loss to 11.56 million yuan, while MetaX Integrated Circuits (688802.SH) cut its first-quarter loss to 98.84 million yuan. Tianshu Zhixin (9903.HK) has gone a step further, forecasting a profit for the first half, although mainly enabled by fair-value gains on financial assets.

Investors welcomed Biren’s improving fortunes. Its shares rose 6.97% to close at HK$38.98 after the profit alert, roughly double the HK$19.60 IPO price in January. Based on brokerage forecasts for 2026 revenue, Biren trades at a forward price-to-sales ratio of about 37 times, broadly in line with Tianshu Zhixin at around 35 times, but below the Moore Threads and MetaX multiples of 75 and 90. Those two firms enjoy a more pronounced scarcity premium as GPU makers with mainland listings. Biren is priced on a par with its Hong Kong-listed peer, Tianshu Zhixin, suggesting investors are already factoring in rapid revenue growth.

A closer look at Biren’s earnings record shows that deliveries of high-end products were concentrated in the second half of last year, when revenue reached about 976 million yuan. Compared with that figure, revenue in the first six months of 2026 still grew by roughly 18% to 33%, indicating that shipments are still accelerating.

As for the bottom line, Biren expects its adjusted loss for the first half to narrow to between 290 million yuan and 360 million yuan, down 35% to 47% from the year-earlier period. With some items falling out of the accounts after the listing, such as non-cash changes in redemption liabilities, the adjusted figure offers a clearer guide to operations. Rapid growth in revenue and gross profit, combined with a slower rise in expenses, pointed to emerging economies of scale, the company said. If revenue continues to outpace spending, profitability should come into view.

In its product mix, Biren has been moving beyond standalone GPUs toward complete computing systems. In 2025, it scaled up deliveries of its BR106 and BR166 products and deployed intelligent computing clusters, including a system with 2,048 GPUs. Large-model customers make their purchasing decisions based on cluster stability and cost, not just on chip performance.

That helps explain why Biren, founded in 2019 by a former president of SenseTime, has been devoting more resources to software and systems. Its software platform BIRENSUPA has expanded support for mainstream frameworks used by large language models, including PyTorch, vLLM and SGLang. The company also completed compatibility work last year for models including DeepSeek, Qwen and Zhipu GLM. One of the biggest hurdles facing Chinese GPUs is that many customers have built their workflows around Nvidia’s CUDA ecosystem. The easier the migration, the cheaper it will be for companies to switch.

High R&D costs to continue

Supplying systems could also mean bigger orders and help chipmakers reduce the risks of a concentrated client base. In its earnings forecast, Biren said it had continued to diversify its customers and deployed key applications at scale, a trend worth watching in future earnings releases. A broader mix of telecom operators, intelligent computing centers and enterprise customers could make revenue more sustainable, easing the reliance on a handful of large projects.

Gross margin is another key test. Biren’s revenue more than tripled last year while gross margin remained at 53.8%. That suggests the ramp-up has not overly relied on price cuts to win orders, leaving more room to absorb costs.

But R&D will remain a drain in the near term. The company spent nearly 1.48 billion yuan on R&D last year, more than its full-year revenue, and will need to keep investing in its next-generation BR20X products, software platforms, optical interconnects and supernode solutions. GPU product cycles move quickly, making it difficult to rein in spending once products start generating sales. Achieving breakeven will still depend on revenue outpacing R&D and operating expenses.

The half-year projection suggests Biren is moving in the right direction. But its eminent peers are also narrowing their losses. The ability to achieve mass shipments, cut customer migration costs and keep large clusters running reliably will be key to future earnings.

To subscribe to Bamboo Works weekly free newsletter, click here

The waste management specialist has received regulatory approval for its Hong Kong listing plan as it seeks growth outside its main market where it wields an effective monopoly

Key Takeaways:

• Junxin has received regulatory approval to list in Hong Kong, aiming to sell global investors on its trash-to-cash story
• The company is looking to export its business model beyond the Central Chinese city of Changsha, where it holds an effective monopoly in waste management

By Warren Yang

In the fast-moving world of Chinese corporates, waste management rarely commands the same glamour as high-flying technologies like AI, electric vehicles or semiconductors. Yet while many companies in those flashy sectors burn through cash, Hunan Junxin Environmental Protection Co. Ltd. (301109.SZ) is quietly doing just the opposite by converting heaps of refuse into cold, hard cash.

Now, the waste management specialist is looking to raise funds for overseas expansion from a Hong Kong IPO for its next growth phase. After securing approval from the China Securities Regulatory Commission (CSRC) for its offshore listing plan late last month, Junxin wasted no time in refiling its prospectus with the Hong Kong Stock Exchange last Thursday, after an earlier filing expired. Domestic heavyweights CICC and Citic Securities are acting as joint sponsors, indicating the listing could be relatively large, perhaps raising $100 million or more.

Overseas investors starved for profitable companies and generous cash dividends will welcome Junxin’s arrival in Hong Kong, which would complement its existing domestic listing across the border in Shenzhen. The company stably generates operational cash flow and pays out more than two-thirds of its net profit as dividends.

To understand the company’s investment thesis, one first needs to appreciate the massive structural shift in China’s environmental policy over the past decade. The era of dumping municipal garbage into unlined landfills or illicit discharge-and-forget waterways is long gone. Under Beijing's aggressive "Zero-Waste City" initiative and strict environmental mandates, waste disposal has transformed from a low-tech sanitation service into a capital-intensive, high-tech utility business.

Junxin sits squarely at the center of this transformation. Founded in 2011 by entrepreneur Dai Daoguo, a former military student who later became interested in environmental protection in the construction industry, the company operates across the entire waste treatment value chain. Its core cash generator is municipal solid waste (MSW) clean incineration power generation, better known as waste-to-energy (WTE). In addition to burning trash to produce electricity for the national grid, Junxin processes sludge, leachate, kitchen waste and fly ash.

Junxin’s revenue comes mostly from sales to the state grid of electricity produced by burning waste and fees — paid per ton for the collecting, compressing and disposing of trash. Junxin further monetizes waste byproducts through resource recovery, extracting and refining food waste into industrial-grade mixed oil for commercial sale to biofuel markets.

Green energy sales account for more than half of the company’s total revenue. Junxin ranked first in average grid-connected electricity generated per ton of waste last year, according to third-party data in its prospectus.

All of this happens at the Changsha Environmental Park in Central China’s Hunan province, one of the largest integrated eco-industrial parks in China. By housing WTE plants, leachate treatment facilities and food waste processing units within a single site using an ecosystem of related technologies, Junxin extracts economies of scale that few competitors can match.

With this facility, Junxin effectively holds a local monopoly over MSW treatment in the urban core of Changsha, Hunan’s capital with a population of more than 10 million. Backed by multi-decade government contracts, the company handles all the city’s municipal waste, sewage sludge, leachate, and fly ash generated across Changsha’s six main districts.

It further solidified its dominance by acquiring Hunan Renhe Environment in late 2024. The deal allowed Junxin to expand beyond its flagship downstream incineration hub and integrate Renhe’s midstream urban logistics. It brought Changsha's central municipal waste transfer and compression network, alongside a food waste treatment project, into Junxin’s ecosystem, uniting midstream waste collection with high-efficiency downstream power generation.

Superior margins

Because all urban waste in Changsha is directed to a single centralized hub owned and operated by Junxin, the company has an effective local monopoly that insulates it from the type of margin-eroding price wars and regional overcapacity that plague waste-to-energy operators in China’s coastal provinces. But it also limits growth opportunities, making Junxin look something like a traditional utility.

Thanks to the Renhe acquisition, Junxin’s revenue jumped about 31% to 2.41 billion yuan ($360 million) in 2024, and grew another 13% to 2.73 billion yuan last year. More importantly, the company’s closely integrated operations give it strong profitability. Jinxin turned a net profit of 716.6 million yuan last year, which translates into a net profit margin of 26%, an enviable figure for any company, let alone a utility operator.

But there’s only so much waste to process in Changsha, or in China for that matter. The country’s domestic waste-to-energy market has reached structural saturation, as rapid urban infrastructure expansion over the past decade has left coastal cities with excess incinerator capacity and intense price competition for trash. And because municipal waste management is locked under multi-decade agreements, it’s effectively impossible for Junxin to expand into other Chinese cities organically.

Facing a growth ceiling at home, the company is turning its sights overseas, looking to capitalize on China’s Belt and Road Initiative to export its high-margin business model to other developing markets. It’s eyeing the Central Asia nations of Kyrgyzstan and Kazakhstan, where underdeveloped waste processing systems and regional power shortages offer uncrowded markets and favorable long-term utility terms.

In Kyrgyzstan, the company invested $95 million to construct Central Asia's first major waste-to-energy plant in the capital of Bishkek. Following this entry, Junxin expanded its regional footprint with additional project agreements in the Kyrgyzstan cities of Osh and Karakol, while signing a preliminary framework exploration agreement in Kazakhstan. International expansion is one of the reasons the company is seeking the Hong Kong IPO.

Of course, overseas expansion is a completely different ball game from the comfortable domestic monopoly the company currently enjoys, as it entails navigating different regulation environments and local politics. So, its future growth is hardly guaranteed.

But Junxin can still appeal to investors looking for a rare cash-generating company with good, stable margins and steady dividends. Trading at a trailing price-to-earnings (P/E) ratio of about 14 for its Shenzhen-listed stock, Junxin commands a premium over state-owned giants like China Everbright Environment Group (0257.HK), which probably reflects its superior net margins and localized monopoly.

Junxin’s business may not be sexy, but it sure is stable. And stability can be a valuable asset in this age of volatility.

To subscribe to Bamboo Works weekly free newsletter, click here

"As much as the high-tech sector has become national interest in China, the same thing has happened in the U.S." – commenting on perils Apple could face as it explores using Chinese memory chips

Key Takeaways:

• Apple's reported plan to use Chinese memory chips highlights a complex balancing act between commercial needs and geopolitical pressures
• Burger King's turnaround after taking a major state-owned partner demonstrates how the right local alliance can revive a struggling Western brand

By Rene Vanguestaine and Doug Young

Multinationals operating in China are increasingly adopting highly tailored, localized strategies to survive and thrive. Whether navigating supply chain shortages in the tech sector or battling fierce competition in the fast-food arena, the playbook is changing. This is playing out now with two major Western names. Apple (AAPL.US) is exploring a strategy to buy memory chips from a leading Chinese producer, while Burger King is having a renaissance after forming a new alliance with a massive state-owned conglomerate.

We’ll start with Apple. According to The Wall Street Journal, the tech giant is reportedly looking into buying memory chips from CXMT (688825.SH), one of China's leading computer memory makers. For decades, the global memory sector toiled in anonymity, producing a commodity for PCs and smartphones dominated by the South Korean duo of Samsung (005930.KS) and Hynix (000660.KS), alongside U.S. giant Micron (MU.US). However, the sudden explosion of AI has created a massive global shortage, leading to spiking prices.

From a purely commercial standpoint, Apple's move makes sense as the company attempts to solve this supply shortage and potentially secure more favorable pricing. But this is where it gets complicated. Apple is acutely aware of the trade tensions between the U.S. and China. To mitigate this, the company has reportedly developed a regional isolation strategy: using CXMT chips exclusively in devices sold within China, while utilizing other suppliers for the rest of the world. We believe this represents a fascinating potential business template for other multinationals that might hesitate to use Chinese components globally.

Will this satisfy Washington and Beijing? The U.S. government has determined that CXMT works with the Chinese defense industry, which makes any partnership a deeply sensitive issue. Additionally, Washington has been actively trying to build a self-sufficient domestic chip industry. As much as the high-tech sector has become a national interest in China, the exact same thing has happened in the U.S. This shift started during the first Trump administration, continued under Biden, and accelerated during Trump's second term. The message from Washington is clear: don't help China build a growing business in the chip sector, use what's available in the U.S., and invest heavily alongside everybody else.

To soothe these political concerns, Apple has already announced it will help Intel (INTC.US) and Micron grow in the U.S. Ultimately, Washington's primary fear is that American companies might engage in technology transfer. As long as Apple simply uses existing CXMT chips, it may be viewed as the lesser evil. However, if U.S. companies ask these Chinese firms to get involved in custom designs, Washington will likely step in.

A state-owned recipe for fast food success

Shifting from tech to fast food, we're seeing another Western giant make a major tweak in its localized strategy. Burger King had been struggling in China in the face of better-run competition from McDonald's (MCD.US) and KFC (YUMC.US). But the brand appears to be turning a corner after its parent, Restaurant Brands International (QSR.US), partnered with Citic, a major state-owned conglomerate.

On a recent earnings call, Restaurant Brands executives noted that under this new partnership, Burger King China recorded another quarter of “double-digit comparable sales and a sequential improvement in unit economics.” In the current environment, double-digit comparable sales are incredibly strong.

What does a massive state-owned conglomerate like Citic bring to the table? We think it brings unparalleled "fire power." Beyond basic benefits like better sourcing of food ingredients and improved pricing, Citic provides tremendous leverage for securing building leases. Being state-owned also gives Citic an additional aura and the ability to get things done, particularly through better relationships with local governments. For a consumer brand, this looks like a win-win.

This state-backed partnership model contrasts sharply with other routes, such as aligning with private equity firms. We recently saw Starbucks (SBUX.US) choose a major local private equity name called Boyu to help navigate a market where it has fallen behind Luckin Coffee (LKNCY.US) in total outlets. There are cases where private equity firms have been very successful, and they can course-correct very quickly if something goes wrong. However, they historically present less certainty than state-owned giants. Because Starbucks is already an established brand, going with a PE firm represents less risk for them than it would for a smaller player. Sadly, smaller or mid-tier brands don't always have the luxury of choosing a giant like Citic. Brands like Tim Hortons and Dunkin Donuts often end up with smaller partners and ultimately struggle or close. These smaller partners simply aren't as efficient in operating and financing as the bigger PE firms, let alone state-owned conglomerates. It's a classic chicken-and-egg situation: major players like Citic want to partner with the biggest names, leaving smaller brands to take what they can get.

The company forecast a move to the black in the first half of the year, but investors are still waiting for its core inference GPU business to become profitable

Key Takeaways:

• Illuvatar expects to report a first-half profit of 60 million yuan to 140 million yuan, buoyed by more than 700 million yuan in paper fair value gains
• The GPU maker raised over HK$7 billion in a July share placement, with plans to use the funds to ramp up production

By Lee Shih Ta

If a chipmaker reports a half-year profit of 60 million yuan ($8.89 million) to 140 million yuan, but only because it logged 730 million yuan to 790 million yuan in investment gains, does it truly count as a breakthrough? A profit alert last week from Shanghai Iluvatar CoreX Semiconductor Co. Ltd. (9903.HK), a graphics processing unit (GPU) maker that debuted on the Hong Kong Stock Exchange early this year, raises exactly this question.

The company said it expects to swing to a profit of 60 million to 140 million yuan in the first half of this year, reversing a 609 million yuan loss a year earlier. But its board pointed out the breakthrough was due to 730 million yuan to 790 million yuan in fair value gains on financial assets stemming from shares it holds in a Shanghai-listed company. In other words, the swing to the black was just a paper profit.

While Iluvatar didn’t name the listed company, available information suggests it is SJ Semiconductor (688820.SH), a recently listed company that it invested in earlier this year. That means the pleasant swing to the black shows Illuvatar is a good investor, though it doesn’t really say much about its core GPU business.

Some simple math to exclude the fair value gains leaves the company with an estimated loss of 590 million yuan to 730 million yuan for the six-month period. That may not be the same as the adjusted net loss many companies provide, since Illuvatar’s final midyear report may still include other non-cash profit and loss items as well as accounting adjustments. Nevertheless, it at least demonstrates that the company’s core chip business is still far from being profitable.

That’s not to say the business isn’t growing rapidly. The company's 2025 revenue nearly doubled, surging 91.6% year over year to 1.03 billion yuan, as its gross margin improved from 49.1% to 54%. Yet, it still recorded a net loss of 1 billion yuan for the year. Excluding share-based compensation and listing expenses, its adjusted net loss stood at 438 million yuan. Heavy R&D spending kept any future profits in the distance, reaching 974 million yuan for the year, equal to 94.2% of its annual revenue and roughly 175% of its gross profit. That shows that despite the rapid revenue growth, hefty R&D investment, an important factor for all tech companies, could continue to delay the company’s eventual move into the black.

In the first half of 2025, revenue from Illuvatar's inference chip business spiked from 21.6 million yuan to 87 million yuan. But its gross margin dropped from 52.9% to 32%, which it blamed on intensifying market competition and price cuts on legacy products to clear inventory.

For all of 2025, revenue from its ZK series inference products reached 339 million yuan, more than tripling year-over-year, boosting its contribution from 18.6% to 32.8% of total revenue. Revenue from its general-purpose graphics processing units (GPGPU) totaled 923 million yuan, or nearly 90% of the total. The pressing question is no longer whether the company can sell its products, but rather when the rapid growth might translate into the economies of scale needed to operate profitably.

Some industry peers offer useful references. Cambricon’s (688256.SH) revenue skyrocketed 453.2% last year to 6.5 billion yuan, propelling it to a net profit of 2.06 billion yuan – its first annual profit since going public. During the same period, revenue for Moore Threads (688795.SH) and MetaX Integrated Circuits (688802.SH) grew by 243.4% and 121.3%, respectively, though that pair reported losses of 1.02 billion yuan and 781 million yuan.

China’s AI chip market has entered a new phase centered on “who can convert shipment volumes into profitability.” Compared to Cambricon, whose profitability was driven by its core business, Iluvatar's recent paper profit doesn’t really count.

Another point worth noting is cash burn. Iluvatar logged a net operating cash outflow of 1.16 billion yuan in 2025, nearly double the level of 2024. Its year-end inventory also more than doubled from 343 million yuan to 710 million yuan, showing how scaling the business requires more working capital support.

New funding round

That ongoing need for cash means a massive financing round completed by Iluvatar in July warrants special attention. The company placed 14.86 million new shares at HK$476 each, raising net proceeds of approximately HK$7.03 billion ($896 million). About 60% of that, or roughly HK$4.22 billion, is earmarked for procuring critical materials and components and for enhancing its supply chain resilience.

The company said that increasing demand for AI training and inference clusters, coupled with a tight supply of certain components like memory, requires it to purchase materials in advance to build up its inventory buffers.

In a positive signal, Reuters, citing unnamed sources, reported in June that Illuvatar was in talks to sell chips to internet giant ByteDance. If such a deal materializes, Iluvatar could deliver 50,000 or more chips to ByteDance this year, mostly for use with its Doubao large model. The report also indicated that domestic GPU and AI chipmakers captured nearly 41% of China's AI accelerator server market in 2025.

Those signals show Illuvatar's inference GPU revenue continues to grow rapidly, as potential major internet clients emerge, giving investors the confidence to provide it with a fresh HK$7 billion cash infusion. A continuation of triple-digit growth could indeed take Illuvatar to the next level. Still, whether it can further convert its top-line growth into higher gross margins, improved operating cash flow, and ultimately profitability for its core business, will be the company’s critical next test.

Illuvatar has yet to disclose core operating metrics for the first half of this year, such as revenue, gross margin, R&D expenses, and operating cash flow, though it’s set to release its midyear report on Aug. 28. While the more than 700 million yuan in unrealized investment gains pushed the company into the black ahead of schedule, that feat is likely to be short-lived. For a GPU maker like Illuvatar, the ultimate question remains when it will truly turn a profit.

To subscribe to Bamboo Works weekly free newsletter, click here

China's largest third-party data center operator reported a 838 million yuan second-quarter profit, as it doubled its 2026 sales target and raised its full-year guidance

Key Takeaways:

• GDS Holdings posted net income of 837.6 million yuan in the second quarter, with its net margin rising to 27.1%
• The data center operator’s new bookings hit 470 MW in the first half of 2026, exceeding all of 2025, prompting management to double its full-year sales target to 1 GW

By Teri Yu

GDS Holdings Ltd. (GDS.US; 9698.HK), the Shanghai-based operator of high-performance data centers, sustained its growth momentum in the second quarter and raised its full-year sales target, buoyed by robust AI-related demand and contract-signing activities. The company last Thursday reported a net profit of 837.6 million yuan ($123.5 million) for the quarter, reversing a net loss of 70.6 million yuan a year earlier, as demand for AI computing capacity continued to reshape its business.

As China’s leading independent data center operator, GDS and its peers are at the leading edge of the AI revolution as key infrastructure providers, supporting the huge volumes of computing demand required by AI.

The company’s revenue for the three months through June rose 6.5% year-on-year to 3.09 billion yuan, while its adjusted earnings before interest, taxes, depreciation and amortization (EBITDA) rose 2.5% to 1.41 billion yuan, translating to a healthy adjusted EBITDA margin of 45.5%. The company also recorded a 959.9 million yuan dilution gain from an external investment during the quarter.

Riding the AI wave

"We delivered solid financial and operational results in the second quarter of 2026, reflecting our continued commitment to disciplined execution," said William Huang, GDS' founder, chairman and CEO. "We are very excited about the opportunities in China ahead of us, driven mainly by AI demand. We are confident in our ability to capture these massive opportunities and expand our business at scale."

The company's core metrics all posted strong gains. Its total committed and pre-committed area rose 18.2% year-on-year to 784,802 square meters, while the utilization rate for area in service climbed to 79.2% from 77.5% a year earlier, reflecting strong sales momentum and accelerating commitment conversion. Its area under construction jumped 28.8% to 170,355 square meters, with a pre-commitment rate of 89.2%, meaning new capacity is being taken well before it comes online.

GDS’ Nasdaq-listed shares rose 6.2% to close at $34.77 on Thursday after the results came out. The stock currently trades at a price-to-earnings (P/E) ratio of 19. Analysts expect the company’s revenue growth to accelerate to 16.4% for the year, according to the average of 24 polled by Yahoo Finance.

Strong AI demand also supercharged the company’s order book. GDS signed 260 MW of new capacity commitments in the second quarter alone, bringing first-half bookings to a record 470 MW — already surpassing the total for all last year. That momentum led management to nearly double its full-year sales target to 1 GW, up from an original goal of 500 MW.

"Our sales momentum is the strongest we have ever seen," Huang told analysts on the company’s earnings call. He noted that GDS won significant new business from each of its three largest hyperscale customers in the first half of the year, while also beginning to build relationships with a new cohort of "emerging AI leaders." Roughly half of its first-half bookings came from newer markets such as Ulanqab and Horinger in the Inner Mongolia region, and Shaoguan in Guangdong province, illustrating how GDS is diversifying its geographic footprint to chase demand where power and land are available.

Beyond signed contracts, customers have also locked in an additional 600 MW of reserved future capacity this year, with management expecting total 2026 reservations to exceed 1 GW — a leading indicator the company says has historically converted into bookings at a 100% rate.

CFO Dan Newman told investors that the company’s backlog swelled to 757 MW by mid-year from 450 MW at the start of 2026, representing roughly 1.6 billion yuan of "booked but not billed" adjusted EBITDA set to flow through as new data centers are delivered and ramp up.

Building for tomorrow

To keep pace with demand, GDS raised its 2026 capital expenditure guidance to around 10 billion yuan from 9 billion yuan, saying it would fund the outlay through a disciplined mix of roughly 60% project-level debt and 40% equity. The company secured 4.9 billion yuan of new debt financing and refinancing during the quarter alone, which management said reflects a highly supportive onshore lending market.

Newman noted the company has deleveraged to 4.7 times net debt to annualized adjusted EBITDA, leaving room to fund expansion while building out its onshore asset-monetization program, including a second potential asset injection into its C-REIT vehicle currently under regulatory review.

Together with its improved financial position and access to multiple funding channels, the new financing gives GDS greater flexibility to accelerate investment in China and convert its record order backlog into operating capacity.

Looking further out, management expects net move-in volumes — the pace at which built capacity is occupied and begins generating revenue — to more than double in 2027, with growth weighted toward the second half of that year, setting up what Newman described as a "significant acceleration of EBITDA growth" heading into 2028.

For the full year, GDS lifted its revenue guidance to between 12.7 billion yuan and 13 billion yuan, up from the 12.4 billion to 12.9 billion yuan it previously targeted, implying year-on-year growth of 11.1% to 13.7%. It similarly raised its adjusted EBITDA guidance to 5.9 billion yuan to 6.1 billion yuan. On a pro forma basis stripping out one-time items, first-half adjusted EBITDA already grew 12.7% year-on-year.

"With strengthened financial standing and funding capabilities to support our business expansion, we remain focused on creating sustainable, long-term value for our business partners and shareholders," Newman said.

With bookings running at record levels, a fast-growing backlog and China's AI infrastructure race still in its early stages, GDS management appears increasingly convinced that the company's best growth chapter is still ahead.

The Bamboo Works offers a wide-ranging mix of coverage on U.S.- and Hong Kong-listed Chinese companies, including some sponsored content. For additional queries, including questions on individual articles, please contact us by clicking here

To subscribe to Bamboo Works free weekly newsletter, click here

The LiDAR technology specialist is jumping on the fast-moving humanoid robot bandwagon with its latest strategic growth initiative

Key Takeaways:

• Hesai has begun generating revenue from robotic actuation modules, while its Kosmo spatial intelligence platform is expected to contribute revenue next quarter
• The company reported a fifth consecutive quarter of profitability in the second quarter, and expects its strategic growth initiative business to break even in 2027 

By Doug Young

It’s never too early to look for the next big thing. That’s a key message in the latest earnings report from Hesai Group (HSAI.US; 2525.HK), which described this year’s second quarter as a “defining milestone” as it turned up its move into robotics. Co-founder and CEO Li Yifan, who also goes by David, said Hesai has evolved into “a full-stack infrastructure platform for robotics and Physical AI — empowering them to see, understand, and act.”

Leveraging its expertise in light detection and ranging (LiDAR) technology, the company has expanded into the fast-growing robotics sector, developing products for humanoid robots that also rely on such systems. Its Kosmo spatial intelligence platform and new robotic actuation modules are two of the company’s key strategic initiatives rapidly gaining commercial momentum in that space.

The company’s second quarter results, released on Tuesday, showed strong initial traction for those two new growth engines. Its revenue for the quarter reached 861 million yuan ($127 million), while its net profit rose 60% year-on-year to 71 million yuan, despite the need for continued investments to build up its new strategic initiatives. The latest quarterly profit also marked Hesai’s fifth consecutive quarter of profitability on a GAAP basis.

First Kosmo orders

Kosmo, one of Hesai’s strategic growth initiatives, is a spatial intelligence platform that combines an AI-powered spatial camera, AI algorithms, 3D spatial data and cloud services into a unified system. Hesai says the platform converts real-world environments into editable, interactive 3D models that can be used for applications in robotics, cultural tourism, film and television production, gaming and advertising.

Hesai delivered initial Kosmo prototypes in July and has received orders from several humanoid robotics companies, including Galbot, the company said. It added that it is in talks with more than 200 prospective partners across industries and expects Kosmo to generate initial revenue in the third quarter of 2026.

Actuation modules start generating revenue

In robotic actuation modules, Hesai said its proprietary modules combine precise control, compact size, high force output and true backdrivability. Compared with other leading products, it says, the modules deliver approximately three times the torque and power density in a package that is 37% smaller. The modules also began generating revenue in the second quarter.

Hesai said its dedicated production line for the actuation modules is fully operational and delivered more than 10,000 units by the end of the second quarter, with shipments expanding from dexterous hands to full-body applications. Shipments are expected to reach six-digit volumes in 2027.

Morgan Stanley projects that cumulative global adoption of humanoid robots will reach approximately 1 billion units by 2050, with the annual market potentially hitting a massive $7.5 trillion in revenue. China is fast becoming a critical piece of that equation, with the investment bank recently raising its forecast for China’s humanoid robot shipments to 50,000 this year, nearly double its previous forecast of 28,000.

The company was the first in its class to become profitable when it recorded its first annual profit of 436 million yuan last year, even as most of its peers continue to lose money. Reflecting that, the company’s stock currently trades at a price-to-sales (P/S) ratio of 6.32, ahead of the 4.87 for RoboSense (2498.HK) and a lowly 0.70 for Seyond (2665.HK), both of which are still losing money.

Strong LiDAR Growth

While the Kosmo and robotic actuation modules initiatives took center stage, Hesai’s latest report showed the company’s core LiDAR business also continued to post strong growth in the second quarter.

The company’s LiDAR shipments rose 78.5% to 628,275 during the second quarter, outpacing its revenue growth rate, showing that the revenue share of relatively low-unit-price ADAS LiDAR is growing rapidly as it gains wider adoption. Within that total, ADAS-use LiDAR for autonomous cars still accounts for the majority, with shipments up 60% to 485,904 units for the quarter. Robotics-use LiDAR is rapidly closing the gap, however, with shipments up 193% to 142,371 units during the period.

On the auto side, the LiDAR business is being driven partly by shifting industry trends toward multi-LiDAR setups for Level 3 (L3) autonomous driving, which include safety redundancy requirements. As that happens, each L3 vehicle is expected to require three to six LiDAR devices, worth $500 to $1,000 per car for companies like Hesai.

The company says its products were featured in 56 models at the latest Beijing Auto Show in April, including leading brands like Geely, Cadillac, Audi, BYD and Chery, as well as electric vehicle specialists like Xiaomi, Li Auto and Leapmotor, and robotaxi operator Pony AI. Its LiDAR products are also used by over 50 embodied AI and robotics companies.

To keep its core LiDAR business at the head of the pack, Hesai has developed Picasso, a single-photon avalanche diode system-on-chip (SPAD-SoC), which it unveiled in April this year and described as the brains behind its next generation of LiDAR products. In its latest report, it said Picasso is now ready for mass production, and the LiDAR unit containing the technology is currently in the customer validation phase.

Analysts are generally bullish on the company, with investment banks like Citigroup, Goldman Sachs and Morgan Stanley all rating Hesai a "buy." Investors are also quite positive, with big names like Schroder Investment Management, Vanguard Capital Management, Fidelity Management & Research and BlackRock all holding stakes of more than 1%.

The Bamboo Works offers a wide-ranging mix of coverage on U.S.- and Hong Kong-listed Chinese companies, including some sponsored content. For additional queries, including questions on individual articles, please contact us by clicking here

To subscribe to Bamboo Works free weekly newsletter, click here

The robotaxi leader’s latest financial report shows its losses remained stubbornly high in the first half of this year, despite a substantial revenue increase

Key Takeaways:

• WeRide reported its revenue surged by 73% to 346 million yuan in the first half of 2026, while it lost 790 million yuan
• The company’s fleet of autonomous robotaxis reached 1,800 vehicles by the end of June

By Lau Chi Hang

Investors frequently ask when autonomous driving will finally succeed, to which some reply that day will come when the driving profession disappears.

The statement may be somewhat exaggerated, but it does reflect the long road autonomous driving faces before it can become truly commercialized. For the enterprises waiting for that day, profits are largely a fleeting dream, and containing losses and simply keeping the lights on is often more realistic. Several notable names, like Tsingtech Microvision, once valued as high as 10 billion yuan; Zongmu Technology, backed by Lenovo and Xiaomi; and former autonomous truck highflyer TuSimple, have all reached the end of their roads or are close.

Even a powerhouse like robotaxi operator WeRide Inc. (0800.HK; WRD.US) faces an uphill road to profitability, which is reflected in its latest earnings report delivered last week. Its revenue rose by a healthy 73.3% to 346 million yuan ($51.31 million) in the first half of the year, including a stellar second-quarter reading of 232 million yuan, up 82.2% year-on-year and doubling from the previous quarter.

Continuing losses

The company has been steadily building up a robotaxi fleet that it hopes will one day become its core business. At the end of July, its global fleet exceeded 1,800 vehicles, up nearly 40% from the 1,300 vehicles it had at the end of April. Average daily rides per vehicle during the second quarter exceeded 21, up 24% quarter-over-quarter, while its registered user base grew by 35% sequentially.

WeRide’s overseas autonomous driving business is now in 13 countries, including a partnership with ride-hailing giant Uber (UBER.US) for European autonomous robotaxi services in Madrid and Zurich.

While all those milestones look encouraging, investors shouldn’t rejoice too early. Despite the surging revenue, WeRide has continued to burn through money, including a loss of 790 million yuan during the first half of the year, similar to the same period last year. The company lost 400 million yuan in the second quarter alone, narrowing by 1.4% year-on-year, while its quarter-over-quarter loss expanded by 3%.

WeRide has lost big sums in each of the last four years, including 1.3 billion yuan in 2022; 2 billion yuan in 2023; 2.5 billion yuan in 2024; and 1.65 billion yuan last year. Its latest loss translates to another 1.6 billion yuan down the drain on an annualized basis, showing its red ink remains stubbornly high despite its revenue gains.

In that context, it’s not surprising the company's Hong Kong-listed stock fell by over 6% the day after its latest earnings announcement, showing investors remain skittish about its longer-term prospects.

Fierce competition

While WeRide’s revenue growth looks strong, the nearly 350 million yuan in first-half revenue the company reported is hardly anything to write home about. The figure remains low on an absolute basis, as the big majority of its business comes from programs still in pilot phases. That will make it difficult for the company to achieve economies of scale needed to drive down costs through mass production.

And even though the company's first-half revenue rose by 73%, its expenses have been growing at a similar clip. Its marketing and advertising expenses, in particular, reached 51.9 million yuan in the first half of the year, up 87%, or even more than its revenue growth. That means the company is achieving its revenue growth in large part on massive advertising and marketing spending. R&D spending is also constantly required as the technologies evolve, though that figure grew by a more modest 24% in the first half to 798 million yuan.

Then there’s the competition, as WeRide attempts to outrace global rivals like Waymo to mainstream consumer acceptance. That race was nicely captured by WeRide Chairman Han Xu, also known as Tony, when he said during a recent interview that: “In every generation, new talents emerge, with each leading the field for merely three to five months.”

That illustrates just how brutal competition is in the autonomous driving industry, where today’s market leader could quickly become yesterday’s news. That means spending is essential to stay in the race, leaving profits somewhere in the distance for everyone.

Elusive L5 goal

Unmanned driving is divided into six levels, with Level 5 (L5) as the highest, defined as truly autonomous. But how long will it take to reach that goal? Han spoke candidly on that topic in his interview, saying, “It is entirely possible that L5 won’t be realized in the next 10 to 20 years.” While a mere flick of a finger in human history, such a timeframe can feel quite distant and uncomfortable for investors.

Of course, WeRide may not need to wait for L5 driving to turn a profit. But realistically speaking, how long will it take to reach that milestone?

Han previously said his goal is to have 1 million autonomous vehicles in operation, quoting an ancient proverb: “Without taking small steps, one cannot complete a journey of a thousand miles.” He explained that 1,000 vehicles is just the current starting point, whereas 1 million is a longer-term objective that must be achieved for success.

That looks quite ambitious, given WeRide's current fleet only consists of 1,800 vehicles. And while the company operates under an asset-light model, its constant need for new investment makes it look like profitability won’t be on the horizon in the next two or three years.

WeRide CFO and head of international Jennifer Li said the company is moving steadily toward self-sustainability, aiming to assure investors the company is beginning to see improvements in its cash flow. Despite that, Li wasn’t any more specific on how much further WeRide must travel before it breaks even, let alone becomes profitable.

To subscribe to Bamboo Works weekly free newsletter, click here

The Chinese producer of magnetic materials has dropped plans to invest in an Australian minerals firm after the target pivoted towards North American partners

Key Takeaways:

• The plan was meant to diversify JL MAG’s supply chain but changes in ownership of the mining assets diluted the deal benefits and amplified geopolitical risks
• China’s tighter controls on overseas investment and cross-border technology transfers are also complicating the quest for rare-earth resources

By Lee Shih Ta

As global competition for rare earths intensifies, China is finding it tougher to gain control of overseas mineral resources.

Although Chinese companies dominate the fields of processing and magnetic materials, they still need to diversify and strengthen their international supply chains. But they risk being shut out of deals as the United States and Australia step up efforts to bypass China in their supply chains. Meanwhile, China itself is tightening oversight of strategic resources and outbound investment, making overseas expansion ever more complicated.

Against this backdrop, JL MAG Rare-Earth Co. Ltd. (6680.HK; 300748.SZ) has axed a two-year plan to invest in Australian rare-earth company Hastings Technology Metals. The Chinese magnetic materials company issued a statement on Aug. 7 saying the conditions for the equity subscription had not been met but it did not expand on why the plan had unraveled. No payment had been made, and the terms had never formally taken effect, the company said.

On the same day, a Chinese mining company suspended a rare-earth project in Laos, citing a need to comply with China’s evolving policies on mineral resources.

In their different ways, JL MAG’s Australia decision and the move by Chifeng Gold (6693.HK) to halt its Mengkham project both represent a slowdown of overseas expansion in rare earths. At a time when China still encourages overseas mineral cooperation, why are companies beginning to pull back?

A changed landscape

For JL MAG, the nature of the investment target itself had changed. Two years ago, the deal with Hastings was intended to strengthen the Chinese company’s supply of raw materials for its high-performance magnetic products. At the time, Hastings owned 100% of the Yangibana rare-earth project in Western Australia, where neodymium and praseodymium accounted for about 37% of total rare-earth oxides.

Hastings later enlisted Australian mining investment company Wyloo to help finance the project. After the transaction was completed last year, Wyloo acquired a 60% interest in Yangibana and took over management control, leaving Hastings with just 40% of the project. If JL MAG had proceeded with its plan for a 9.8% stake in Hastings, its indirect exposure to Yangibana would have been diluted to roughly 3.9%. The structure of the core asset underlying the agreed price had changed significantly.

At the same time, Yangibana’s strategic direction had shifted. Wyloo, Hastings and Canada’s Ucore signed a framework agreement last year to explore supplying up to 37,000 metric tons of rare-earth concentrate annually and to evaluate midstream processing in the United States. Ucore positioned the partnership as part of a North American “ex-China” supply chain, linked to the U.S.-Australia cooperation framework for critical minerals.

An investment originally meant to strengthen JL MAG’s resilience risked becoming embedded in a Western supply chain designed to exclude China. Even putting political considerations aside, the changes at Hastings and Yangibana were enough to call the commercial value into question.

Chifeng Gold’s move, meanwhile, points towards a trend for tighter Chinese regulation of the rare-earth industry to keep control of key technologies. The company cited a “progressively improving” policy framework for its Laos suspension, without detailing specific rules. New regulations that took effect in July have tightened oversight of outbound investment and cross-border transfers of restricted technologies. Rare-earth mining, smelting and magnetic-material technologies had also previously been included under stricter export licensing rules. Although those requirements were later suspended, they still suggest a broader direction of regulation.

Tighter controls

The regulatory scope even covers investment, consulting and joint R&D arrangements, showing that the authorities are closely monitoring whether China’s expertise in mining, smelting and magnetic materials risks being transferred overseas along with such projects.

The Mengkham project is still in the trial-mining stage, with the permit currently being renewed. The project made a loss of about 54 million yuan ($8 million) for Chifeng Gold in 2025, meaning the suspension should barely affect overall results. Meanwhile, a joint venture between Chifeng Gold and Xiamen Tungsten became embroiled in an environmental controversy late last year, highlighting how Chinese companies seeking mineral resources in Southeast Asia must also contend with rules on licensing, the environment and corporate responsibility.

JL MAG is in a stronger position to walk away from Hastings than it was two years ago. China Northern Rare Earth and China Rare Earth together accounted for about 72% of the company’s total procurement last year, while JL MAG also recycled 3,681 metric tons of rare-earth raw materials. Long-term domestic supply deals and recycling capacity therefore underpin its raw-material security. Revenue rose 14% to 7.72 billion yuan last year, while net profit jumped 142% to 706 million yuan. Net profit for the first half of this year is projected to rise by 31% to 51%.

JL MAG’s Hong Kong-listed shares edged down 0.3% to HK$17.90 in the first session after the announcement, suggesting investors see little earnings impact from the cancelled deal. Still, the company’s current supply is concentrated in the two Chinese rare-earth groups, leaving a continued need to diversify going forward.

Australian resources are increasingly being integrated into Western supply chains, while China itself is tightening oversight of overseas investment and strategic resources. For JL MAG, finding a viable overseas project may be considerably harder than it was two years ago.

To subscribe to Bamboo Works free weekly newsletter, click here   

The car technology company has filed to list in Hong Kong, as it pivots from automotive communication boxes to automated accident emergency call systems

Key Takeaways:

• Yodosmart has renewed its Hong Kong listing application as a maker of vehicle communication, accident emergency call, and sensing and domain control products
• Accident emergency call systems sales surged more than fourfold in the first five months of 2026, overtaking vehicle communication as the company’s top revenue source

By Lee Shih Ta

As Chinese car exports accelerate, breaching the 7 million-unit threshold last year, automakers are finding themselves facing not only new competition in unfamiliar markets, but also speed bumps from local safety standards that must be cleared. Such requirements, which lie hidden in the shadows of the global expansion wave, are bringing some unexpected bounty to manufacturers in the supply-chain for safety-oriented systems.

One of those, Hangzhou Yodosmart Automotive Technology Co. Ltd., which renewed its Hong Kong listing application last week, is unearthing fresh growth from those compliance requirements. The company supplies in-vehicle communication, emergency call, and sensing and domain control products tailored to a wide range of vehicles. Lately, it has found a major new cash cow in emergency call, or eCall, systems that contact rescue centers and transmit location and vehicle data when accidents occur.

Yodosmart’s pedigree came from its founders’ background at research institutes when it launched a decade ago. Chairman Li Wei previously worked at the No. 52 Research Institute of China Electronics Technology Group Corp. and Cethik Group for over 20 years, while general manager Lu Chaohong previously headed Cethik's smart automotive operations.

Grounded in this academic and product development background, Yodosmart quickly worked its way into the supply chain of a top-10 domestic automaker just two years later. The company began volume shipments for its SUV-use vehicle communication products for a top-10 manufacturer in 2020, and by 2022, its deliveries exceeded the 500,000-unit mark.

Its core product, the Telematics Box, or T-Box, functions as the vehicle's “communication box.” It bridges cellular networks with the cloud, enabling vehicle condition monitoring, remote control and software updates. Moving from design wins to mass supply typically takes nine to 24 months, with subsequent production orders extending for two to three years. Consequently, the company's revenue remains tightly tethered to life cycles for individual vehicles produced by its clients.

Turbocharged growth in 2024

Multiple 4G vehicle communication projects completed their production ramp-up and advanced into large-scale delivery in 2024, turbocharging Yodosmart’s revenue that year to 398 million yuan ($59 million), nearly double the previous year. But that growth was fleeting, punctuated by model life-cycle dynamics.

By 2025, as automakers retooled their production schedules and model mixes, Yodosmart’s 4G product revenue tumbled from 332 million yuan in 2024 to just 240 million yuan last year. Newly commercialized 5G products picked up some of the slack, but failed to plug the shortfall. Still, the company managed to salvage overall revenue growth of 12.8% last year, as its newer eCall and sensing and domain control businesses took the baton. In effect, Yodosmart’s growth engine pivoted from 4G mass production toward safety compliance products fueled by demand from export-bound models.

The company began bundling emergency call and vehicle communication solutions into export models in 2023, securing multiple international certifications the following year. That effort has paid off in booming revenue for the eCall business, which surged more than fourfold year-over-year to 78.76 million yuan in the first five months of 2026, equal to 43.9% of total revenue.

That jump helped the eCall business eclipse Yodosmart’s older vehicle communication segment for the first time, resulting in overall 37.1% revenue growth to 179 million yuan for the five-month period. More crucially, the eCall business boasts a 30% gross margin — more than double the 12.6% for vehicle communication — signaling this newer revenue stream should also help to lift Yodosmart’s profitability.

Profitability lags business pivot

That said, the revenue transition has yet to fully filter down to the company’s bottom line. Yodosmart's net profit grew from 10.94 million yuan in 2023 to 40.15 million yuan in 2024, but then dipped 1.4% to 39.57 million yuan in 2025, primarily weighed down by listing fees, climbing personnel costs, and dwindling government subsidies. Its adjusted net profit still grew 14.5% year-on-year to 57.81 million yuan in 2025.

In the first five months of 2026, the company swung to a net profit of 4.38 million yuan from a 630,000 yuan loss a year earlier, while its adjusted profit more than doubled to 13.14 million yuan. These figures indicate the newer eCall ramp-up has begun to bolster the company’s core earnings, though a tiny 2.4% overall net profit margin remains notably low.

The nascent bottom-line improvements, while encouraging, don’t signify that Yodosmart’s transformation is fully complete. Despite the shift from 4G to eCall products, the company’s legacy and new businesses are all anchored by the same core client, which appears to be Chery, based on descriptions in the listing document. As of the end of May 2026, its largest customer accounted for 77.4% of its revenue, with its top five clients collectively sweeping up 96% of its sales. The company is bringing a new factory online in the city of Wuhu, which is Chery’s headquarters, in East China’s Anhui province. Yet the added capacity remains heavily tailored for the client.

While such close collaboration can strengthen customer stickiness, it also makes adjustments more difficult for models from other customers.

Thus, Yodosmart next major task will be not merely selling more eCall units, but validating its product template across a broader array of automakers. China's mandatory standard for in-vehicle emergency call systems, slated to take effect on July 1, 2027, could help to redirect demand from export models back to the domestic market. However, 5G vehicle communication and sensing and domain control still represent a minuscule fraction of Yodosmart’s overall revenue and have yet to prove they can become another new growth curve.

Its high degree of specialization and unique product mix make it difficult to identify direct Hong Kong-listed peers for Yodosmart. Shares of closely related Shenzhen-listed counterparts Flaircomm Microelectronics (301600.SZ) and Gosuncn Technology (300098.SZ) are down between 20% and 30% this year.

For investors, Yodosmart's attraction is its eCall business that has replaced 4G as a new growth engine, and China’s 2027 mandatory standard that could boost demand at home for those eCall products. Nonetheless, the company’s overwhelming reliance on a single customer, and its relatively low profitability are important caution signals. At this stage, Yodosmart looks more like a relatively mature automotive electronics supplier rather than deserving a valuation premium as a high-growth tech stock.

To subscribe to Bamboo Works weekly free newsletter, click here

Revenue nearly doubled this year for the leading provider of green shipping equipment and systems, as it renews its attempt at a Hong Kong IPO

Key Takeaways:

• Zhejiang Energy Marine has renewed its Hong Kong listing application, aiming for a first-to-market premium in the booming field of green shipping equipment
• The company’s revenue nearly doubled in the first five months of this year, as two of its newer businesses overtook its original exhaust gas emission control systems

By Doug Young

The U.S.-Iran conflict may be wreaking havoc on global shipping, but that doesn’t seem to be affecting Zhejiang Energy Marine Environmental Technology Co. Ltd., which bills itself as the global leader in green shipping equipment and systems. Instead, the company seems to be thriving on booming demand for cleaner-running ships that are the backbone of world trade.

Zhejiang Energy Marine is hoping investors buy into its high-growth story, as it submitted an updated application for a Hong Kong IPO last Friday after its original January application lapsed. While its story looks quite positive, driven by demand for cleaner-running ships, the company is also notable for the choppy nature of its business.

That’s not difficult to understand, since its work isn’t very standardized and is highly project-based, in a global shipping sector where the top 10 companies control around 85% of the market, according to its listing document. The largest of those shippers, the privately held Mediterranean Shipping Co., or MSC Group, appears to be Zhejiang Energy Marine’s biggest customer by far, which is a double-edged sword that we’ll discuss in more detail shortly.

Zhejiang Energy Marine operates in a space that’s growing rapidly, as shipping companies try to install the most cutting-edge technology to reduce their emissions. The global green shipping equipment and system market is expected to grow by an impressive 31.7% annually between 2025 and 2030 to reach 151.6 billion yuan ($22.5 billion) by the end of that period, according to third-party market data in the listing document.

Zhejiang Energy Marine was founded in 2018 as a maker of vessel exhaust gas emission control and cleaning systems. It was a product of China’s experimentation at that time with “mixed ownership reform,” which attempted to breathe new life into inefficient state-run entities by bringing in private investors. In this case the state-run entity that provided the company’s initial business was Zhejiang Energy Group, while the private-sector partner was Wang Xinru, who has a long career working in China’s state-dominated shipping industry.

Reflecting that public-private hybrid, Wang is currently the company’s president, while its Chairman Guo Jinrong has a long background in the government of Zhejiang province. Such hybrid management can be good when everyone agrees on strategy, and also provides strong channels to government entities that are key for things like funding and permits. But it can also become problematic when disagreements occur between the state-owned and private stakeholders. In such cases the state-owned parties – who are sometimes more interested in politics than profits – almost always win.

While internal politics remains a major risk factor going forward, the company is doing quite well right now, based on the data in its listing document.

Its revenue jumped 90% in the first five months of 2026 to 2.51 billion yuan from 1.32 billion yuan a year earlier, as two of its newer business areas gained major new momentum. The company started out mainly selling vessel exhaust gas emission control and cleaning systems, which accounted for nearly all of its revenue as recently as 2023. But two newer segments, vessel energy efficiency enhancement systems, and vessel retrofitting services, have been growing rapidly since then, and both passed the original vessel exhaust gas emission control and cleaning systems in terms of revenue this year.

Sputtering original business

The company’s legacy vessel exhaust gas emission control and cleaning systems grew 25.5% to 767 million yuan in the first five months of this year to account for 31% of total revenue. But that segment was falling steadily before that, with revenue down by a third between 2023 and 2025. The drop owed to sharply falling prices, which saw the average system tumble from a price of 15.3 million yuan in 2023 to just 5.18 million yuan this year. The company blamed the steep drop to its shift from doing more retrofitting work on existing ships to installing more systems on new ships, with the former typically costing much more than the latter.

But vessel energy efficiency enhancement systems did much better, with revenue up 144% year-on-year to 860 million yuan in the first five months of 2026, as that segment became the company’s biggest breadwinner at 34% of total revenue. Unlike the legacy business whose prices have been falling, average selling prices for vessel energy efficiency enhancement systems have been on a sharp uptick, rising from 1.79 million yuan in 2024 to 2.9 million yuan in the first five months of this year.

Vessel retrofitting services rose by a similar 145% to 800 million yuan, making up 32% of total revenue.

Gross margins for the three main business segments are quite variable, with retrofitting services notably low in the 10% to 18% range. That reflects the non-standardized nature of the business, with most sales involving high degrees of customization and different levels of support services.

But one of the most notable things, and biggest risk factors, for Zhejiang Energy Marine is its extremely heavy reliance on a single customer, which, while not named directly, appears to be MSC Group. That single customer supplied 68% of the company’s revenue in the first five months of this year – a huge total by any measure, and one that could quickly change if the relationship deteriorates or MSC finds better or cheaper products.

What’s more, the company’s second-largest customer is a small company with a fleet of just four vessels engaged in dry bulk shipping. That means the customer, which accounted for 13.5% of Zhejiang Energy Marine’s revenue in the first five months of this year, is likely to disappear as soon as its small fleet is upgraded and equipped with the latest green equipment.

That said, this company is quite solid in terms of its financial health. Its operating cash inflow has ramped up steadily in step with its business, rising to 756 million yuan in the first five months of this year from 663 million yuan for all of 2025. That’s helped the company to build a war chest of 1.5 billion yuan in cash by the end of this May, up from 794 million yuan at the end of 2025. And on the bottom line, the company also saw its profit more than double to 622 million yuan in the first five months of this year from 277 million yuan in the year-ago period.

Zhejiang Energy Marine would be the first from its class to go public if it completes its Hong Kong IPO, with major rivals like Feen Marine, Panasia Co. and Innomotics GmbH all still private. That first-to-market status, combined with its relatively solid financials, despite its heavy reliance on a single customer and fading original business, could prove enticing for investors looking for exposure to the global shipping industry.

To subscribe to Bamboo Works free weekly newsletter, click here

The former e-commerce company wants to become a high-tech infrastructure operator, adding new momentum to its current core chip distribution business

Key Takeaways:

• Ingdan has unveiled an “AI token factory” offering computing capacity for high-tech companies
• The chip distributor is already benefiting nicely from its expansion into AI chips, estimating its revenue potentially doubled in the first half of 2026

By Warren Yang

If there’s one thing Ingdan Inc. (0400.HK) excels at, it’s catching major technology waves quickly.

The company went public in Hong Kong back in 2014 when it was known as Cogobuy, which sold general electronic components to businesses. Following a 2019 restructuring to capitalize on the rapid rise of advanced technology, it shifted its focus to wholesale distribution of integrated circuits (ICs) and AI internet of things (AIoT) services, eventually renaming itself as Ingdan, meaning “hard egg,” in 2022 to reflect the new identity.

As generative AI captivates the world, Ingdan last month announced its newest journey –building a “vertical AI token factory.” When it unveiled the new business, the company said it had already secured potential orders exceeding $1 billion. That sounds rather promising, though how much of that will turn into actual sales remains to be seen.

Ingdan currently generates most of its revenue from its chip distribution and technology services like custom hardware design, which puts it in pole position to capitalize on the current AI boom.

Last Friday, the company said it expects to post revenue of 12 billion yuan ($1.67 billion) to 14 billion yuan for the first half of this year, up as much as 100% year-on-year. It estimated that its profit from operations grew 60% to 100% year-on-year during the six-month period from 275.6 million yuan a year earlier. Management attributed the explosive profit growth to strong demand for hardware used for cutting-edge applications like AI data centers and robotics.

Investors seem to appreciate the company’s pivot to its current focus centered on AI products and services. Ingdan shares have rallied 85% in the past year, easily outperforming the broader Hong Kong market.

Now, Ingdan is betting that as the AI revolution accelerates, the real recurring value is moving from AI hardware sales to processing power. Unlike what the name suggests, the company’s AI token factory has nothing to do with the cryptocurrency, blockchain, or digital coins that enchanted investors for most of last year. In AI, a token is simply the fundamental unit of data, such as a fragment of text, code, or visual data, that an AI model processes and generates.

With its token factory, Ingdan is looking to provide specialized, full-stack computing infrastructure for enterprise clients in fields like humanoid robotics, autonomous vehicles and vertical AI models. The company is essentially positioning itself as a high-tech utility, charging customers based on the volume of computing work performed.

Traditional chip distribution that is Ingdan’s current mainstay is essentially a low-margin wholesale middleman business, with distributors buying products in bulk and reselling them for modest, one-time markups. Reflecting this economic reality, Ingdan’s gross profit margin is quite low, at just about 7% last year.

By contrast, the AI token factory model transforms those same chips into proprietary infrastructure, with Ingdan retaining hardware ownership and selling computing output as an ongoing service. Instead of capturing a single slice of profit at the point of sale, Ingdan can collect high-margin, recurring service fees over the lifetime of the hardware.

And Ingdan can cross-sell its new AI token services to its established network of more than 10,000 corporate clients, converting hardware buyers into recurring users of its computing capacity. By enabling its existing clients to replace heavy upfront hardware expenditure with usage-based fees, Ingdan can scale its computing platform with virtually zero customer acquisition cost, at least initially.

Different skill set

Yet this new business pivot can be easier said than done, as operating computing infrastructure requires a completely different skill set than traditional hardware wholesaling. While chip distribution relies heavily on procurement logistics and trade financing, running a distributed computing platform requires power supply management and other resources that incur heavy costs and require different expertise.

Ingdan plans to deploy its token factory capacity using a hybrid model, combining its own proprietary data center clusters with third-party computing nodes scattered across domestic and international locations. But maintaining consistent uptime and high usage rates across such a fragmented network could quickly strain operating margins if capacity utilization drops even slightly below specific levels.

Furthermore, high-performance AI chips are very expensive, requiring immense upfront capital expenditure that can weigh on the company’s balance sheet. And that’s not the only financial implication. The value of AI chips depreciates rapidly as newer, faster products hit the market every 12 to 18 months. That means Ingdan will need to start booking depreciation costs in its income statements, which will eat into its bottom line.

The quality of Ingdan’s touted initial order book also warrants some skepticism. The company’s headline-grabbing figure of more than $1 billion sounds impressive, but what it has lined up are merely “intent service orders,” which are non-binding.

There’s no guarantee that the company can convert these expressions of interest into real agreements, especially because the high-tech sector is notoriously volatile, with high cash burn rates and uncertain commercial longevity. If the AI startups that Ingdan is targeting struggle to scale or secure funding, their usage-based token consumption could quickly evaporate.

And obviously, Ingdan is entering a competitive arena dominated by large cloud hyperscalers run by tech titans like Alibaba, Tencent and Baidu. To survive, it will need to carve out a niche as a specialized AI computing orchestrator, leveraging its expertise in robotics and IoT rather than trying to compete on generalized cloud computing power.

Ingdan shares currently trade at a price-to-earnings (P/E) ratio of 23, a lofty level for a hardware wholesaler. By comparison, Smart-Core Holdings (2166.HK), an electronic component distributor, is far lower at about 11. Much of that premium owes to Ingdan’s strong stock gains over the last year, and the current gap suggests investors are banking on the company’s ability to adapt. 

Ultimately, though, Ingdan’s AI token factory pivot is a bit of a high-risk, high-reward strategic gamble. Its successful building of a steady pipeline of orders while keeping capital expenditure under control could transform the company from a low-margin hardware trader to a high-margin tech infrastructure operator. But execution could just as easily falter because of rising costs or customer churn, which would leave investors with dubious shares of a company tripped up by a transformation that failed to flower.

To subscribe to Bamboo Works weekly free newsletter, click here

Energy storage stocks are undergoing a major correction, including a plunge of more than half for Deye’s Shanghai-listed shares from a peak in May

Key Takeaways:

• Deye Technology has applied for a second listing in Hong Kong, reporting its profit surged 75% in the first four months of this year
• The new energy inverter maker gets 80% of its revenue from overseas markets, subjecting it to risks from trade policies and foreign exchange rate fluctuations

By Cheng Shui Tong

Hong Kong’s IPO wave may be ebbing, but some notable names continue washing up in the steady string of new filings nonetheless. One of those, Ningbo Deye Technology Corp. (605117.SH), harkens from a corner of the new energy sector providing energy storage inverters that are a critical component in solar and wind power stations. But the timing may be slightly off, as this sector, which was once an investor darling, has lost much of its star power in the last few months.

That said, Deye boasts some relatively strong financials. Its profit has been growing steadily over the last three years, rising from 1.79 billion yuan ($265 million) in 2023 to 3.17 billion yuan last year, according to its preliminary prospectus filed late last month. A month ago, Deye, which is already listed in Shanghai, projected more strong profit gains this year, saying it expects to report a profit of 2.67 billion yuan to 2.73 billion yuan for the six-month period, up 75% to 79% year-on-year.

Deye’s gross margin is also relatively high, ranging between 38% and 39.2% over the last three years. It has maintained such high margins by adjusting its product mix, and also by expanding in sales overseas where margins tend to be higher than for China.

Energy storage inverter leader

Deye’s bread-and-butter is energy storage inverters that convert direct current generated by solar and wind farms into alternating current that meets power grid requirements. According to third-party market data in its listing document, the company ranked first in the global residential energy storage inverter market last year with 20.6% share. It also produces environmental management equipment, such as dehumidifiers, solar air conditioners and heat exchangers.

Among Deye's two main businesses, new energy boasts the higher gross margin, which stood at 39.9% in the first four months of this year. Energy storage inverters were the star of that segment with a gross margin of 51.1% during that time, while energy storage batteries and PV inverters clocked in at 31.1% and 30.2%, respectively. Environmental management appliances were the company’s laggard, with a gross margin of 25.4%.

Big business overseas

A key factor keeping Deye’s margins high has been its pursuit of overseas markets. Exports rose from 58% of its sales in 2023 to 79.7% last year, and climbed further to 87.6% in the first four months of this year. Within that figure, Europe was the largest destination, accounting for as much as 45% of sales as EU members raced to install more renewable power in response to challenges created by the Russia-Ukraine war. Deye's overseas gross margin reached 40% last year, far higher than the 28% for domestic sales.

While things generally look positive for Deye, there are also some red flags in its business. Leading those is its heavy reliance on overseas sales, which could place it in the crosshairs of international trade conflicts. Illustrating that, the U.S. recently announced a ban on Chinese power inverters on national security grounds, causing related new energy stocks to plunge. Deye wasn’t too affected by the decision, since North America accounts for only 3.3% of its business. Still, similar moves in an increasingly protectionist Europe and other regions are always possible.

Another important risk lies in the foreign exchange rates. Because the company relies so heavily on exports, a strengthening Chinese yuan this year has resulted in net foreign exchange losses of 172 million yuan in the first four months of 2026 alone. The company states that a 5% appreciation of the yuan against the U.S. dollar would result in 240 million yuan in foreign exchange losses this year through April. A similar appreciation against the euro would generate about 11.4 million yuan in foreign exchange losses over the same period.

China is the world's largest producer of energy storage products that play a key role by storing excess electricity generated by wind and solar farms for later use. The market has grown steadily in recent years, but competition has also become fierce. Reflecting that, the average price of Deye's energy storage inverters fell by about 20% from 8,567 yuan in 2023 to 6,763 yuan in 2025, only to rebound somewhat to 7,015 yuan this year. Its selling price for energy storage batteries also dropped from 6,930 yuan in 2023 to 4,933 yuan in 2025, but then bounced back to 5,325 yuan this year.

Roller coaster stock prices

Deye filed to list in Hong Kong back in January this year, and only refiled last month after that initial application lapsed. That timing may be critical, as investor sentiment towards new energy IPOs has cooled markedly over that time. A case in point is Guoxia Technology (2655.HK), an energy storage company whose shares more than tripled after their IPO last December, giving it a sky-high leading price-to-earnings (P/E) ratio of more than 200 times. But the stock has cratered more recently, and now trades below its IPO price. Sigenergy (6656.HK), a provider of energy storage systems, has followed a similar trajectory. Its shares initially more than doubled after its April IPO, only to give back all those gains and more to trade below their IPO price.

Deye's Shanghai-listed shares haven’t been spared in the bloodbath, also down by half from their peak in May. Shenzhen-listed peers Sungrow (300274.SZ) and Ginlong (300763.SZ) have recorded similar declines, showing that investors remain concerned about stiff competitive and the potential for protectionist measures to affect exports. More broadly, market sentiment has also shifted away from high-growth emerging industries back toward traditional sectors, pressuring new energy stocks across the board.

Deye's performance looks strong, but that hasn’t resonated lately with investors. Its Shanghai-listed stock currently trades at a trailing P/E ratio of about 33 times. While that still looks relatively solid, the figure would drop significantly on a forward basis if its profit continues to grow strongly this year. That could ultimately work to the company’s advantage with the Hong Kong listing, since a pricing at today’s more reasonable valuation levels could draw in not only sector bulls, but also some bargain hunters.

To subscribe to Bamboo Works weekly free newsletter, click here

The former Lenovo sibling is betting on AI and data platforms to escape its traditional systems-integration business

Key Takeaways:

• Digital China Holdings expects to report a first-half loss after its DCITS subsidiary made a 333.6 million yuan provision linked to a contract dispute
• The company has spent decades trying to move beyond hardware-heavy IT services, but its latest AI push has yet to deliver higher margins and stabler profits

By Hu Minghe

IT services stalwart Digital China Holdings Ltd. (0861.HK) was supposed to be leaving its old business model behind. Instead, a dispute over a traditional technology contract has reminded investors why the company’s two decades of trying to reinvent itself have yet to yield convincing results, including the latest to overhaul its business with AI.

The Hong Kong-listed technology services company warned last week that it expects to report a loss of 50 million yuan ($6.9 million) to 70 million yuan for the first half of 2026, reversing a profit of about 15 million yuan a year earlier. It blamed the loss mainly on its Digital China Information Service Group (DCITS)(000555.SZ) subsidiary, which is expected to report a loss of 240 million yuan to 390 million yuan.

The loss stems mainly from a contractual dispute involving a procurement agreement with Beijing Urban Construction Intelligent Control Technology. Digital China said a first-instance court ruled a DCITS subsidiary must pay 333.6 million yuan in connection with the disputed contract, though DCITS has appealed the decision.

Regardless of the final outcome, the case highlights a broader weakness in the traditional systems-integration model that has long been Digital China’s main business. Projects using this model often involve large upfront purchases, customized development and long payment cycles, leaving companies exposed when contracts run into problems.

Search for higher-value technology

Digital China was created in 2001 when parent Legend Holdings separated its technology businesses, with Lenovo (0992.HK) retaining its PC operation while Digital China inherited its distribution and enterprise technology businesses.

At the time, Digital China looked like a promising tech company riding China’s early corporate computer boom. Companies across the country were racing to build internal networks and upgrade their information systems, creating big demand for businesses that could connect hardware, software and services from different suppliers.

But even at its birth, the company faced a profitability challenge. In a 2001 interview at the time of its creation, founder and Chairman Guo Wei said the company’s long-term opportunity was not simply selling third-party technology products, but moving toward more standardized software and services with recurring revenue streams. The company wanted to break away from traditional systems integration and build higher-value businesses.

The ambition was clear, but the reality was more complicated. At the time, around 1.2 billion yuan of Digital China’s roughly 1.3 billion yuan in annual systems-integration revenue still came from hardware sales. That business, broken out as its distribution business in its reports, generated a gross margin of only 8%, compared with 13.16% for Lenovo’s PC business.

More than two decades later, that same challenge applies. Digital China has continued to expand its systems-integration business. But the growth has come with limited pricing power. These projects often require large purchases of third-party equipment and extensive customization, producing much lower margins than simpler and more standardized software-related services. Its software development and technical services business generated a gross margin of 15.9% last year, compared with just 7.9% for systems integration.

The reason is structural. Systems integrators often operate as project contractors rather than technology-product companies. They compete for large contracts, purchase equipment from other vendors and customize solutions – a model that’s difficult to scale.

That model was once essential when Chinese enterprises were building their first computer networks two decades ago. But as cloud computing, standardized enterprise software and software-as-a-service (SaaS) products have matured, many technology functions no longer require the same level of customized integration.

Customers still need integrators to manage complex systems in the current environment, but those services are increasingly viewed as commodities rather than differentiated technology. And those services can often be performed by smaller vendors that charge lower prices, rather than big dedicated companies like Digital China.

Digital China is not alone. PCI Technology (600728.SH), which helps companies upgrade their IT systems through software, hardware and integration solutions, faces a similar challenge. Its traditional enterprise digital-transformation business accounts for most of its revenue but also generates low gross margins.

Digital China’s ongoing dispute shows why the economics of systems integration can be unforgiving: even after years of completing projects and generating revenue, one large contract problem can have an outsized impact on earnings.

Digital China has also faced internal turbulence lately, including a highly publicized divorce between Guo Wei and former wife Guo Zhengli, who previously served as the company’s COO. The divorce attracted media attention because of the couple’s long involvement with the company and their historical shareholding ties.

AI transformation

Digital China’s latest answer to improving its prospects is AI, or at least that’s how the company is framing things. It describes its latest strategy as “Data x AI” and “AI for Process,” aiming to turn its experience working with enterprise customers into reusable platforms and products that can achieve the scale its project-based work could never find.

Its main example is Yanyun, a data-intelligence platform designed to combine information from different corporate systems and make that data available for business applications and AI tools. The company argues that the platform can reduce the need for repeated customization by creating standardized data assets that can be reused across customers. It has also developed AI agents such as Xiao Jin, which are designed to automate supply-chain and operational decisions, while products such as Kejie Cloud Warehouse target smaller businesses with more standardized cloud services.

The idea is straightforward: software platforms and AI agents should theoretically scale more efficiently than traditional consulting-style projects. If successful, Digital China could move from a business that earns money by repeatedly building customized systems for individual customers into one that sells more standardized technology platforms to many customers.

But the financial evidence of such a transformation remains limited. Digital China generated 21 billion yuan in revenue last year, yet its profit was a miniscule 31 million yuan. Its data-intelligence business generated billions of yuan in revenue but contributed only a modest amount of profit.

That means investors are still waiting for proof that the company’s AI strategy represents a genuine business-model change rather than simply another layer added on top of a project-based services business with low margins and limited scalability.

A quarter century after leaving Lenovo’s shadow, Digital China is once again trying to answer the same question Guo Wei faced when the company was created: how to turn systems integration expertise into sustainable profits. This time, the answer may depend not on installing more systems, but on whether AI can finally find a way to make those systems more standardized and perform easily repeatable work, thus squeezing better margins from the process.

To subscribe to Bamboo Works weekly free newsletter, click here

"It's always all about money. Anything AI requires a tremendous amount of investment." – on Kuaishou’s decision to spin off its Kling AI video unit

Key Takeaways:

• Kuaishou's Kling spinoff reflects a growing trend of tech giants seeking massive standalone valuations for their AI units to fund rapid development
• A recent light aircraft crash in Beijing has exposed the severe safety risks of China's heavily hyped low-altitude economy, likely triggering intense regulatory scrutiny

By Doug Young and Rene Vanguestaine

China's technology landscape contains a tale of two vastly different frontier sectors. On one hand, we're witnessing an astronomical rush into AI, underscored by a massive new strategic spinoff plan. On the other, an actual small plane crash has brought the heavily hyped low-altitude economy firmly back to earth. While both of these lie at the cutting edge of innovation, they're currently on completely different trajectories.

We’ll start with the high-flying AI sector. Kuaishou (1024.HK) is preparing a major strategic move to spin off Kling, its AI video unit. The short video operator announced it will bring in around 20 new investors to support the service. This group will pump around 20.5 billion yuan — or nearly $3 billion — into Kling in exchange for 17% of the company, valuing the AI unit at around $18 billion. Kuaishou will continue to hold a majority 68% stake, while the remainder will go into various incentive plans.

The rationale here is simple: It's all about money. Anything related to AI requires a tremendous amount of investment. While Kuaishou is already public and could theoretically do a follow-on offering to raise such funds, investors wouldn't necessarily be interested pumping more capital into the parent company. But a pure AI play? Given the current technological climate, that could appear to many investors as the most valuable investment of their lifetimes.

By raising money specifically for the AI-related business, Kuaishou can deliver a much higher valuation than it would get for the company as a whole. And as long as it retains a substantial 68% majority, it's a meaningful way to raise the capital needed to build that business as quickly as possible. In this sector, you want to move fast and keep potential competitors in the rear-view mirror. Without enough money, moving fast is impossible.

Interestingly, this spinoff includes a clause stating that private investors are entitled to a refund if Kling doesn't go public by 2031. While five years down the road might not seem long to some, in terms of AI, that's like an eternity. In the U.S., major players like OpenAI and Anthropic want to move to market with IPOs very quickly. It's too soon to tell exactly how attractive a Kling IPO will be, as the company could either become wildly successful or be entirely out of business by then.

We believe there are going to be more companies doing this. This is one of the first times we've seen this sort of spinoff for an AI company with a wealthy parent, mirroring discussions we've seen surrounding Baidu (BIDU.US; 9888.HK) spinning off and separately listing its AI chip unit. Even highly capitalized giants like Alibaba (BABA.US, 9988.HK) or Tencent (0700.HK) might eventually follow suit. They may not financially need to, but at some point, the market might offer such a high standalone valuation for their AI businesses that it becomes an offer they simply can't refuse.

Grounding the low-altitude economy

While AI soars, China's low-altitude economy is facing a major setback. The sector was already quite slow to lift off despite massive industry and government hype, and a recent incident in Beijing has only deepened those troubles.

On June 26, a light aircraft crashed into Beijing's tallest skyscraper, Citic Tower, known to locals as China Zun. The crash killed the pilot and injured 13 people on the ground. While Beijing hasn't said much publicly, we think central leaders are likely quite alarmed that this type of accident could happen in such highly restricted airspace. The pilot reportedly received his license a few years ago and was known to have mental issues.

This is undeniably bad news for the makers of electric vertical takeoff and landing aircraft (eVTOLs), that were already facing commercialization difficulties. EHang (EH.US), the only publicly traded Chinese company in this space so far, scored a huge milestone last year when it became the first to win a type certificate for its eVTOLs from China's aviation regulator. Its stock initially shot up, but has since fallen back to earth as people realize the skies won't be filled with flying taxis anytime soon.

The Chinese government has long pushed this industry, liking the high-tech aspect that enhances China's global reputation. But putting flying objects into crowded urban environments is a massive risk. We understand that this emerging business needs tremendous amounts of regulation. Historically, aviation everywhere has been extremely regulated, requiring regular physical and mental fitness tests for pilots.

There's a massive disconnect between the hype — delivering packages and conducting building inspections by drone, and offering air taxi services — and reality. On the ground in China, there are very few products actually in use. Like helicopters in the U.S. that occasionally suffer mechanical or human failures despite a century of development, eVTOLs rely on mechanics that can fail. Putting tens of thousands of these objects into the hands of people with varying levels of flying and maintenance skills could be a recipe for disaster. The technology does have valid applications, such as delivering goods faster to inaccessible countryside locations. But mass urban commercialization is currently an illusion best left for the movies. For now, we'll have to wait a little longer for the Jetsons to come to China.

The company’s Hong Kong listing bid faces multiple headwinds, including its own unstable financials and cooling enthusiasm on robotics stocks

Key Takeaways:

• Topstar has filed for a Hong Kong IPO, reporting steadily falling revenues over the past three years
• The company is actively expanding its industrial robotics business in its bid to jumpstart revenue growth

By Lau Chi Hang

Agile robots performing flips and martial arts on China’s “Spring Festival Gala” Lunar New Year program never fail to dazzle, even as this year’s extravaganza is mostly memory by now. Even so, entertainment value isn’t worth a whole lot without practical, real-world applications. From that perspective, industrial models are the primary commercializing force in the robot sector for now, while the performing variety remain mostly a curiosity.

Against that backdrop, Guangdong Topstar Technology Co. Ltd. (300607.SZ), looks like a solid bet, at least conceptually, as a maker of industrial robots for the plastic injection molding sector. The company is banking on its proven products to lure investors to its planned Hong Kong IPO, following its filing of a listing application last week.

Founded in 2007, Topstar started out focusing on plastic injection molding processes before pivoting toward Cartesian and multi-joint robots. Listed on Shenzhen's Nasdaq-style ChiNext board in 2017, the company has recently introduced its own vertical large model for industrial robots. Last year, it rolled out intelligent single- and dual-arm robots and unveiled China's first intelligent humanoid robot specifically for use in injection molding scenarios.

Tumbling revenues

Despite its recent move into industrial robotics, nearly half of Topstar’s revenue still comes from its older intelligent energy and environmental products. Combined with injection molding equipment and computer numerical control (CNC) machine tools, industrial robots and automation systems only account for slightly more than 20% of the company’s total revenue, showing it’s still highly dependent on older product lines.

Topstar’s revenues have been falling steadily over the last three years, dropping by more than a third from 4.55 billion yuan ($672 million) in 2023 to 2.87 billion yuan in 2024. They fell further to 2.51 billion yuan last year. That uninterrupted string of declines translates to a contraction of more than 40% over the last three years.

Its bottom line has also been less than stellar. The company logged a profit of 106 million yuan in 2023, but then swung to a 240 million yuan net loss in 2024, before returning to the black last year with a 73.14 million yuan profit. It remained in the black with a 42.85 million yuan profit in the first quarter of 2026, but its overall profitability is clearly volatile.

Topstar attributed the massive top-line slump to a deliberate downsizing of its intelligent energy and environmental businesses. Its listing document shows that segment shrank from contributing 59% of the company's revenue in 2023 to 36.5% last year, before plummeting to just 5.6% in the first quarter of 2026.

“Addition and subtraction” strategy

Topstar expects to conclude the ongoing scale-back of its intelligent energy and environmental business by the end of this year. While the segment’s revenue will end up a fraction of its former self, overall profitability is set to notably improve. The business has continually lost money in recent years, maintaining a double-digit gross loss margin that has dragged down the company’s overall margins. By slimming down the segment, Topstar is sacrificing revenue but positioning itself for more sustained profits.

While cutting back its loss-making operations, Topstar has been pooling its resources to aggressively scale up its industrial robotics division. That restructuring helped to lift the company’s overall gross margin from 17.6% in 2023 to 32.5% in the first quarter of this year, showing its “addition and subtraction” strategy is bearing fruit.

Still, it’s worth noting that despite its improving margins, revenue from Topstar’s industrial robotics and automation systems business also fell from 969 million yuan in 2023 to 685 million yuan last year. The company said it strategically decided to reject some new orders to focus on serving core clients in the consumer electronics sector, causing its automation system sales to drop from 2,500 units to 1,300 units over that period.

Things started looking up this year. Automation system sales hit 1,000 units in the first quarter of the year, up by a sharp 230% from the year-ago period, while revenue from the segment also surged by a smaller 81% to 320 million yuan. All this shows that the company’s revenue mix is still quite unstable, as it searches for a return to sustainable growth.

Cash burn and mounting inventories

On the financial front, Topstar’s operating cash flow was negative in the first quarter of 2026, as it recorded an outflow of 130 million yuan — up 119% from the same period last year. Its net cash burn across the entire company reached 295 million yuan for the quarter, in stark contrast to its 380 million yuan net inflow a year earlier.

Inventory turnover days have been on an equally troublesome trajectory, lengthening by 72% from 90 days in 2023 to 155 days last year, before further ballooning to 230 days in the first quarter of this year.

A key element to Topstar's revenue stabilization is the overseas market, whose contribution has been growing steadily. Over the past three years, foreign sales grew from 11% of overall revenue in 2023 to 26.6% in the first quarter of this year.

The overseas expansion owes mostly to Vietnam and Indonesia. But even here, a closer look reveals that underlying revenue generation from overseas sales is less inspiring than the rising ratio of export sales to its overall mix. Instead, that rise is mostly a byproduct of the company’s plunging revenues in its home China market.

Revenues from Vietnam climbed 22% year-over-year to 500 million yuan in 2024, only to sink 27.7% to 365 million yuan last year, before rocketing more than sevenfold year-over-year to 99.7 million yuan in the first quarter of 2026. Indonesia delivered 14.97 million yuan in revenue in the first quarter of 2026, up from zero in the same period last year. Meanwhile, revenues in other foreign regions contracted by nearly 10% to 28.18 million yuan in the first quarter of 2026. As the data plainly shows, the company's international revenue remains volatile and patchy, and total sums are still quite minor.

Judging from its performance and recent financials, Topstar is in the process of navigating a pivotal transition that’s likely to determine its future. It’s trying to strip out its money-losing businesses, while expanding other areas with better potential, resulting in top-line contraction and an unstable bottom line. Its pivot appears to be mostly on track for now. But the strategy is still in an early stage, and more time is needed to gage the outcome.

To subscribe to Bamboo Works weekly free newsletter, click here

The telecoms software company said its revenue fell up to 19.2% in the first half of 2026, implying contraction of up to nearly 30% in the second quarter

Key Takeaways:

• AsiaInfo’s revenue fell between 11.5% and 19.2% in the first half of 2026, as strong gains for its AI business failed to offset accelerating declines for its traditional telecoms business
• The company is betting on AI infrastructure partnerships and space communications to revive its fortunes, though those two areas account for just 15% of its sales

By Doug Young

As report cards go, it probably rated a C+. That’s our assessment of a new earnings preview from telecoms software provider AsiaInfo Technologies Ltd. (1675.HK), which shows the company’s revenue deteriorated sharply in the second quarter after appearing to stabilize last year and into the start of 2026.

That slippage comes as China’s three major telecoms carriers, which are AsiaInfo’s biggest customers, cut back their overall spending. Making matters worse, the carriers are making especially big cuts on traditional network spending, which is AsiaInfo’s main focus.

The company is racing to develop new products and services tailored for the emerging AI computing and satellite networking sectors to replace fading demand for its traditional products. But the transition has been far from smooth, as gains for its newer business are failing to offset the sharper contraction of the company’s main traditional telecoms business.

AsiaInfo said it expects to report revenue of 2.1 billion yuan ($310 million) to 2.3 billion yuan for the first half of this year, down 11.5% to 19.2% year-on-year. The company reported a milder 6.9% revenue decline in the first quarter, meaning the situation deteriorated in the three months to June. Some calculations based on the first-quarter numbers and the latest half-year forecasts show the company’s revenue tumbled by 15.6% to 28.6% in the second quarter.

The bottom-line trends were slightly better, which saved AsiaInfo from getting an even lower grade on our report card. The company said it expects to report a first-half net loss of 460 million yuan to 490 million yuan, more than double the 202 million yuan loss a year earlier. But it already reported a 308 million yuan loss in the first quarter, showing its loss narrowed sharply to about 167 million yuan in the second quarter on a sequential basis.

AsiaInfo cited a number of factors behind the weak forecasts, led by reduced spending by China’s major telecoms carriers. That group has been reining in their capex spending over the last few years, following a major buildup with the rollout of 5G networks in 2019. Spending by the country’s four largest players fell 10.3% last year to 315 billion yuan from 351 billion yuan in 2024. They are on track to spend about 290 billion yuan this year, which would represent another 7.9% decline.

The company also blamed the revenue decline on its own decision to scale back or discontinue some of its less profitable businesses, including “certain inefficient ICT projects for government and enterprise clients” as well as other highly labor-intensive businesses. Its bottom line also took a hit from layoffs, as well as heavy spending on its newer initiatives related to AI, space and low altitude economy products and services.

Investors gave a strong thumbs-down to AsiaInfo’s latest preliminary results, and have been relegating its stock to the back of the class for quite some time now. The shares fell 4.9% on Friday, the day after the forecast came out, and are down 63.6% over the last 52 weeks. The decline is somewhat ironic coming during the current AI boom, since telecoms stocks were some of the biggest beneficiaries during the last big boom for internet stocks during the dot-com bubble of the 1990s.

AI infrastructure shift

AsiaInfo is racing to cater not only to a new generation of AI and space telecoms infrastructure builders, but also to its traditional telecoms carrier base that is shifting its spending priorities in that direction. In their latest discussions, China’s four leading carriers have indicated they are reallocating about a third of their capex budgets for the remainder of 2026 away from traditional network spending and into newer areas like AI computing and cloud infrastructure.

At the same time, a newer generation of private companies like Alibaba, Tencent and Baidu are spending heavily to build up their AI capabilities, including major spending on infrastructure. Finally, even newer companies like DeepSeek and Moonshot AI are developing their own open-source AI models and also spending on supporting infrastructure.

To address that changing landscape, AsiaInfo launched its “AI First” strategy earlier this year, which breaks out two new product segments – smart digital business and smart connectivity products – to complement its traditional telecoms software. The smart digital business has been the most promising so far, involving cultivation of relationships with key major clients on data, models, applications and other operations. In its first quarter update in April, AsiaInfo revealed that partners under that segment include big names like Alibaba Cloud, Volcengine and Moonshot’s Kimi, while it also has ecosystem partnerships with Nvidia and ABB.

The company’s core telecoms system business was the clear laggard last year, dropping 8.9% to 4.78 billion yuan, though even at that level it still accounted for about three-quarters of total revenue. The smart digital business moved in the other direction, rising 34% last year to 807 million yuan, accounting for 12.8% of revenue. That contribution has almost certainly grown even higher this year, as the company disclosed that revenue from the smart digital business nearly doubled in the first quarter of this year, even as overall revenue fell 6.9% during the three-month period.

The smart connectivity segment, which includes space and low altitude economy products and services, has taken off more slowly. But it got a major lift earlier this year with the launch of AsiaInfo’s “Satellite-Terrestrial Intelligent Connectivity” product portfolio, which includes satellite gateway core networking equipment, spaceborne base stations and spaceborne core networking software and hardware.

In its first-quarter report, AsiaInfo said it is working with low-earth satellite companies like Gesi Aerospace and Spacecom. But the segment is still quite small, generating just 125 million yuan in revenue last year, about 2% of the total, and only up 6.2% year-on-year.

The company’s uphill road is reflected in its bottom line, including a 79% profit decline to 114 million yuan last year. The first half of the year tends to be a weak period for the company, so it’s quite possible AsiaInfo could still report a profit this year despite losing money in the first six months. The company is also largely ignored by the analyst community, with zero coverage among the large pool of market watchers surveyed by Yahoo Finance.

While a C+ isn’t great, it’s certainly still a passing grade and, in our view, could signal potential for improvement. But AsiaInfo certainly has its work cut out, and will need to make faster progress in its AI and space transition to win back investors and regain some of its former dot-com bubble glory of earlier days.

To subscribe to Bamboo Works free weekly newsletter, click here   

The company posted triple-digit growth in the first half of 2026, with revenue for the six-month period already equal to 70% of its 2025 total

Key Takeaways:

• Deepexi said it expects to report its revenue grew by 101% to 120% year-over-year in the first half of 2026
• The agentic AI company's price-to-sales valuation is significantly lower than that for better-known peer Palantir

By Bai Xin Rui

As the global AI frenzy runs its course, people are increasingly less impressed with the simple conversational chit-chat that once wowed the world. Instead, enterprises are craving more practical AI in the form of “digital employees” who can understand and make intelligent decisions based on their business scenarios.

Deepexi Technology Co. Ltd. (1384.HK) is trying to meet that demand with its enterprise-level large language models and agentic AI applications, and, based on its latest upside financial forecast, issued earlier this month, is making strong progress in that direction. That forecast showed the company expects to report its revenue more than doubled in the first half of 2026, rising between 101% and 120% to between 266 million yuan ($39 million) and 291 million yuan.

Deepexi attributed the growth to the upgrade of its FastAGI enterprise-level AI solution to its DeepexiOS AI-level enterprise operating system platform solution. The upper limit of its expected revenue range would already equal 70% of the 415 million yuan that Deepexi recorded for all of 2025.

Origins in Huawei, Alibaba Cloud

Deepexi traces its history back to 2018 with its founding by executive directors Zhao Jiehui and Yang Lei. Both previously worked at Huawei, and Zhao subsequently went on to work at e-commerce giant Alibaba’s cloud computing unit. Deepexi launched its FastData enterprise-level data intelligence solution in 2019, with a focus on enterprise management.

Before the latest triple-digit gains, Deepexi's revenue grew 70.8% for all of last year. Within that total, contributions from the FastData traditional data intelligence business decreased to 38.7%. As its revenue growth was accelerating, the company attracted big-name cornerstone investors to its Hong Kong listing last October, including Hillhouse, China Minsheng Bank and China Merchants Group.

The engine behind the accelerating growth is the DeepexiOS AI-level enterprise operating system, which integrates Deepexi’s enterprise large model, the FastAGI enterprise intelligent agent platform, and its FastData Foil enterprise data fusion platform. The operating system is used to construct a multi-layered matrix of AI agents, or “digital employees,” specifically tailored for different business scenarios.

The current field of enterprise AI solutions in China is still mostly at the retrieval-augmented generation (RAG) + knowledge graphs phase of development. Such solutions essentially use “external retrieval plugin” models, which are highly susceptible to losing their value as the context windows of large models expand. Deepexi is trying to carve out an alternative path, targeting complex business scenarios in specific industries, such as manufacturing, which accounted for 51% of its revenue last year, and retail consumption, which made up 30%.

Take the manufacturing industry, for example, which involves massive amounts of information such as engineering drawings, scientific computations and sensor data. Deepexi deploys its core weapon, an enterprise ontology corpus database of up to 1.1 TB, to directly replace professional roles, such as AI process validation engineers and NC researchers. That provides the company with a dual moat of “industry knowledge + model capabilities,” which the company bills as its core competitive advantage.

It’s worth noting that the concept of AI agents, or intelligent agents, was included in China’s national policy documents for the first time this year. This top-level inclusion, which signals strong support from Beijing, has prompted large state-owned enterprises, industrial giants, and government agencies to significantly tilt their IT procurement budgets toward the use of intelligent agent platforms and digital employees.

China's Ministry of Industry and Information Technology (MIIT), a leading high-tech regulator, has rolled out procurement policies specifically targeting AI and digital employees, offering special subsidies as high as 30% to 50% aimed at transforming the manufacturing sector. Such policies have substantially lowered procurement costs for small- and medium-sized enterprises (SMEs) as well as larger discrete manufacturing plants, providing strong policy tailwinds for companies like Deepexi.

More cost-effective than Palantir

Deepexi's implementation and delivery cycle requires only four to six weeks, which is faster than the higher-profile U.S.-listed Palantir's (PLTR.US) timeframe of approximately eight to 12 weeks, according to a report from Soochow Securities. Deepexi's accuracy rate for complex problems reaches 92%, far higher than the 78% rate for Palantir. More importantly, the average project quotation for Deepexi Technology is a relatively modest 3 million yuan, which is substantially less than Palantir rates that typically range between 5 million yuan and 8 million yuan.

Deepexi is well positioned for strong growth as demand for agentic AI takes off. Revenue from AI applications for the manufacturing industry alone is expected to grow at an average annual rate of 65% from 2025 to 2029, according to third-party data in Deepexi’s financial disclosures.

While Deepexi currently possesses a competitive advantage through its highly specialized products, it could quickly lose that high ground later if other more generalized operators like OpenAI, Anthropic or Llama can encroach on its domain. That could happen as maturing technology enables breakthrough capabilities, including long-context reasoning, coding, and long-term task planning capabilities. That could drive down costs to very low levels, flooding the market with cheap solutions directly tied to general models equipped with external RAG plugin.

That means Deepexi will need to keep spending huge amounts on R&D to maintain its technological advantage, presenting one of the biggest risks for AI companies in general.

Deepexi isn’t limiting itself to China, and is accelerating its expansion into overseas markets like the Middle East and Southeast Asia. But given that its core enterprise data, industrial process drawings, and other similar information all constitute extremely sensitive information, figuring out how to comply with China’s stringent local data security regulations could also present a significant challenge for the company in that global expansion.

Soochow Securities estimates that Deepexi's revenue will surpass 900 million yuan this year, which could lift the company to its first-ever annual profit of around 36 million yuan. That would give the company a price-to-sales (P/S) ratio of 12.7 times using this year’s sales, which is still well behind Palantir's trailing multiple of approximately 55 times. That could help Deepexi's stock, which initially soared after its IPO but has more recently given back most of the initial gains, to regain some of its lost momentum.

To subscribe to Bamboo Works weekly free newsletter, click here

The provider of network equipment and solutions has flagged up a jump in first-half earnings, driven by demand for high-density AI computing

Key Takeaways:

• The company expects its net profit for the first half of this year to surge between 60% and 70%, boosted by rising sales of high-performance network solutions
• It is also spending around $49 million to acquire a Shanghai-based provider of network technology, aiming to strengthen its research and manufacturing capability 

By Lee Shih Ta

The rapid expansion of AI computing is driving demand for the switches, cables and optical transceivers that help to connect high-speed networks.

A jump in projected profits at newly listed FS.com Ltd. (3355.HK) underscores this trend. The company, which supplies networking equipment and solutions to enterprise customers worldwide, served notice this week that it expects its half-year revenues to rise by at least 25% and its net profits to increase by 60% or more.

Citing growing demand for high-performance products tailored for AI clusters and data centers, the company forecast revenues for the first six months of this year would range between 1.75 billion yuan and 1.78 billion yuan ($258 million and $262 million), translating into a rise of up to 27%.

The bottom line for the half year was forecast to range from 437 million yuan to 465 million yuan, a leap of between 60% and 70%.

Based on that, the company’s net profit margin would rise from about 19.5% in the first half of last year to between 25% and 26.2%, indicating that new revenue is being more readily converted into profit.

The growth was driven by accelerating sales of value-added products with transmission speeds of 100 gigabits per second, or above. Revenue from the high-performance segment grew around 29% to 1.07 billion yuan in 2025, accounting for just over 36% of total turnover, while gross margin rose from 44.8% to 49.5%, mainly driven by increased selling prices and a greater contribution from high-margin products.

Rising network demand

Growth in high-performance network solutions is even faster, projected at 45.4% in the first half after a rise of 40.8% in the first quarter, well above the overall pace. By comparison, revenue from general network solutions edged up a mere 3.6% to 1.55 billion yuan in 2025, as momentum shifts toward AI computing clusters and data centers.

FS.com has been upgrading its systems and management platforms for cloud-based networks, expanding beyond sales of optical modules, switches and cables into network design, deployment and management services. The firm’s proprietary platform integrates product information, consulting solutions, warehousing and delivery. As order volumes grow, the company said it was able to spread the fixed costs of R&D and service teams across a larger revenue base.

Margins have not been lifted by AI demand alone, benefiting also from price increases related to U.S. tariffs, as well as a bigger income share from high-speed optical modules and fiber-optic cables. FS.com has also scaled back purchases from Chinese suppliers while sourcing more products from Southeast Asia to ease the tariff impact and make its supply chain more resilient.

The strategy has helped to protect profits, but gross margin could come under pressure if the supply of high-speed products increases, or rivals cut their prices to win orders. With specifications for optical modules and switches evolving rapidly, FS.com will likely need to maintain its R&D investment to stay competitive.

Geographic concentration presents another risk. The United States supplied about 53.6% of revenue last year, up from 46.8% in 2024. The U.S. market is the biggest source of growth, but the company is left vulnerable to changes in trade relations, tariffs and rules of origin.

Buying a loss-making company

With an increasing focus on high-performance network solutions, FS.com is expanding beyond sales of individual products into the business of equipment development, systems integration and large-scale delivery. In early July, it announced plans to acquire the equipment developer Shanghai Baud Data Communication Co. Ltd. for 330 million yuan, aiming to gain greater control over core products and shorten delivery cycles.                                                                    

Baud primarily develops and manufactures routers, switches, optical access products, wireless networking products and network security products. It has its own network operating system, as well as software and hardware research teams and production facilities in Shanghai. FS.com plans to combine Baud’s research and manufacturing capabilities with its own customer base, network solutions and overseas sales channels.

But the acquisition could increase pressure in the near term. Baud logged revenue of about 463 million yuan last year and made a net loss of 64.58 million yuan, followed by another net loss of 35.81 million yuan in the first five months of this year. It remained in a net liability position at the end of May. If the integration proceeds more slowly than expected, additional losses and manufacturing costs could offset some of the margin gains generated by the AI-related business.

FS.com stock is not expensive at its current level, but it is not cheap enough for investors to ignore the risks. The company’s stock edged down 0.11% to HK$35.12 after the positive profit alert, leaving it about 15% below the issue price when the firm listed in March. The stock trades at a price-to-earnings ratio of around 20.6 times, below the roughly 25 times for telecoms and network solutions provider ZTE (0763.HK; 000063.SZ).

Investors appear to doubt whether the company can sustain its high margins and rapid growth over the long term. If demand for AI networking keeps rising, driving sales of high-performance products, the stock could have scope to rise. But investors are likely to remain cautious for now.

To subscribe to Bamboo Works weekly free newsletter, click here

Rebounding lithium prices returned the miner to the black this year, but falling spot prices and planned supply restarts are testing the durability of its rebound

Key Takeaways:

• Ganfeng Lithium returned to the black in the first half of the year with a profit of 3.65 billion yuan to 4.6 billion yuan for the six-month period
• Rebounding lithium prices have already pulled back from recent highs, as Ganfeng hopes to leverage volume ramp-ups at its mines to lower costs

By Lee Shih Ta

After a rough period of nearly three years, the lithium market has finally rebounded this year, driving a rapid recovery for miners like Ganfeng Lithium Group Co. Ltd. (1772.HK; 002460.SZ). This industry leader, whose business spans lithium mining, lithium salts, and battery production, is once again at the forefront of a nascent recovery for its cyclical sector.

Ganfeng said last week it returned to the black in the first half of 2026 with a profit of 3.65 billion yuan ($539 million) to 4.6 billion yuan, reversing a net loss of 531 million yuan a year earlier. Excluding non-recurring items, the first-half profit ranged between 3 billion yuan and 4.2 billion yuan, compared to a year-ago loss of 913 million yuan.

It’s worth noting that Ganfeng’s return to the black extended from the first into the second quarter of this year. The company, whose lithium products are a key component in new energy batteries, previously reported its revenue rose by 143.8% year-over-year in the first quarter to 9.2 billion yuan, while its profit totaled 1.84 billion yuan, or 1.42 billion yuan excluding non-recurring items.

Based on calculations using those figures, the company earned a profit of 1.81 billion yuan to 2.76 billion yuan in the second quarter. Ganfeng’s first-half profit is already more than double 1.61 billion yuan profit it recorded for all of 2025, reflecting improvements to its financial health with rebounding lithium prices and higher utilization rates.

The latest profits also include income from asset disposals, as the company sold a portion of its shares in PLS Group during the period, and recorded an increase in investment income from associates and joint ventures. But price hikes for lithium salts, alongside Ganfeng’s ramp-up in output, alongside higher battery production and sales, were the main pillars of this year’s turnaround.

Price rebound

Oversupply in 2024 and 2025 caused lithium prices to hover at low levels for a prolonged period, forcing a large number of higher-cost mines to halt production or delay development. Other factors also affected the industry, including a suspension of production at some of Ganfeng’s lithium mines in Jiangxi province in the middle of last year, and Zimbabwe’s tightening of lithium raw material exports at the beginning of this year.

The resulting decline in inventory, combined with growing demand for data center-based energy storage products created by the rise of AI, have helped to further support lithium prices. As of early June, lithium hydroxide contracts on the CME were up 86% over the last year, returning above the $20,000-per-metric-ton mark for the first time since the end of 2023.

That said, lithium prices have recently pulled back from recent highs. As of July 16, the spot price of battery-grade lithium carbonate in China was about 151,000 yuan per metric ton, down about 10.9% over the past month, although still significantly higher than lows from the middle of last year. Rising expectations for resumption of production in the Jiangxi mining areas, combined with growing expectation for the restart of suspended projects in regions such as Australia, are once again raising concerns that growing supply may outpace demand.

During the latest downturn, the nature of lithium demand has also changed. Electric vehicles (EV) still account for roughly 70% of demand for lithium batteries, but global EV sales only grew by 0.9% in the first five months of this year, while China’s sales actually fell by 15%. Grid-scale energy storage has rapidly picked up the slack, with global installations up by more than 20 times over the last five years to account for about 15% of battery demand last year. The storage industry’s use of lithium iron phosphate (LFP) technology has created a strong second demand curve for the metal, making lithium’s recovery prospects stronger than for other battery metals such as cobalt and nickel.

Ramping up self-owned mines

For Ganfeng, rising lithium prices are only half of its profitability equation. The other half stems from its self-owned resources. The company’s latest profit forecast points out the ongoing addition of new capacity from its lithium projects is an important factor driving its improving cost structure. Phase one of its Goulamina project in Mali is expected to add 506,000 metric tons of lithium concentrate in annual capacity, after already producing 336,600 metric tons last year. Its Cauchari-Olaroz salt lake project in Argentina produced 34,100 metric tons of lithium carbonate last year, with a target of 35,000 metric tons to 40,000 metric tons this year. And the first phase of its Mariana salt lake project in Argentina has also begun production, with annual capacity of 20,000 metric tons of lithium chloride.

Smooth development of those projects will further lessen Ganfeng's reliance on externally purchased lithium ore, which could magnify profits during periods of high prices. And even when prices pull back, self-ownership of its lithium supplies could provide a better buffer. As it produces more of its own lithium and prices recovered, the company's overall gross profit margin last year rose more than 4 percentage points to 15.8% from 11.4%. That figure surged to about 29.7% in the first quarter, based on calculations using revenue and operating costs from the first quarter of this year.

Ganfeng's shares have largely moved in sync with lithium prices lately. When expectations for tighter supplies heated up, the company's Hong Kong-listed shares rose swiftly in May to a 52-week high of HK$91.20 on May 8. But the shares nosedived after that as lithium carbonate prices pulled back, closing at HK$39.84 on July 16, down by more than half from their peak. The stock also fell 4.5% over the two days after its earnings forecast last week.

The analyst community is generally positive on the company, with Soochow Securities maintaining a “buy” rating on its estimates that lithium carbonate prices will range between 150,000 yuan and 180,000 yuan per metric ton this year. But that may be overly optimistic, given the most active lithium carbonate contract had already sunk below that level, to about 147,000 yuan per metric ton, on July 16.

The recent plunge in Ganfeng’s share price seems to indicate that market focus has shifted from the company’s individual performance to the broader outlook for lithium prices. If those prices continue to fall, pressures from inventory, liabilities and overseas investments could continue to haunt the company’s stock in the months ahead.

To subscribe to Bamboo Works weekly free newsletter, click here

The SLC NAND flash memory maker is capitalizing on strong demand for its chips, which lifted it into the black this year

Key Takeaways:

• XTX Technology has filed to list in Hong Kong, reporting it swung to a profit in this year’s first quarter on booming demand for its memory chips
• Prices for the company’s SLC NAND flash memory more than tripled during in the first three months of 2026 from a year earlier

By Bai Xin Rui

The global AI explosion is shining a spotlight on the importance of related infrastructure, touching off parallel booms for related hardware like servers, computing chips and data centers. That’s lit a fire under many companies’ stocks, especially makers of computing and memory chips used in high-end AI computing.

Now, XTX Technology Inc., one of China’s leading fabless code storage flash memory chip makers, is hoping to ride that wave to a Hong Kong IPO, shifting gears from its previous plan to list on the country’s domestic markets in Shanghai and Shenzhen. The company submitted its IPO application to the Hong Kong Stock Exchange earlier this month, with heavyweight Citic Securities and the smaller GF Securities as joint underwriters.

XTX was founded in April 2014 by Chairman Long Dongqing, a chip veteran whose resume includes engineering stints at Japan’s Renesas (6723.T), Europe’s STMicroelectronics (STM.US), and America’s Freescale Semiconductor, now part of Dutch firm NXP (NXPI.US). Long currently holds 34.9% of the company, while Ningbo Hongshan Zhisheng holds 9.6%. Pu Xun, a managing director at private equity firm HSG, formerly Sequoia China, also serves as a non-executive director.

XTX designs and sells code-storage flash memory chips, a category that includes NOR flash and SLC NAND flash, which are manufactured by third-party fabs. As AI gains traction, it is rapidly generating strong demand for such chips to meet the technology’s high performance computing needs. XTX ranks fourth globally among fabless makers of SLC NAND flash, its main product. Boasting high endurance and low error rates, such chips are widely used in server system drives, industrial equipment, caching solutions and medical and military equipment.

The global market for SLC NAND flash was worth $5.1 billion in 2025. But with the widespread adoption of AI, the figure is expected to balloon to $42.7 billion by 2030, growing at a breakneck average annual rate of 53% over that time, accounting for 7.9% of the overall flash memory market, according to third-party market data in the listing document.

Global tech giant exodus

Current global leaders in the SLC NAND flash sector include multinationals such as U.S.-based Micron Technology (MU.US) and Japan's Kioxia (285A.T). But most of those have announced plans to discontinue or downsize their legacy SLC NAND flash production to pivot toward the mid- to higher-end of the memory market.

Despite their retreat, demand for legacy SLC NAND flash remains strong, with downstream clients aggressively stockpiling inventory in the second quarter of this year. As supplies shrink, the average price of SLC NAND flash skyrocketed between 130% and 150% in the first half of this year alone.

Analysts expect price increases to moderate slightly in the second half of the year, but are still forecasting additional gains of 70% to 75%. The strongest gains are expected for industrial and automotive-grade products, which require more rigorous specifications.

XTX derives most of its revenue from SLC NAND flash. Its revenue reached 224 million yuan ($33 million) in the first quarter of this year, up 77.4% year-over-year. SLC NAND flash generated 149 million yuan during the quarter, accounting for about two-thirds of total revenue and up 120% annually. The segment’s gross profit hit 101 million yuan, up more than 10-fold year-on-year, accounting for 81% of the company’s overall gross profit.

Sliding volume, soaring prices

It’s worth noting that XTX’s sales volume actually fell this year, even as its revenue and profit soared. The company sold just 11.5 million units in the first quarter, down 34.4% year-over-year. Constrained by a tight supply of semiconductor wafers, one of the main components for chip manufacturing, XTX strategically allocated its limited inventory to higher-value clients.

Skyrocketing prices for SLC NAND flash more than offset the lower chip output, as XTX’s average selling price more than tripled from 3.89 yuan per unit in 2025 to 13.04 yuan per unit this year. That surge lifted the company to a 75.89 million yuan profit in the first quarter of this year from a 2.7 million yuan net loss a year earlier.

While surging prices lifted XTX into the black, ongoing shortages of upstream wafers remain one of the company’s biggest risk factors over the near-term, especially if its sales volumes continue to slide. Potential reversals by international giants Micron and Kioxia, if they decide to re-focus on the segment again, remain another major variable that could dictate whether XTX can sustain its newfound profits.

At the same time, the global memory chip sector has historically been very cyclical. Its boom-and-bust cycles were deeply intertwined with the smartphone and PC sectors, which traditionally depressed their valuations below 15 times forward price-to-earnings (P/E). Such levels were far lower than the multiples enjoyed by other semiconductor peers, such as CPU or GPU firms, whose valuations often exceed 30 times.

Investors are currently laser-focused on whether explosive growth in AI demand could perhaps strip memory chip stocks of their cyclical label and lead to a longer-term expansion, lifting the stocks to higher multiples. But skeptics question if real-world AI applications can meaningfully boost these chipmakers’ earnings, since prices will inevitably come down as producers boost their capacity. Should the AI narrative fail to materialize or stumble, memory chip valuations across the board could come under intense pressure.

Despite their recent stock run-ups, shares of Micron and Kioxia still only trade at modest estimated P/E ratios of 8 times and 9 times, respectively. That means XTX might need to be conservative in pricing its IPO shares to attract investor interest. It could try to be more aggressive by targeting investors who believe that memory chip stocks are set to shed their cyclical nature, though that could scare away many of the AI skeptics. Either way, XTX looks relatively well positioned to ride the upside of expanding valuations as long as chip prices keep rising and that translates to growing profits.

To subscribe to Bamboo Works weekly free newsletter, click here

The Chinese enterprise software pioneer is pitching AI as its next growth engine, even as the technology forces investors to rethink the future of the global software industry

Key Takeaways:

• Yonyou is seeking a Hong Kong listing after years of weak revenue growth and heavy losses, as AI reshapes the economics of enterprise software worldwide
• The Chinese enterprise software leader argues AI will make enterprise software more valuable, even as some bet AI will eventually make traditional software vendors irrelevant

By Hu Minghe

When Wang Wenjing borrowed 50,000 yuan – a huge sum at the time – to start his accounting software company in Beijing in 1988, his mission was straightforward: replace paper ledgers with computers.

Nearly four decades later, the founder of Yonyou Network Technology Co. Ltd. (600588.SH) faces a much bigger challenge. Instead of persuading companies to adopt his software, he must now convince investors that software itself will remain indispensable in the age of artificial intelligence.

The challenge arrives at a delicate moment. Yonyou, China’s largest enterprise software and service provider by revenue, filed its third Hong Kong listing application last week, after two previous applications lapsed, following several difficult years. Its revenue rose only marginally to 8.86 billion yuan ($1.24 billion) in 2025 from 8.82 billion yuan a year earlier, after falling from 9.44 billion yuan in 2023. Meantime, its net loss narrowed to 1.35 billion yuan last year from 2.07 billion yuan in 2024. At the same time, the company says aggregate AI-related customer orders jumped from essentially zero two years ago to 1.88 billion yuan by the end of this year’s first quarter.

Yonyou is hardly alone in navigating the transition to an AI era that could either make or break the company. Enterprise software companies worldwide are confronting their biggest disruption since cloud computing. For decades, software vendors enjoyed one of technology’s most attractive business models: recurring subscriptions, high switching costs and predictable customer retention. AI is beginning to challenge all three assumptions.

Rather than logging into enterprise software, company employees may increasingly ask AI agents to retrieve data, prepare budgets, reconcile accounts or approve purchases – functions traditionally handled by software developed by companies like Yonyou.

Nearly every major enterprise software company is racing to adapt. Domestic rival Kingdee (0268.HK) is rolling out AI-powered enterprise resource planning (ERP) products, while global competitors Salesforce (CRM.US) is betting on Agentforce, SAP (SAP.DE) has launched Joule, and Oracle (ORCL.US) is integrating generative AI throughout its enterprise applications. Even Microsoft (MSFT.US) has embedded Copilot across much of its software portfolio. The race reflects a growing belief that the next-generation of enterprise software will revolve around AI agents instead of menus, dashboards and forms.

Investor caution

Investors have responded cautiously. Software stocks have lagged the broader technology sector for much of this year as money flowed instead toward companies supplying AI infrastructure such as chips, servers and memory. Traditional software vendors have come under pressure as investors debate whether AI will weaken their subscription-based models and make it easier for companies to build customized internal applications that skirt the need for third-party software.

Yonyou also faces additional challenges as AI arrives before it finishes tackling older problems. Unlike many software companies now grappling with the AI disruption, Yonyou’s performance was already deteriorating years earlier as its transition from traditional enterprise software to cloud services proved slower and more expensive than expected.

Management says increasingly complex projects for large enterprise customers have lengthened implementation cycles and delayed revenue recognition. Continued investment in its YonBIP cloud platform, AI development and higher depreciation and amortization costs have further weighed on the company’s profits. Those traditional challenges were facing software enterprises before. Now, AI simply raises the stakes.

The transformation is already reshaping Yonyou itself. The company’s workforce fell from 24,949 employees at the end of 2023 to 17,747 by the end of March this year, a reduction of nearly 30%. Chinese media also reported multiple rounds of layoffs this year, affecting implementation, sales and other business units as the company streamlined operations and accelerated its AI strategy. Rather than simply cutting costs, the restructuring reflects a broader effort to reduce the labor-intensive model that has long characterized China’s enterprise software industry.

That restructuring highlights one of AI’s biggest implications for enterprise software. Historically, companies like Yonyou generated much of their revenue through labor-intensive implementation, customization and maintenance services. AI-assisted coding, automated testing and customer support could reduce the need for thousands of engineers and consultants, making enterprise software companies more scalable. But the same technology also makes it more practical for customers to develop their own software or rely on AI-generated workflows instead of buying traditional third-party products like Yonyou’s.

On the winning side of history

Yonyou is telling investors it can stay on the winning side of that equation.

The company launched its YonGPT large language model in 2023 and introduced Yonyou BIP Enterprise AI last year. Rather than replacing traditional enterprise resource planning (ERP) systems, management argues AI agents will need trusted enterprise data, business rules and workflow engines that ERP platforms like itself already provide. In that vision, ERP becomes the operating system behind enterprise AI.

Whether customers agree with that vision remains the million-dollar question.

AI is arriving at a pivotal moment for China’s enterprise software industry. For years, the sector has lagged its U.S. counterpart in both software spending and profitability. Chinese companies have generally invested less in IT, relying on relatively inexpensive labor instead, and preferred highly customized systems over standardized subscription software. As a result, many domestic software vendors have operated more like consulting firms as opposed to software companies, relying on large, customized implementation teams that are difficult to scale.

Yet AI could also become the industry’s biggest opportunity. China’s enterprise software market remains less mature than those in developed economies, leaving considerable room for digitalization as companies adopt AI. HSBC analysts recently argued investor concerns that AI will replace software vendors may be overdone. They said Chinese companies such as Yonyou could instead benefit because they already possess deep expertise in enterprise workflows, regulatory requirements and business data that large language models alone cannot provide.

Yonyou sits at the nexus of those competing narratives. Its prospectus highlights many of the industry’s long-standing challenges, including increasingly customized projects that have slowed product delivery, and weaker IT spending by government and public-sector customers. But it is also betting those same enterprise relationships will become an advantage as companies deploy AI across finance, manufacturing and supply chains. The company says 78% of Fortune China 500 companies use its products, while its 3.8% market share still makes it China’s largest enterprise software provider in a highly fragmented market.

That leaves investors to decide the answer to a question facing software companies worldwide, well beyond China.

For nearly 40 years, Yonyou has successfully navigated one technological transition after another, from paper bookkeeping to PC software, and from software licenses to cloud subscriptions. Its pitch to Hong Kong IPO investors is that it can survive one more such challenge.

To subscribe to Bamboo Works weekly free newsletter, click here

The maker of power units for heavy-duty crypto computing is looking to raise IPO capital for a switch into systems for AI servers and data centers

Key Takeaways:

• Boco Electronics has filed for a Hong Kong listing for a second time, citing plans to invest in R&D and strengthen its presence in the AI computing market
• Its power products for AI computing achieved a gross margin of 47.7% last year but accounted for just 0.7% of revenue

By Lee Shih Ta

At its core, the AI race is a competition for power, reliant on vast amounts of energy to drive servers and data centers.

With power supply at a premium, equipment manufacturers that serve the energy-guzzling crypto market are looking to adapt their technology to meet the rising demands of AI. One of those suppliers is Hangzhou Boco Electronics Co. Ltd., which recently reapplied for a Hong Kong IPO.

Boco makes power supply units for high-density computing as well as energy conversion and storage systems for industrial and residential use. It has been in the market for powering digital-asset mining for five years and is now using its experience as a springboard into the supply chain for AI data centers.

While the end uses differ, the underlying technologies overlap. Like data centers, cryptocurrency mining rigs need to operate reliably for extended periods under heavy processing demands and thermal loads.

Boco founder Yin Guodong graduated from Harbin Institute of Technology and spent more than a decade at Taiwan-based power supply giant Delta Electronics (2308.TW). There he rose through the R&D ranks to become technology director, leading the global development of high-end power supply equipment. After leaving Delta in 2016, he stayed in the server power sector, establishing ties with Shenzhen MicroBT, which designs and manufactures crypto-mining hardware.

Founded in 2021, Boco Electronics launched its first mining-rig power system the same year, with orders and investment from MicroBT. Income from MicroBT accounted for 89.9% of Boco’s revenue in 2023, but that figure had plunged to about 9% by the first quarter of this year. Still, Boco’s five largest customers provided 93.5% of its revenue during the same period, leaving it exposed to any purchasing shifts within its small client base.

Boco’s revenues have climbed from 261 million yuan in 2023 to nearly 1.05 billion yuan ($154 million) in 2025, while net profit jumped from 4.24 million yuan to 97.13 million yuan. Citing third-party data, the IPO paperwork described Boco as the second-largest provider of power systems for high-performance computing in mainland China by revenue in 2025, with a market share of 15.8% there and 6.9% globally.

As AI drives up server power consumption, performance needs are converging with those of mining rigs operating under sustained loads. But the products cannot simply be switched from one use to the other. They must go through a lengthy process of being redesigned, rigorously tested and then certified by server manufacturers and cloud computing customers.

AI revenue still marginal

Boco’s AI power supply products began to be adopted by customers in the fourth quarter of 2024, but the business is still far from taking over as a growth engine.

In 2025, products for mining machines and energy storage systems (ESS) generated revenue of 719 million yuan and 318 million yuan, accounting for 68.8% and 30.5% of the total respectively. Income from the AI computing sector contributed just 7.43 million yuan, a mere 0.7% of topline income.

AI-related revenue surged in the first quarter from 158,000 yuan to 1.35 million yuan, while sales volume doubled from 600 to 1,200 units. Yet the business still accounted for only 0.6% of revenue, albeit with a high margin. Its gross margin reached 47.7% in 2025, well above the 26.6% for specialized computing power equipment and 13.2% for ESS products. Boco could enjoy an earnings boost if it can bring more customers on board and put its AI-targeted technology into mass production.

In the first quarter, Boco’s revenue rose 35.3% to 215 million yuan from the year-earlier period, but the company swung to a net loss of 5.32 million yuan from a profit of 7 million yuan. R&D expenses surged about 70% to 28.5 million yuan, rising from 10.5% to 13.3% of revenue, while the company also recorded foreign exchange losses. The core business also came under pressure, as Boco cut prices to stay competitive as cryptocurrency values fell. The gross margin for its core specialized server business fell to 19.7% in the first quarter from 22.5% a year earlier, adding to pressure for a business transition.

Boco is seeking to integrate its power technologies for computing and energy storage, developing 20 kW liquid-cooled units, a power distribution system delivering 800 volts as direct current, and integrated solar-storage microgrids. Its goal is to expand beyond standalone power supplies into comprehensive systems spanning power supply, storage and load management for AI data centers. But those products remain at an early commercial stage, with no dedicated production line as yet.

Competitive pressure is also intense. Delta Electronics has launched a 120 kW rack-level power supply for AI servers and a data center solution integrating power supply, liquid cooling and energy management. U.S.-based Vertiv (VRT.US) operates across the spectrum of data center infrastructure, from uninterruptible power systems and power distribution to liquid cooling. Both companies are far ahead of Boco in market share and technology.

Even so, investors have been keen to buy into AI power and cooling stocks. Delta Electronics shares have risen about 86% so far this year, while Vertiv has climbed around 73%, showing that investors are willing to price in AI-driven demand well in advance. But AI power supply still contributes less than 1% of Boco’s revenue, while its microgrid solutions are far from launch ready. Its earnings are still fundamentally tied to mining-rig power products. Boco’s valuation will depend on how well the company can deliver new customers, orders and profits.

To subscribe to Bamboo Works weekly free newsletter, click here