As creation of content expressed in multiple formats grows, video generation costs are reshaping companies’ AI computing budgets

By LeadLeo Research Institute

Multimodal content creation is moving from a niche use of large models toward the forefront of AI applications. LeadLeo research shows that its share of nine core business application categories rose from 0.6% to 11.9% in six months, making it the fastest-growing category. The change goes beyond a shift in the rankings: Generating a 10-second, 1080p video consumes about 350,000 tokens, hundreds of times as many as a text task. The implications extend to content platforms, marketing departments and anyone planning computing capacity purchases.

Fast-changing rankings

The report examined a sample of about 700 large-model applications. In the first half of 2025, text content creation ranked first with a 23.7% share, followed by question-and-answer assistants, data processing and analysis. Multimodal content creation accounted for just 0.6%, placing last among the nine categories.

Six months later, the rankings began to change. Multimodal content creation rose to 11.9%, while intelligent customer service climbed from 5% to 9.4%, and AI search from 3.1% to 4.7%. The report identified these as the three fastest growing categories, with multimodal content creation posting the largest gain.

The core applications remained strong. Text content creation still ranked first in the second half, with a 19.5% share, while question-and-answer assistants and data processing also continued to account for substantial share. Multimodal applications captured a larger share of the growth without replacing text-based uses.

Why did multimodal applications grow? The report attributes the increase to continuing improvements in multimodal technology in 2025, which accelerated demand in content production, information retrieval and customer interactions. Content-heavy applications were among the first to gain traction: The number of games on the Steam platform using generative AI rose from more than 1,000 in 2024 to more than 7,818 in 2025, an increase the report puts at 681%.

Revenue from video generation services is also growing. Kuaishou (1024.HK) said its Kling AI service generated more than 850 million yuan ($127 million) in revenue in the second quarter of 2026, up more than 200% year-over-year. Spending on AI-generated short-video marketing materials on its platform rose more than 70% in the same period. From video production to advertising, multimodal applications are entering more commercial settings.

Changing unit of account

The shift in application share also changes how computing demand is measured. The report estimates that generating a 10-second, 1080p video consumes about 350,000 tokens, hundreds of times as many as a text task. When an application moves from writing copy to making video, the computing requirements for a single task are on a different scale.

The way models are used adds to that demand. The report observed a shift from one-off generation toward continuous inference and end-to-end tasks. As models become embedded in frequently used workflows such as search, marketing, customer service and office work, the usage volume, frequency of interactions and inference load all rise.

Token usage nationwide is growing as well. China’s National Data Administration said average daily token use exceeded 140 trillion in March this year. The National Bureau of Statistics later said the daily figure had reached “several hundred trillion,” without providing a more precise number. The measure covers all types of AI applications and reflects overall growth in model usage.

Taken together, these figures change how companies calculate their computing budgets. While the initial question was whether a model could perform a task, costs now need to be measured in terms of token usage volume multiplied by consumption per task. As work shifts from text to video and from one-off generation to continuous use, costs may grow faster than the user base.

Who pays first?

The cost pressure won’t be shared equally. Multimodal growth is concentrated in content-heavy applications such as games, leaving internet content platforms and AI-native companies among the first to bear the costs. The report says these customers need high concurrency, elastic capacity and low-cost usage models, making model-as-a-service (MaaS) and cloud-based inference services a fit for usage-based pricing.

The scope of the figures matters. The application shares are based on a sample of about 700 and reflect changes in that sample’s usage mix, not the distribution of industry revenue. Whether multimodal applications continue to grow also depends on how quickly video generation costs fall. The report identifies pricing and the maturity of available services as key variables affecting computing demand.

For companies, the implications come down to three considerations: including multimodal production costs in annual budgets; specifying how model usage will be billed in procurement contracts; and assessing, task by task, which processes warrant an upgrade from text to video generation. In planning an AI budget, companies first need to determine how many tasks will move from text to video and from one-off generation to continuous inference.

Multimodal AI is more than just another feature option: It can multiply the computing required for a single task hundreds of times. Rather than focusing only on model rankings, companies should calculate how many tasks in their business will shift from text to video and from one-off generation to continuous inference. The first shift affects the experience, while the second affects the cost.

LeadLeo Research Institute is an original content platform for research on banks and companies and an innovative digital research service provider with nearly 100 senior analysts. You can contact the platform at CS@leadleo.com

This commentary is the views of the writer and does not necessarily reflect the views of Bamboo Works

To subscribe to Bamboo Works weekly free newsletter, click here

As China’s humanoid robot industry moves into mass production, commercial use remains constrained by technical limits and uncertain returns

By CLS Marketwatch

Robotics have become a flavor of the day in China, increasingly known for high-profile products with imaginative flourishes. But as pioneering companies get set for mass production, entertainment performances and data research still account for over 60% of applications, while penetration into the industrial and service sectors has only just begun. Amid all the hype, attention is increasingly turning to practical questions that arise once the spotlight dims: In what scenarios can these costly humanoid robots truly “get to work,” and can they truly achieve commercial viability?

Shipments tell only half the story

Global humanoid robot shipments reached 22,000 units in the first half of this year, up nearly 300% year-on-year. China's domestic market accounted for nearly half of that, with humanoid robots sales hitting the 10,000-unit level – a ninefold increase from the previous year. Some giants have already established, or are close to establishing, annual production capacity of tens of thousands of units, with critical hardware achieving mass production as well.

But high growth in shipment volumes doesn’t equate to major progress in commercial applications. A breakdown of global sales by application exposes a clear structural imbalance: entertainment and performance applications account for 33.6% of shipments, while data production and scientific research applications make up 27%, exceeding 60% of the total combined. Service and guidance applications represent about 19% of shipments. And the share that has truly entered intelligent manufacturing and warehousing logistics stands at less than 20%, meaning the integration of robots as tools into frontline production scenarios remains very limited.

This data exposes a core contradiction: while the industry has acquired manufacturing capacity to deliver tens of thousands of units, genuine production-oriented application scenarios have not yet become the primary demand driver. A large number of products remain at the technical demonstration stages, primarily focused on entertainment functions, with a long way to go to the factory floor.

What's holding things back?

The transition to practical industrial applications faces three core barriers.

One is the severe lack of generalization capabilities in embodied intelligence. While current systems show impressive task success rates in fixed scenarios, the rates drop sharply when objects or environments change, directly limiting the ability of robots to perform in complex and ever-changing real-world settings.

The industry also has yet to form a technically unified and practically deployable pathway across vision-language-action and other large-model architectures. Humanoid robots need to overlay the brain's reasoning and task-decomposition functions on top of the cerebellum's motion control capabilities, while also achieving precise coordination between the two. This is the greatest weakness in the current technological framework.

Even if intelligence-related issues are addressed, engineering challenges remain just as severe for mass production. The absence of unified industry-wide standards makes it difficult to ensure consistency across production batches, becoming a core pain point in ramping up production. A comprehensive standardized system for manufacturing efficiency, yield rates, and hardware and software has yet to be fully established across the numerous phases, from planning to stable large-scale deliveries. In short, the industry still lacks a proven methodology to guarantee product consistency, stability, and reliability at large-scale production levels.

What ultimately determines a business’ fate is the viability of its economic model. Purchasing decisions by industrial customers follow a strict logic: they will only place bulk orders if the robots can pay for themselves within two to three years. Given their current high bill-of-materials (BOM) costs, humanoid robots are only economically competitive for replacing labor in certain positions where recruitment is difficult and labor costs are high.

Long road from the exhibition floor

In light of these bottlenecks, many in the industry generally believe commercial adoption of humanoid robots will proceed in a stepwise manner. High-risk areas such as inspection and security have already seen a first wave of adoption; the second tier includes physically demanding scenarios like heavy lifting and warehouse sorting, with tangible cases already emerging in relatively standardized workflows such as factory assembly and coffee making.

Beyond this, robot applications will extend to more complex business services. Meanwhile, consumer-facing household deployments are expected to face the longest rollout cycle. The first robots to enter homes are likely to be for companionship and interaction-oriented functions, before progressing to performing complex household chores.

Notably, some scenarios have started employing a "human-robot collaboration" model. For instance, in airport baggage transfer, robots handle about 80% of standard items, while humans manage the remaining 20% of irregularly shaped pieces, maximizing efficiency and reliability. Many firms are also experimenting with a shift from selling equipment to charging based on workload, making the supplier responsible for service outcomes. Viability of this model may propel more potential customers to make procurement decisions.

Just a few years ago, vivid imagination was the biggest driver of what humanoid robots might be able to do someday. Today, however, discussions are centered on more practical matters like yield and repeat purchase rates, and return-on-investment cycles. Tens of thousands of units shipped is a milestone, but doesn’t equal a revolution. That said, the industry is steadily bridging the gap between the exhibition floor and real-world applications, banking on scenarios like high-risk inspections, heavy-lifting logistics, human-robot collaboration and pay-for-performance models.

CLS Marketwatch provides insights and analysis on China’s industries. You can contact the author at liujingyi@cls.cn

To subscribe to Bamboo Works weekly free newsletter, click here

Nvidia is moving beyond chips into AI model gateways, while Chinese models are expanding their global reach, as competition shifts to other areas

By Lee Shih Ta

Chip giant Nvidia’s (NVDA.US) new agreement to acquire Hugging Face for $12.93 billion is much more than the purchase of a website that hosts AI models. Nvidia has long dominated the AI chip market and the Cuda software ecosystem, while Hugging Face connects more than 18 million developers and about 200,000 companies with over 3 million models. If completed, the transaction will take Nvidia beyond its roots in computing infrastructure into model discovery, testing, downloading and deployment, bringing it closer to the gateway where companies decide which models to use and where to run them.

Nvidia’s move is not isolated. Microsoft (MSFT.US) offers more than 11,000 models through Microsoft Foundry, while Amazon’s (AMZN.US) Bedrock and Alphabet’s (GOOG.US) Google Model Garden also combine models, development tools and cloud services into enterprise deployment platforms. Meta (META.US), meanwhile, is using its open-weight Llama models to expand its developer ecosystem. The technology giants are pursuing different strategies, but the competition has expanded beyond model capabilities to developers, computing power and enterprise deployment gateways.

Chinese open-weight models are emerging as key players in this shift. A report released by Hugging Face in March showed that models developed in China accounted for 41% of downloads on its platform over the past year, surpassing the share of U.S. models. The Qwen family of models from Alibaba (9988.HK; BABA.US) has spawned more than 113,000 derivatives, exceeding the combined total for Google and Meta. Models including DeepSeek, Qwen, GLM from Z.AI (2513.HK), Kimi from Moonshot AI and those from MiniMax (0100.HK) are increasingly being used by developers for fine-tuning, compression and redeployment, with some beginning to serve as the technical backbone of overseas projects.

Chinese companies primarily provide foundation models, Hugging Face connects models with developers, and Nvidia supplies chips and development tools. The three do not compete directly, but collectively influence how companies adopt and deploy models, while everyone seeks to capture developers, customer relationships and revenue within the industry’s value chain.

From downloads to overseas projects

Humain, an AI company under Saudi Arabia’s Public Investment Fund, recently unveiled the Arabic-language humain-m3 model. Built on MiniMax-M3, the model was further pre-trained on more than 1 trillion native Arabic tokens and has 428 billion parameters. It is currently available in research preview on the Humain Node. Humain plans to make the model weights available after completing safety training and alignment.

For Saudi Arabia, adopting an existing Chinese model as its foundation can help Humain shorten development time and focus its resources on Arabic-language data, localized applications and enterprise services. It also provides Chinese AI companies with another path into overseas markets. However, Humain still controls the local platform and customer relationships, while cloud service providers and chipmakers can also generate revenue from deployment demand. How much value MiniMax can capture will depend on licensing, joint development and ongoing service arrangements.

Nvidia’s revenue rose 106% year-on-year to $96.2 billion in its latest quarter, including $89 billion in data center revenue. But major customers, including Microsoft, Amazon, Google and Meta, are all developing their own AI chips. Acquiring Hugging Face would give Nvidia earlier access to demand for model development and deployment, while allowing it to reinforce its hardware advantage through software and inference services. Nvidia has pledged that Hugging Face will remain open and continue supporting different clouds and computing hardware, without requiring users to adopt Nvidia chips. Still, combining the two companies could bring more development activity into Nvidia’s software and computing ecosystem.

Turning influence into revenue

Chinese companies have begun trying to generate revenue from application programming interfaces, private deployments for enterprises, model fine-tuning, AI agent tools and cloud computing. Alibaba has a relatively complete commercial ecosystem, with Qwen expanding its developer base and Alibaba Cloud providing the computing power and enterprise services needed to run the models.

Alibaba’s revenue from AI-related products reached 12.4 billion yuan ($1.85 billion) in the quarter through June, marking its 12th consecutive quarter of triple-digit growth. While the company didn’t disclose how much of that revenue came directly from Qwen, the complete value chain formed by its models, cloud computing and enterprise services provides a clearer path to monetization.

The challenge is greater for independent model developers such as MiniMax, Z.AI and Moonshot AI. Open weights can help them rapidly grow usage, derivative models and brand influence, but training, inference and overseas services remain costly. If developers download their models and deploy them on other cloud platforms, these companies may gain little beyond name recognition, while chipmakers, cloud platforms and systems integrators capture the revenue. Overseas licensing, joint development and enterprise deployment projects therefore provide a better test of their business models than download figures alone.

Nvidia’s acquisition of Hugging Face shows that value in the AI industry is being redistributed among models, platforms, computing infrastructure and enterprise services. Chinese models have already established an advantage in downloads and developer adoption, while humain-m3 shows that they are beginning to enter large overseas projects.

The next test is whether model developers can retain control of customer relationships and revenue as platforms, computing providers and cloud giants compete to capture value. Download figures can demonstrate influence, but only sustained spending by enterprise customers can translate such influence into a durable business.

Lee Shih Ta is an editor at Bamboo Works.

You can contact him at shihtalee@thebambooworks.com

To subscribe to Bamboo Works weekly free newsletter, click here

The advanced technology is spreading from consumer desktops to high-end manufacturing, driven by exports, lower prices and AI adoption

By CLS Marketwatch

As the key driver behind additive manufacturing, 3D printing technology has become a transformative force by disrupting traditional production paradigms. Today, however, this industrial-grade technology is rapidly making its way into households. From toys and flowerpots to footwear, users can now produce items on command using simple materials and ready-made models. At the same time, building on a robust manufacturing foundation and supply chain strengths, the 3D printing industry is working its way into China's high-end manufacturing systems at an unprecedented pace.

3D printing’s wild ride

3D printing is one of the fastest-growing segments of China’s manufacturing sector. In the first half of 2026, the country's 3D printing equipment output jumped by 48.5% year- on-year, ranking first in terms of growth among all major industrial products.

Meanwhile, China is also filling the global demand gap for 3D printing. The country exported a cumulative 3.62 million 3D printers in the first half of 2026, with export value reaching 9.61 billion yuan ($1.43 billion), up 90.2% and 109.3% year-on-year, respectively, according to official data. Based on the current growth rate, full-year exports are expected to climb to 8 million units this year. In the long term, global revenue from 3D printer sales is expected to hit $115 billion by 2034, growing by an average 18% annually over the next 10 years. Notably, the consumer market is expected to be the key growth driver.

Growth in China's consumer 3D printer market currently follows two main tracks. Efforts by industry leaders Bambu Lab and Creality (3388.HK) to extend their sales networks, ramp up public outreach, and provide hands-on product demonstrations have greatly raised public awareness and demand for 3D printers. At the same time, “3D printing farms” engaged in mass production are also emerging across China, further expanding application scenarios and demand for consumer-grade products. According to incomplete statistics, China is currently home to several thousand 3D printing farms of varying scale, with a cumulative 300,000 units of equipment installed and growing.

What’s powering the consumer boom?

Three key factors are contributing to the booming demand for consumer-grade 3D printers in China: rapidly falling equipment prices, simplified operation, and a more mature model ecosystem. Among those, falling prices are the most direct driver of the consumer-market boom. Mainstream 3D printing products currently sell for thousands of yuan, or typically less than $1,000, making them cheaper than many smartphones.

As prices have dropped, lower usage barriers have further stimulated the market. In the past, users generally found 3D printing complex and intimidating. Today, deep integration of AI has solved problems involving difficulty of use. Advanced features such as auto- leveling, real-time monitoring, and automated troubleshooting have transformed devices from professional tools requiring manual calibration into plug-and-play smart terminals, effectively removing operational obstacles for ordinary users.

An increasingly rich resource ecosystem has also played a pivotal role in retaining users. The MakerWorld online model community,  operated by Bambu Lab, is a case in point. Using heavy subsidies to incentivize creators, the platform has accumulated nearly 3 million model resources and a large base of active creators, with model resources growing at a rate of nearly 100,000 per month. Meanwhile, Creality, is working to incorporate AI-assisted modeling tools that empower users to create their own designs.

Industrial 3D: The patient giant

Compared with the consumer side’s explosive growth, industrial-grade expansion has been more muted. Global shipments of industrial-grade 3D printing equipment rose by 18% year-over-year in the first quarter of 2026, while the Chinese market increased by 29% – both figures far below the consumer-grade equipment sector. But from an application value perspective, the industrial-grade market still commands a larger base and has already started large-scale deployment in certain high-value scenarios.

Indeed, 3D printing offers numerous unique and irreplaceable advantages in modern industrial manufacturing. These include the ability to vastly shorten R&D and delivery cycles; support small-batch customized production without the need for mold tooling; and offer an exceptionally high degree of design freedom, enabling the fabrication of complex structures difficult or impossible to achieve through traditional manufacturing.

As a result, this emerging manufacturing technology is rapidly proliferating across various industries. In the high-tech aerospace sector, critical components such as rocket engine thrust chambers and nozzles are now being mass-produced using 3D printing. Moreover, companies like Farsoon Technologies (688433.SH) are leveraging the unique advantages of 3D printing to helping aerospace firms overcome technical challenges in manufacturing lightweight, complex, and irregularly shaped structural components. Similarly, precision structural parts, including hinges for foldable smartphones, titanium alloy frames for smartwatches, and drone components, as well as liquid-cooling heat dissipation parts in the low-altitude economy sector, are all accelerating their adoption of 3D printing processes.

China's 3D printing industry has generally established a competitive advantage in the consumer-grade segment, underpinned by supply chain cost efficiencies, intelligent integration, and a thriving content ecosystem. Nevertheless, those scale advantages on the consumer side won’t automatically translate to the same dominant position on the industrial side. While industrial-grade 3D printing has achieved breakthroughs in high-value scenarios such as aerospace, healthcare, and precision manufacturing, further in-depth R&D and refinement are still required in certain areas to achieve a transition to next-generation manufacturing powered by 3D printing.

CLS Marketwatch provides insights and analysis on China’s industries. You can contact the author at liujingyi@cls.cn

To subscribe to Bamboo Works weekly free newsletter, click here

The performance gap between leading Chinese and U.S. AI models has narrowed to less than 3%, yet U.S. private AI investment is still 20 times higher than China’s

By Lee Shih Ta

Judging solely by model rankings, the AI race between China and the U.S. appears to be close to parity. Since DeepSeek burst onto the scene last year, Chinese models including Alibaba’s Qwen have continued to improve, and U.S. companies no longer enjoy the overwhelming lead they held just a few years ago. Stanford University’s 2026 AI Index shows that as of March, the performance gap between leading U.S. and Chinese models had narrowed to just 2.7%, with models from the two countries taking the lead in different rankings.

But the capital picture tells a very different story.

The same report shows that U.S. private AI investment reached $285.9 billion in 2025, compared with just $12.4 billion in China, a difference of 23 times. In other words, even as the performance gap between the two countries’ top models fell below 3%, the gap in private capital spending behind them remained huge.

Chinese AI startups raised more than 110 billion yuan ($16.2 billion) in the first quarter of this year, up 185% year on year. But PitchBook data shows global AI venture funding reached $255.5 billion over the same period, with U.S.-based OpenAI, Anthropic and xAI accounting for 67.3%, or about $172 billion. Those three companies alone raised more than 10 times as much as all Chinese AI startups combined during the period. While the two datasets use different methodologies, they still point to the heavy concentration of global AI venture funding in a small group of leading U.S. companies.

The picture is similar when the focus shifts from fundraising to actual infrastructure investment. Goldman Sachs estimated late last year that China’s leading internet companies would invest more than $70 billion in AI-related areas in 2026, equivalent to just 15% to 20% of the amount U.S. hyperscalers were forecast to spend. Capital Group, meanwhile, estimates that capital spending by major U.S. technology companies will rise from $425 billion in 2025 to $791 billion this year, compared with an increase from $57 billion to $118 billion for their Chinese counterparts. That would narrow the gap only slightly, from about 7.5 times to 6.7 times.

Private investment, however, is only part of China’s AI capital picture. In 2025, China launched its 60 billion yuan National Artificial Intelligence Industry Investment Fund. This year, authorities were also reportedly considering a plan to invest around 2 trillion yuan over five years in a nationwide network of AI data centers, though details of the proposal are still a work in progress.

The U.S. government is also increasing its AI spending. Brookings data on federal AI contracts through March this year show that about $7.2 billion in funding had been allocated. The figures aren’t directly comparable, but they still point to a clear difference in the structure of capital spending: U.S. AI expansion is more heavily driven by technology companies and private capital, while the government and state-owned enterprises play a more prominent role in China.

Can state capital bring in private investment?

If China already has large amounts of state capital flowing into AI, why has private investment failed to expand at anything close to the scale seen in the U.S.? The answer starts with commercial returns. Amazon, Microsoft and Google all have enormous global cloud businesses. Adding data centers and GPUs is not simply a bet on the future of AI; those investments can also be monetized directly by selling computing power, models and software services to companies around the world.

China’s cloud market, by comparison, remains much more domestically focused, while intense price competition among large language models has put pressure on monetization. Technological capabilities may be improving rapidly, but that progress does not necessarily translate into revenue and cash flow at the same pace. Goldman Sachs notes that Chinese hyperscalers still derive 90% to 95% of their revenue domestically, while paid usage of leading Chinese chatbots remains limited.

Profit pressure at China’s largest technology companies also constrains their ability to invest. Alibaba’s (BABA.US; 9988.HK) capital expenditure jumped 75% year on year to 67.68 billion yuan in the April-June quarter, while its net profit fell 75%. The company subsequently raised about $10.2 billion through a share placement to support further AI development. Baidu’s (BIDU.US; 9888.HK) second-quarter revenue fell 4% year on year, while its net profit dropped to 2.3 billion yuan. For such companies, massive AI spending translates more directly into pressure on profits, cash flow and financing needs.

State capital can take on longer-term projects such as data center and power grid construction, domestic chip development and fundamental research. But it primarily addresses the question of who will provide the money. It cannot guarantee that customers will be willing to pay for AI services. If demand and profits fail to grow alongside investment, policy-driven supply could result in low utilization or inefficient capital allocation.

The real test, therefore, is whether state investment can be converted into commercial demand and profits, which in turn could attract more private capital.

That demand gap can also be seen in the revenue mix of Chinese AI infrastructure suppliers. High-speed optical module supplier InnoLight (3308.HK; 300308.SZ) saw revenue from overseas markets surge 209.9% to 39.62 billion yuan in the first half of this year, while revenue from Mainland China rose just 7.7% to 2.16 billion yuan. Overseas markets now account for nearly 95% of its total revenue.

Sitting at the heart of the AI data-center investment boom, InnoLight’s results offer a useful barometer for global AI capital spending. The company’s interim report noted that Alibaba, Tencent (0700.HK) and Baidu spent a combined 64.7 billion yuan in capital expenditure in this year’s first quarter, up 18% year-on-year. By comparison, Microsoft, Amazon, Meta and Google recorded combined capital expenditure of $164.9 billion in the second quarter, up about 86%. A leading Chinese AI infrastructure supplier is therefore seeing most of its incremental demand come from overseas, underscoring how Chinese companies have become important suppliers to the global AI infrastructure boom even as demand growth in their home market remains considerably weaker.

Yet lower private investment does not mean China’s AI capabilities must lag by the same proportion. Restrictions on advanced chips complicate the comparison further. Even if Chinese companies were prepared to spend the same amount as their U.S. counterparts, they would not necessarily be able to obtain the same computing power. That has forced them to improve model efficiency, lower inference costs and accelerate adaptation to domestically produced chips.

That helps explain, at least partly, how private investment in China can be so much lower even as the performance gap between leading Chinese and U.S. models is just 2.7%.

The big question is whether greater efficiency can continue to offset the large capital gap. AI is moving toward agents, multimodal systems and robotics, while large-scale inference will require more chips, electricity and data centers. If U.S. companies continue to outspend their Chinese counterparts by hundreds of billions of dollars each year, their greater computing resources and capacity for experimentation could eventually translate back into a wider technological advantage.

America’s spending spree also carries the risk of overinvestment. If massive AI infrastructure spending ultimately fails to generate sufficient revenue, today’s capital advantage could turn into tomorrow’s depreciation burden, debt and idle capacity. For China, how effectively state capital can turn computing power, chips and infrastructure into market demand, corporate profits and sustained private investment will determine how far its model can go.

China has already shown that lower levels of private capital aren’t preventing it from rapidly closing the technological gap in AI. But as AI enters its next phase, with greater dependence on computing power, electricity and large-scale inference, it remains unclear how long efficiency gains can continue to offset the capital gap. As the model gap narrows, the ability to turn capital into demand, profits and reinvestment could become a new dividing line in the U.S.-China AI race.

Lee Shih Ta is an editor at Bamboo Works.

You can contact him at shihtalee@thebambooworks.com

To subscribe to Bamboo Works weekly free newsletter, click here

China’s AI office race is heating up as tech giants and software vendors push agents beyond chatbots toward execution of workplace tasks as ‘digital employees’

By CLS Marketwatch

The movement of AI technology beyond general-purpose conversation to in-depth specialized scenarios has gained momentum this year. Among the trend’s many focus areas, the workplace, characterized by high frequency usage and ability to perform essential tasks at affordable prices, has been a primary front for AI agent commercialization. A growing array of office-focused AI agentic products, each attracting tens of millions of monthly visits, has quickly emerged, attesting to genuine demand for AI-powered productivity-enhancing tools.

Who will own your digital desk?

China’s workplace-oriented AI assistant sector began undergoing a critical shift starting in the second quarter of this year. The previous model, where major internet enterprises pursued multiple product lines in parallel and tested different technological paths separately, is being replaced by a more consolidated tactic. Several dominant tech giants, including Tencent (0700.HK), ByteDance, Baidu (BIDU.US; 9888.HK), and Alibaba (BABA.US; 9988.HK), have integrated their AI agents, large language models (LLMs) and AI chatbot businesses with their office-focused AI assistant businesses. Meanwhile, office software provider Kingsoft Office (688111.SH) launched two workplace-oriented AI agents in mid-July, covering both individual and organizational use cases.

The flurry of business restructuring and product launches marks the entry of the office-focused AI agent industry into a phase of competing for users through the desktop as a primary gateway. There are currently four main types of market participants: big tech giants, leveraging their proprietary LLMs, cloud infrastructure and productivity suites to secure a foothold in the infrastructure of large corporations; traditional office software vendors, which are building on their existing user bases to embed AI agents into processing workflows for individuals and smaller firms; native LLM providers, who focus on delivering underlying frameworks; and specialized tool providers, which concentrate on specific point solutions, primarily catering to niche needs.

Vision vs reality

China’s office-focused AI agent market has already cultivated considerable users. As of June, aggregate monthly visits to 17 mainstream desktop-based AI-native office-oriented agent platforms in China exceeded 60 million. Among them, Tencent’s WorkBuddy recorded 20.97 million monthly visits, maintaining a commanding lead. This shows that user habits for AI-powered office tools are gaining rapid adoption and frontrunners are beginning to emerge, as the market continues to grow.

In terms of demand, the consumer and business markets are showing distinct differences. Consumer users tend to have shorter decision-making processes and fewer concerns about system integration and related issues. With the growing penetration of AI, an increasing number of individual users are experimenting with AI-backed office solutions through easy-to-adopt scenarios such as PowerPoint creation, drafting meeting minutes, and document processing, and have already demonstrated a clear willingness to pay. By comparison, business clients usually have longer decision-making cycles due to concerns over data security, efficiency improvement and iteration and other factors. Yet once the business market takes off, it promises to become a long-term, stable, and sustainable revenue source.

But multiple bottlenecks are still constraining AI-powered office work. One is a lack of reliability in executing complex tasks. While current AI agents excel at text generation, they struggle to perform in-depth operations across different business software products and varying corporate workflows. At the same time, widespread data silos within individual enterprises and token consumption costs are difficult to manage, creating considerable cost pressures and implicit data leakage risks. Collectively, these issues make businesses reluctant to more actively integrate AI-backed office solutions into their daily operations.

Growing wave

In the upstream computing power supply chain, a demand-pull effect for workplace AI has already begun to materialize. Kingsoft Office has said its profit rose 210% to 264% in the first half of 2026, citing the positive contribution of AI office market demand to its performance; Kingdee International Software (0268.HK) said it returned to profitability in the first half of the year, benefiting from efficiency gains driven by AI; and Yonyou Network Technology (600588.SH) reported that AI-related contract signings reached 910 million yuan ($135 million) in the first half of the year, reflecting high-speed year-over-year growth.

Furthermore, the proliferation of workplace-focused AI assistants will drive a sustained increase in inference workloads. Unlike the widely discussed computing power consumption associated with LLM training, AI-supported workflows primarily involve high-frequency, small- to medium-scale inference, which is unlikely to generate the explosive surge in demand for computing power seen during the training phase but rather will exhibit a steady rise in demand. These different demand structures are giving rise to multiple changes across the industrial chain and pushing business models toward greater diversification.

Looking ahead over the next three to five years, the ability of intelligent agents to complete tasks – that is, achieve closed-loop execution across multiple software systems rather than merely generating text content – will determine whether AI-powered office solutions can evolve from auxiliary tools into digital employees. The maturity of device-cloud synergy technologies and sustainability of business models, including whether on-device computing power can effectively offload cloud-side pressure, whether token costs can be effectively managed, and whether customization costs can keep decreasing, will collectively determine whether the industry can successfully realize its commercial transformation.

The rapid rise of office-focused AI agents mirrors how large-model technology is migrating from general-purpose capabilities to specialized scenarios. Yet despite high expectations for office applications, a bustling competitive landscape does not equate to a mature commercial ecosystem. Fundamental issues such as operational reliability, data security, and cost structures reveal that for these agents to truly assume a core role in productivity, they still need to bridge multiple gaps between technical delivery and commercial viability. That said, there’s no denying that AI-supported office agents have, from their early inception, already demonstrated immense economic and productive potential.

CLS Marketwatch provides insights and analysis on China’s industries. You can contact the author at liujingyi@cls.cn

To subscribe to Bamboo Works weekly free newsletter, click here

China’s scrutiny of offshore insurance income is forcing a rethink of return advantages for Hong Kong policies, which could weaken the appeal of dividend and savings products

By Lee Shih Ta

Hong Kong insurance policies have long been popular with Mainland Chinese, offering higher long-term returns, multi-currency options and access to globally diversified investments beyond traditional protections. But now this cross-border insurance business faces another cost that customers will have to factor into their calculations: personal income tax on offshore insurance income.

The issue surfaced on Aug. 5, when Chinese media reported that tax authorities in Beijing, Hangzhou and other cities had begun reviewing offshore insurance policies held by Chinese residents, including policies issued in Hong Kong. Some policyholders were reportedly required to pay a 20% personal tax on income such as policy dividends and interest earned on prepaid premiums.

Subsequent reports indicated that realized gains from policy surrenders, reductions in coverage and dividend withdrawals could also fall within the tax net. China’s State Taxation Administration responded on Aug. 7 by saying the requirement for Chinese residents to declare and pay tax on overseas income has long been in place, stressing that the enforcement was “not a new policy” and was not specifically targeting Hong Kong’s insurance market.

Nonetheless, insurance stocks wasted no time reacting to the news. The selloff hit London first, where Prudential plc (PRU.L) plunged as much as 13% on Aug. 5 after the original reports. The pressure spread to Hong Kong the following day, with AIA Group (1299.HK) tumbling as much as 8.5% intraday on Aug. 6, and Hong Kong-listed Prudential (2378.HK) also coming under heavy selling pressure. The reaction showed how quickly investors came to view the tax crackdown as a potential threat to demand from Mainland Chinese for Hong Kong insurance.

Those concerns are not without basis. Mainland visitors contributed HK$62.8 billion ($8 billion) in new premiums to Hong Kong in 2024, up 6.5% year-on-year and accounting for about 29% of new premiums. The Hong Kong Insurance Authority subsequently suspended separate publication of Mainland visitor statistics for 2025 while reviewing the reporting methodology for non-local policyholders. Even so, Hong Kong’s overall long-term insurance market continued to expand rapidly, with new premiums surging 50.6% to HK$330.9 billion in 2025.

Cross-border wealth management edge

Much of the appeal of Hong Kong insurance to Mainland customers has come from dividend and savings products. In addition to offering multiple currency options, including the U.S. and Hong Kong dollars, insurers can invest across global markets, giving such policies the potential to offer higher returns than comparable products on the Mainland. UBS estimates that returns for Hong Kong policies are around 6% to 6.5%, compared with about 3% for similar Mainland products. Combined with their foreign-currency exposure, global investment opportunities and wealth-transfer functions, they form a comprehensive cross-border wealth management tool.

But it is precisely this return advantage that a 20% tax would erode first.

Based on the cases disclosed so far in Beijing and Hangzhou, as well as interpretations from industry professionals and tax lawyers, taxable income may extend beyond policy dividends and interest on prepaid premiums to gains realized when policyholders surrender a policy, reduce coverage or withdraw cash dividends. In other words, the cases reported to date suggest that the 20% rate is primarily applied to gains deemed taxable income, rather than simply applying the 20% rate to the policy’s total premiums or cash value. But detailed nationwide rules on how gains from different types of policies should be defined and calculated have yet to emerge.

There is a clear tax rationale behind the move. Chinese residents are subject to tax on their worldwide income, meaning overseas interest, dividends and certain investment gains have long been subject to reporting requirements. As the Common Reporting Standard (CRS) has increased transparency around cross-border financial assets, rules on offshore income that were once difficult to enforce have become easier for tax authorities to administer. In the first five months of this year, taxpayers who received overseas income paid about 13 billion yuan in back taxes.

That means the impact can vary significantly by policy type. Pure protection products with no savings component — such as policies primarily covering death, medical expenses or critical illness — should be less affected. Dividend and savings policies are more exposed, particularly when policyholders surrender policies, reduce coverage or withdraw dividends, turning accrued gains into realized income. For returns that remain accumulated within a policy and have not yet been withdrawn, publicly available information does not yet point to a uniform nationwide method of taxation. It would therefore be misleading to simply cut the expected returns on Hong Kong policies by 20%.

How much of the return gap will remain?

Goldman Sachs believes a key question is how much of the return gap between Hong Kong policies and comparable Mainland products will remain after taxes are paid. As long as a meaningful gap persists, long-term growth drivers for business from Mainland visitors may not be fundamentally affected. UBS holds a similar view. It notes that China’s current individual income tax rules still lack detailed guidance on how insurance-policy returns should be taxed, while returns on domestic Mainland policies are not generally taxed in practice. Even if taxation reduces the relative appeal of Hong Kong products, their higher returns, multi-currency features and broader investment universe should remain competitive advantages.

More important is whether tax authority scrutiny now seen in individual Chinese cities develops into broader enforcement. If such enforcement gradually becomes routine, the tax cost of offshore policies will become a standard part of the purchase calculus for Mainland customers.

In the short term, taxation alone may not be enough to reverse the flow of Mainland customers buying insurance in Hong Kong. But if taxation of offshore policies becomes routine across China, dividend and savings products marketed primarily on their higher returns will face growing pressure. The real test is whether Hong Kong products can retain a sufficiently large return advantage over their Mainland counterparts after tax costs are taken into account.

But returns are not the only source of Hong Kong insurance’s competitive appeal. Multi-currency options, global asset allocation, insurance protection and wealth-transfer functions should continue to support cross-border demand. The current tax crackdown may therefore not immediately change the broader trend of Mainland customers buying policies in Hong Kong. But if enforcement continues to widen, the challenge for investors in insurance stocks will shift from a short-term share-price shock to more fundamental adjustments to assessments of new business growth, product mix and customer demand.

Lee Shih Ta is an editor at Bamboo Works.

You can contact him at shihtalee@thebambooworks.com

To subscribe to Bamboo Works weekly free newsletter, click here

Text-to-video has widened the scope of AI-generated content, while world models are pushing the technology into gaming, robotics and manufacturing

By CLS Marketwatch

AI has achieved many astonishing leaps in content generation over the last few years, across images, videos and audio, blurring the boundary between the digital and real worlds. This trajectory includes the fast-emerging field leveraging AI to generate digital content (AIGC). Yet, as visuals grow sufficiently lifelike and audio increasingly natural, a pivotal question arises: where should AI head next? As the industry diversifies and iterates, an increasingly clear answer is emerging: AI needs to move beyond just "generating materials" toward "understanding the world."

The next frontier: World model

If the past two years have seen AIGC master content “imitation,” then the core challenge for the next phase is enabling AI to step beyond the screen and acquire the ability to interact in real-time and solve practical problems within real-world scenarios.

Here, versatility of the underlying technology infrastructure lays the groundwork for cross-domain transfer, as computing clusters, massive-scale cleansed datasets, and model training capabilities required for training video models are highly compatible with the technical underpinnings of world models. Nearly all leading text-to-video developers are venturing into world model development. And marginal costs remain manageable while value remains significant, since substantial infrastructure investments already in place can be easily redeployed.

In contrast, gains from text-based large language models are approaching their ceiling, while video and game generation models remain firmly on a clear growth track, which offers a well-defined return profile for investment in world model technologies. In view of this, industry players agree that the next phase calls for AI to be able to "understand the world".

From pixels to physics-grounded authenticity

With this goal in mind, the world model has achieved a systematic leap in capabilities compared to traditional AIGC tools. Most importantly, the world model’s training data is largely derived from interaction data in the physical real world, rather than simply video footage. This enables the model to internalize fundamental physical principles such as gravity, inertia, and the attenuation of light and shadow. In highly dynamic and complex scenarios, such as explosions, particle systems and fluid effects, the visuals generated by world models exhibit greater physical coherence, fundamentally reducing the incidence of visual anomalies. This not only elevates visual quality but also underscores that generated content possesses an “authentic” foundation that can be verified against the physical world.

Furthermore, traditional video generation tools require users to wait for the final output to be produced after imputing a prompt, which follows a linear, one-time generation process where users passively receive the output. In contrast, world models support full-cycle real-time feedback, adjustment and rendering. Users can intervene and optimize throughout the entire generation process without waiting, thus advancing the creative workflow from an iterative loop of "generate-modify-regenerate" to a continuous interactive stream.

We can take Shengshu AI’s product portfolio as an example. Its three product lines – the Vidu S1 real-time interactive model, Vidu generative world model, and Motubrain world action model for embodied intelligent robots – correspond to the progressive capabilities of world models, from understanding to interaction to action, forming a complete technological ecosystem that spans both digital content and physical entities.

However, the commercial value of world models ultimately needs to be validated in specific real-life scenarios, with the gaming industry serving as a primary application arena. As one of the most advanced forms of digital content, games inherently demand physical simulation, real-time interaction, and three-dimensional spatial understanding, which align closely with what world models can offer.

Gaming heritage powers physical intelligence

Notably, technical capabilities accumulated from the gaming realm are now being reverse-engineered to empower the physical world. For instance, NetEase’s (NTES.US; 9999.HK) SmartEase embodied intelligence brand has migrated 3D virtual technologies and underlying technical frameworks originally developed for the games “Sword of Justice” and “Naraka: Bladepoint” into construction machinery, enabling the pre-training and troubleshooting of models for equipment like excavators and loaders within virtual 3D environments. These AI capabilities, cultivated through game R&D, have been applied to the construction of intelligent systems for construction machinery, producing intelligent excavators and loaders.

At the same time, the fusion of embodied intelligence and physical AI is improving. Examples include Giga AI, which has developed a full-stack world model product matrix covering a wide range of application scenarios, from content creation and driving simulation to embodied intelligence. Mature world models, equipped with general-purpose understanding capabilities, are expected to enable future robots to observe environments and anticipate actions much like humans do. Rather than requiring task-specific training, these robots will rely on vast stores of world knowledge to autonomously execute complex operations.

AIGC is undoubtedly at a critical turning point of evolving from a “producer of digital content” to an “interpreter of the physical world.” As the core technological foundation of this paradigm shift, world models are redefining the way AI understands and generates materials, empowering machines with the ability to learn about space, anticipate motion, interact in real time, and execute complex tasks just as humans do. Applications have already taken off in fields such as gaming, film and TV drama, embodied intelligence and industrial manufacturing. At this stage, AI’s development mirrors the early days of mobile internet, with fusion across a wide range of industries only just beginning.

CLS Marketwatch provides insights and analysis on China’s industries. You can contact the author at liujingyi@cls.cn

To subscribe to Bamboo Works weekly free newsletter, click here

At least 25 Hong Kong-listed companies could switch to China’s accounting standards this year, as a boom of dual listings in both markets raises concerns over financial comparability

By Lee Shih Ta

Hong Kong’s role in connecting Chinese companies with global capital rests on more than Hong Kong dollar trading and access to offshore financing. It also relies on disclosure, auditing and financial reporting systems familiar to international investors. For Chinese companies, listing in Hong Kong is, in a sense, an exercise in translating their businesses into a financial language that can be compared with global peers. But as more Chinese companies standardize their reporting under China Accounting Standards for Business Enterprises (CASBE), Hong Kong’s role as that financial reporting “translator” is beginning to change.

A review of Hong Kong Stock Exchange filings shows that, as of July 26, at least 25 Hong Kong-listed companies had announced or proposed switching to CASBE, far more than the two recorded over the same period last year and the low single-digits two years earlier. The companies span semiconductors, electronics manufacturing, construction machinery, agriculture and livestock, healthcare and infrastructure. Most are China-incorporated companies with dual listings on one of China’s domestic markets in Shanghai and Shenzhen, and a second listing in Hong Kong, often called A+H companies. Many only recently completed their Hong Kong listings.

For such companies, International Financial Reporting Standards (IFRS) used in Hong Kong are often an extra layer added during the Hong Kong listing process rather than the foundation of their day-to-day financial systems. Victory Giant (300476.SZ; 2476.HK), which listed in April, said it had always prepared its financial statements under CASBE and only engaged a Hong Kong accounting firm to prepare information under IFRS for its Hong Kong offering. Less than two months after listing on April 21, it announced that it would stop preparing a separate set of IFRS financial statements and proposed appointing a Mainland accounting firm to audit both its Shenzhen and Hong Kong financial statements, citing faster disclosure, simpler procedures and lower audit fees.

In fact, the Hong Kong Stock Exchange has allowed Mainland-incorporated issuers to use CASBE since 2010 and permits approved Mainland accounting firms to audit their Hong Kong-listed financial statements. The real question, then, is why so many companies are suddenly choosing this long-established route.

A direct driver is the changing mix of Chinese companies listing in Hong Kong. Some 24 companies completed A+H listings in the first half of 2026, already surpassing the 19 recorded for all of 2025. These companies already had complete CASBE-based accounting and audit systems in place. They often wait until preparing their first financial report or reappointing an auditor after their Hong Kong listings before deciding whether to continue maintaining two reporting systems. This year’s switching wave may therefore be viewed as a lagging consequence of the current A+H listing boom.

Looking ahead, tighter scrutiny of red-chip listing structures, involving Chinese companies incorporated in offshore locations, could further amplify the trend. If some companies planning Hong Kong listings shift from offshore structures to issuing H shares directly through Mainland-incorporated entities, the proportion of Mainland-incorporated issuers will rise, potentially increasing the number of Hong Kong-listed companies using CASBE.

Changes in the market’s funding mix may also reduce the appeal of maintaining IFRS financial statements. Average daily turnover through the Southbound Stock Connect, which allows Mainland investors to trade Hong Kong-listed stocks, rose from HK$48.2 billion ($71.2 billion) in 2024 to HK$121.1 billion in 2025, while Southbound trading accounted for about 23% of Hong Kong cash equities turnover at the end of the fourth quarter. Mainland investors are already familiar with CASBE. For companies whose businesses and investor bases are concentrated mainly in Mainland China, the incremental benefit of maintaining a second set of financial statements under IFRS may therefore be diminishing.

In the broader policy context, the switching wave may also serve as one indicator of how China’s financial system is evolving. Beijing has set the goal of building China into a financial powerhouse, while emphasizing stronger domestic financial infrastructure, professional services and cross-border financing capabilities. The ability of more Mainland companies to retain their China-based accounting and audit systems while still tapping offshore capital through Hong Kong is consistent with that direction.

New earnings yardstick

Companies are switching mainly to reduce the cost of maintaining two sets of financial statements. But for investors, the two standards cannot be treated as interchangeable. Although CASBE and IFRS standards have converged in many areas, differences remain in places such as biological assets and fair value measurement, and can be large enough to materially alter reported earnings. The size of those differences also depends on a company’s business model and asset structure.

Hog farming company Dekon Food (2419.HK) offers a clear example. Its net profit for 2025 was 1.42 billion yuan ($210 million) under CASBE, but only 531.7 million yuan under IFRS, a difference of 889.9 million yuan. The direction was reversed in 2024, when its profit under CASBE was 932.8 million yuan lower than that under IFRS. The company attributed the differences mainly to the fair value measurement of biological assets and related accounting treatments.

The case shows that CASBE does not necessarily inflate profits, nor is IFRS invariably more conservative. For the same company, switching accounting standards can produce a difference of nearly 1 billion yuan in annual profit. Directly comparing figures under the new reporting basis with historical IFRS earnings could lead investors to misjudge the pace of growth, while using unadjusted net profit to calculate a price-to-earnings (P/E) ratio could produce an entirely different valuation.

The impact of switching standards is not determined by industry alone, but also by a company’s asset structure and the nature of its transactions. Where the main areas of difference are not material, profit, earnings per share and return on equity may change little. But for accounting involving biological assets, major impairments, business combinations or fair value measurement, differences between the standards can be large enough to alter core financial indicators.

The switching wave reflects a growing number of Mainland companies making use of a long-standing option: retaining their existing accounting and audit systems while accessing Hong Kong’s financing and trading platform. For Hong Kong, this may mean its role is evolving from simply “translating” Mainland companies into an international financial language, to accommodating parallel systems that remain different but continue to converge. For investors, the real issue is whether a change in reporting basis weakens the comparability of historical data and raises the cost of understanding, adjusting and verifying earnings.

Lee Shih Ta is an editor at Bamboo Works.

You can contact him at shihtalee@thebambooworks.com

To subscribe to Bamboo Works weekly free newsletter, click here

AI-driven memory demand is raising costs and squeezing smartphone and PC makers, while creating opportunities for Chinese chipmakers

By CLS Marketwatch

The global memory chip market is witnessing a profound transformation this year sparked by the rising tide of AI. Since 2022, the booming AI market has prompted the world’s three largest DRAM manufacturers to shift a massive portion of their capacity toward the AI sector. At the same time, traditional DRAM product prices continue to climb, creating cost pressures throughout the supply chain – from upstream wafers all the way down to the consumer market. That, in turn, has led to a wave of price hikes across the entire consumer electronics industry in China, for everything from smartphones to laptops.

The AI-hungry beast

Since 2022, the global DRAM market has undergone a supply-side transformation: AI technologies represented by large models have entered a period of explosive growth, driving a sharp surge in demand for high-performance memory chips. As such, the world's three major manufacturers – Samsung (005930.KS), SK Hynix (000660.KS; SKHY.US), and Micron (MU.US) – have strategically shifted their focus toward AI servers and enterprise-grade storage.

By 2026, enterprise-grade memory chips needed for AI servers took up about 50% of DRAM wafer production capacity, and this figure is expected to rise further to around 60% in 2027. As AI server manufacturers have locked in substantial wafer capacity in advance through long-term contracts, the consumer electronics sector continues to bear the brunt of limited capacity.

The hardest hit by this shift are mature products such as DDR4. The three major manufacturers plan to phase out their DDR4 production lines, yet current mid- to low-end smartphone processors still rely primarily on DDR4 and lower specifications. As existing capacity continues to shrink and new supply fails to materialize, the supply-demand gap will keep widening. Moreover, prices for traditional DRAM have risen about 700% since 2022.

Upstream booms while downstream bleeds

Persistent increases in memory chip prices have triggered a dramatic redistribution of profits along the supply chain – terminal players are under severe strain, while upstream manufacturers are enjoying record profits.

The smartphone market has borne the brunt of supply shortages. Taking phones costing $1,000 or less as an example, the share of memory components as a proportion of total costs has soared from 10% to 15% of the total to 40% to 50%, and has even approached nearly two-thirds of total costs in certain models, leaving virtually no profit margin. In response, many low-end smartphone manufacturers have been forced to downgrade specifications, and some have even resorted to use of secondhand chips to maintain production.

In this context, many manufacturers have found themselves unable to continue production, as evidenced by the shutdowns of some smaller ODMs. Meizu (001229.SZ), a long-established Chinese smartphone manufacturer, suspended its smartphone business in March. At the same time, Oppo, Xiaomi (1810.HK) and other Android brands have opted for collective price hikes of 300 yuan to 500 yuan ($44 to $74), while some manufacturers have even hiked prices a second time. Price increases for new models in the second half of the year are expected to be even higher, with some models jumping as much as 1,000 yuan.

The PC and tablet sectors haven’t been spared either. By the third quarter of 2026, the share of memory components as a percentage of total PC costs had climbed to 40% to 60%. Brands including Lenovo (0992.HK), Asus (2357.HK), HP (HPQ.US) and Dell (DELL.US) have all adjusted their prices over the past year. As prices soar, global laptop shipments in 2026 are expected to decline by about 8% year-over-year.

In sharp contrast, upstream players are enjoying a golden age of profitability. Global memory giant Micron’s revenue for the third quarter of its fiscal year surged 346% year-over-year, with net profit skyrocketing more than 13-fold. Chinese memory leader ChangXin Memory anticipates its net profit in the first half of 2026 reached 50 billion yuan to 57 billion yuan, a more than 22-fold increase from the same period last year.

A tale of challenges and opportunities

Looking ahead, the memory shortage for consumer electronics is unlikely to change in the near term, though the pace of price hikes may gradually moderate. DDR5 16GB is expected to see quarter-on-quarter growth of about 19.2% in the third quarter of 2026, with a further increase of around 10% in the fourth quarter. LPDDR is also expected to climb about 20% quarter-on-quarter in the third quarter.

DRAM supply growth is poised to accelerate over the medium to long term as newly built wafer fabs come online over the next two to three years. And once the demand peak driven by AI infrastructure subsides, DRAM supply and demand are likely to become more balanced. But consumer electronics will have to endure the dual challenges of elevated costs and subdued demand during this nearly three-year supply shortage transition.

Beyond the memory question, localization presents another noteworthy factor in the current climate. Chinese memory chip makers, represented by ChangXin and Yangtze Memory Technologies, are rapidly gaining share in the PC, smartphone and consumer SSD markets, and are now meeting demand for mid- to high-end smartphones, despite tight supply. The nearly three-year supply shortfall presents a critical window of opportunity for China's domestic memory chip makers to fill the gap. Yet in the premium HBM and AI enterprise-grade memory sectors, domestic alternatives are still struggling to achieve large-scale substitution.

Overall, as demand for AI computing power explodes, the reallocation of memory chips by the three major manufacturers has led to supply shortages and escalating costs for consumer electronics makers. Products such as smartphones and PCs are all facing severe challenges in supply, costs and profitability, while industry-wide tensions continue to mount. The consumer electronics industry will remain under pressure until new capacity comes onstream. In this environment, Chinese memory chip manufacturers can seize this prime opportunity to accelerate their penetration into the memory chip supply chain.

CLS Marketwatch provides insights and analysis on China’s industries. You can contact the author at liujingyi@cls.cn

To subscribe to Bamboo Works weekly free newsletter, click here

China’s drive to stamp out cutthroat competition is starting to stabilize express delivery prices, but uneven earnings raise questions about the recovery’s sustainability

By Lee Shih Ta

Delivering more parcels while earning less from each was once the clearest illustration of the cutthroat competition – known locally as “involution” – plaguing China’s express delivery sector.

China’s express delivery parcel volume rose 13.6% in 2025 to 198.95 billion, while average revenue per parcel fell 6.3% year-on-year to about 7.51 yuan ($1.11). That trend has begun to change this year. Data from the State Post Bureau showed parcel volume rose 5.2% in the first five months of 2026 to 82.87 billion, while revenue increased 7.2% to 635.37 billion yuan. That implies average revenue per parcel edged up about 1.9% to 7.67 yuan. In May alone, revenue grew 9.5%, easily outpacing a 5.7% increase in parcel volume and implying a roughly 3.6% rise in revenue per parcel.

That doesn’t mean express delivery companies have raised prices across the board. The national average is also affected by international shipments, returns, individual customer parcels and changes in product mix. Still, revenue growth once again exceeding parcel-volume growth suggests the industry’s yearslong price decline is beginning to ease.

Last July, the State Post Bureau voiced its opposition to involution-style competition and convened a meeting with express delivery companies, calling for the lawful regulation of destructive low-price competition. Around the same time, the city of Yiwu in Zhejiang province raised its regional minimum price from 1.1 yuan to 1.2 yuan per parcel. Some cities in Guangdong later required e-commerce parcels weighing 0.1 kilograms to be priced at 1.4 yuan or more. While these were regional measures, they provided an important signal in two of China’s largest express delivery hubs.

In January, the State Post Bureau went further by making a crackdown on involution-style competition one of its priorities for the year. It called for stronger oversight at the source, a better balance in the distribution of benefits between company headquarters, their franchisees and their couriers, and the use of transparent regulation to address unreasonable practices such as unequal delivery fees and automatic penalties triggered by customer complaints. This shows the campaign is expanding beyond curbing low prices to cover profit-sharing and operating practices across franchise networks.

In practice, the burden of price wars is often ultimately borne by franchisees and last-mile deliverers. Company headquarters can lower costs through automation, higher parcel volumes and more efficient sorting and transportation, but smaller players cannot reduce rent, labor and delivery expenses at the same pace. If headquarters see their revenue improve while franchisees still rely on cutting delivery fees paid to couriers or imposing heavier fines to stay afloat, the industry cannot truly escape involution.

Easing price pressure

In the first five months of this year, YTO’s (6123.HK; 600233.SH) cumulative revenue per parcel still edged down about 0.8% year-on-year, though that was a big improvement from the 6.2% drop in the same period last year. Revenue per parcel at STO Express (002468.SZ) and Yunda (002120.SZ) rose about 13.4% and 9.4%, respectively, over that time. But STO’s figure was affected by the consolidation of Danniao Logistics into its results, and an increase in higher-priced reverse-logistics business. Yunda’s parcel volume fell about 4.4% over the same period, producing a clearer pattern of lower volume but higher prices.

At ZTO (2057.HK; ZTO.US), the industry leader by parcel volume, first-quarter revenue per parcel from its core express delivery business rose 8.2%, partly due to bigger contributions from its key-account and return parcel businesses. The associated pickup and delivery costs also increased. These figures suggest the yearslong decline in pricing is under control, with some companies improving profit per parcel through changes in product mix, lower costs and recovering margins.

SF Holding’s (6936.HK; 002352.SZ) revenue per parcel also recovered, rising 4.8% year on year to 13.75 yuan for its express logistics business in May. But SF operates mainly through a directly managed network and has a larger proportion of time-definite express and mid- to high-end services. Its revenue per parcel is therefore far higher than that of franchised e-commerce delivery companies, and movements in the figure are more heavily influenced by its product mix and customer demand.

Overall, while parcel pricing has stabilized to varying degrees across the sector, the reasons have differed for each company. Some benefited from an improved product mix, others achieved higher prices alongside falling parcel volumes, while some relied mainly on lower costs to restore margins. This suggests the anti-involution campaign has yet to produce an industrywide profit inflection point, leading investors to wait before revaluing the sector.

Differing company stock performances also point to variable market views on each company’s prospects. STO’s Shenzhen-listed shares have gained about 9.3% so far this year, while YTO and Yunda have risen 4.7% and 1.2%, respectively. ZTO’s Hong Kong shares have gained about 16.9% over the same period, while SF remains in negative territory. YTO’s shares rose by their daily 10% limit on July 1 after the company said its profit rose between 69% and 86% in the first half of this year, showing investors are more focused on whether improved pricing can translate into actual earnings.

The divergence may reflect differences in earnings structure. ZTO relies mainly on China’s franchise-based e-commerce delivery market, meaning a stabilization in parcel pricing and lower unit costs should, in theory, feed more readily into profits and cash flow. SF is larger, but its operations span time-definite express delivery, freight, international logistics and supply-chain services, making the anti-involution campaign only one of many factors affecting its performance. ZTO’s relative share-price strength may therefore reflect a market view that it has a clearer path to improved profits, rather than signaling a broad revaluation of the express delivery industry.

The anti-involution drive can raise the industry’s price floor, but it can’t create competitive barriers for individual companies. As parcel-volume growth slows, companies will still have an incentive to fight each other for existing customers, and price competition could re-emerge through rebates, subsidies and discounts for major clients. China’s express delivery industry will only truly escape the cycle of delivering more while earning less when competition shifts from chasing parcel volume through low prices to differentiating themselves through factors like delivery speed, returns services, supply-chain capabilities and customer mix — and when their franchisees and couriers can also share in the gains.

Lee Shih Ta is an editor at Bamboo Works.

You can contact him at shihtalee@thebambooworks.com

To subscribe to Bamboo Works weekly free newsletter, click here

China’s power sector is opening the door for specialized embodied robots, but scaling up remains a costly, complex climb

By CLS Marketwatch

As a core medium for AI interacting with the physical world, embodied intelligent robots have made critical evolutionary advances over the past few years as industry chains mature. Within that broader category, specialized embodied intelligent robots, designed for high-risk industrial environments such as power, mining, maintenance and emergency response tasks, are grabbing increasing attention. With its vast infrastructure, highly standardized operational scenarios, and long-accumulated experience in use of automation, the power industry is considered a testing ground for mass deployment of specialized embodied intelligent robots.

The great leap of embodied robotics

Backed by concerted policy efforts and industry initiatives, embodied intelligent robots are taking significant strides toward large-scale adoption. Against the backdrop of such intensive initiatives, demand has been further strengthened by the introduction of industry standards and support measures, as well as a concrete target to deploy tens of thousands of units by the end of 2026.

This year alone, customers have placed large orders for embodied intelligence from sectors such as warehousing, logistics and power, with the State Grid Corp. of China’s purchasing plans drawing particular attention. China’s largest power grid operator plans to procure around 8,500 embodied intelligence devices during the year, with a total investment of approximately 6.8 billion yuan ($1 billion). Robot procurement by power grid operators has entered a broader new upcycle, and the market size could reach 10 billion yuan in the near term.

Power leads the way

As specialized embodied intelligent robots are rolled out, the pace of adoption is varying considerably across different fields. The power sector has taken the lead, while others remain largely in earlier stages. Nonetheless, inspired by the power sector’s progress, market expectations for large-scale robot deployment in other industries, such as mining, oil and gas, and rail transit, have risen sharply.

From a medium-term perspective, grid-related investments are expected to exceed 5 trillion yuan in coming years, which will stimulate substantial demand for robots. As investment in new power systems grows, demand is projected to grow further still. However, the power industry demands a high degree of operational precision, where drone technology is already well-developed and better suited. That raises questions about whether embodied intelligent robots can successfully gain a foothold there.

Meanwhile, adoption of embodied intelligent robots in mining, oil and gas, and rail transit is notably lagging. While demand has picked up somewhat, company feedback indicates a near-term surge is unlikely, primarily due to two constraints. First, cost pressures remain challenging, and the economic viability of mass utilization has yet to be fully validated. Second, non-standardized operational scenarios place high demands on robot adaptability, making it difficult for general-purpose products to directly meet complex on-site requirements. Companies are optimistic about the long-term industry outlook, but have yet to translate that to a significant uptick in demand so far.

Partnerships pave the path forward

Notably, general-purpose robotics and specialized robotics manufacturers haven’t competed against each other as much as previously anticipated. Instead, they’ve increasingly shifted toward collaborative R&D and complementary strengths.

Having cultivated extensive expertise in fields like power, mining and nuclear for years, specialized robotics makers have developed mature product lines, including wheeled and tracked robots as well as drones, and accumulated a wealth of scenario-specific insight and operational data. Yet their weakness lies in a lack of production capacity for humanoid robots and relatively limited knowledge in cutting-edge areas such as embodied intelligence algorithms. On the other hand, general-purpose robotics firms possess advanced technologies and scaled delivery capabilities, but lack in-depth understanding of specific operating environments, safety standards, and workflows in specific industries.

In such an environment, joint R&D has emerged as an ideal model. For example, Hangzhou Shenhao Technology (300853.SZ), a leading power sector player, has teamed up with Deep Robotics; mining giant Citic Heavy Industries (601608. SH) is jointly developing an industrial-grade, explosion-proof robotic dog with Unitree Robotics; and in the nuclear industry, Jingye Intelligent Technology (688290.SH) has also formed a strategic partnership with Deep Robotics. Specialized robotics makers undeniably have better understanding of specific application contexts and possess vast amounts of accumulated data that are difficult to replicate in the short term.

Looking at price trends and corporate strategies, the cost of humanoid robots is falling rapidly. Humanoid robot shipments in China totaled 14,400 units last year, but are expected to leap to between 100,000 and 200,000 units in 2026. Unitree is a case in point of how fast prices are falling. The unit price of its humanoid robots dropped from 593,400 yuan in 2023 to 260,700 yuan in 2024, and fell further to 167,600 yuan in the first three quarters of 2025. Operating in a market where pricing has yet to stabilize, most specialized robotics makers are choosing to focus their resources on independent R&D of core components and software algorithms.

Overall, 2026 is shaping up as a starting point for mass adoption of specialized embodied intelligent robots in China. The power sector, with its vast infrastructure, highly standardized operations and sizable procurement plans, has become the first to unlock a billion-dollar market opportunity. While other specialized sectors offer promising long-term potential, they are still constrained by cost pressures and the need to adapt to specific operational scenarios, thus slowing their adoption rate. On the whole, the industry's path from pilot projects to full-scale adoption is still awaiting further validation.

CLS Marketwatch provides insights and analysis on China’s industries. You can contact the author at liujingyi@cls.cn

To subscribe to Bamboo Works weekly free newsletter, click here

Unlike overseas rivals that rely on subscriptions and APIs, Chinese AI vendors are monetizing through cloud platforms, project contracts and compute usage

By LeadLeo Research Institute

As commercialization of large AI models accelerates globally, overseas and Chinese markets have begun to diverge structurally in usage patterns, competitive dynamics and how value is distributed. Overseas markets are built mainly around mature subscription models and direct API payments. China’s market, in contrast, is dominated by enterprise usage, platform-based delivery and free or low-cost customer acquisition. The gap reflects different choices between open-source and closed-source models, and is also shaped by policy, supply concentration and payment culture. The situation is unlikely to reverse in the near term.

Overseas markets start with cases like ChatGPT’s roughly 700 million weekly active users and extend into enterprise APIs and developer subscriptions, creating a full payment ladder from individuals to companies. That gives model developers relatively broad monetization coverage based on usage volume. China’s market is more polarized. Consumer-facing products such as Doubao and Tencent Yuanbao mostly use free models, meaning direct commercial value from individual users is limited. Billable token consumption is concentrated mainly on the enterprise side and is often handled through cloud vendor platforms, meaning much of the actual usage monetization may not flow directly back to the original model developers.

Differences in the value density of use cases are a key reason subscription models have struggled to take shape in China. Overseas mainstream applications are concentrated in high-token knowledge work such as code generation and professional analysis, where a single call can create enough business value to support standardized billing. Large-scale deployments in China, however, are more often efficiency tools such as customer-service Q&A, marketing copywriting and document processing. These tasks usually involve shorter context windows and more fixed output formats, resulting in lower token density and lower spending per task. That leaves subscriptions without enough per-call value to support them.

Open- and closed-source strategies have further reshaped the payment structure. Under a closed-source model, model weights are not accessible, and all usage must pass through authorized billing points. Token consumption can flow back to the original developer, which can also directly accumulate customer data and strengthen its position in renewal negotiations. Under an open-source model, companies can download weights and deploy models themselves. The computing power consumed by private deployments is handled by cloud vendors or enterprises, while the original model developer’s billable touchpoints are largely limited to cloud-hosted inference. Customer relationships accumulate at the cloud platform and systems-integrator layers, making project-based delivery the main payment model.

In China, policy, supply concentration, procurement practices and payment culture have collectively reinforced the project-based model. High-value customers in finance, government and healthcare tend to prefer private deployments because of data-security requirements. The coexistence of several high-quality open-source models has narrowed capability gaps and shifted pricing power toward buyers. Free-use expectations formed during the mobile internet era, combined with aggressive price competition among vendors, have further weakened users’ willingness to pay. Corporate budgets are also commonly approved on a project-by-project basis, making recurring subscription spending a poor fit. Customized projects, by contrast, are often easier to implement.

Overseas markets, in contrast, have moved toward strategic alliances as barriers rise across computing power, cloud infrastructure and model capabilities. Data centers take years to build, Nvidia (NVDA.US) GPUs have made computing power a scarce resource, and AWS, Microsoft Azure and Google Cloud together control more than 60% of the global cloud market. At the same time, the cost of training frontier models continues to climb. No single company can easily achieve full vertical integration, prompting alliances among players such as OpenAI, Microsoft (MSFT.US) and Nvidia. Each controls a different part of the chain — model capabilities, cloud infrastructure, enterprise channels or the GPU ecosystem — creating end-to-end coordination.

Such alliances are also squeezing the resources, channels and pricing for independent platforms. Computing power is allocated first within the alliance, leaving outside platforms at a disadvantage in both cost and timing. Model capabilities are being embedded into cloud platforms, office software and enterprise services, turning them into the default entry point for corporate customers. Bundled pricing for computing power, models and cloud services also makes it difficult for independent platforms to match their cost advantages. As a result, the independent platform is gradually shifting away from basic model access toward higher-value capabilities such as governance, orchestration, security and compliance.

Money in the cloud

For China’s mainstream vendors, model revenue essentially lies in the monetization of infrastructure consumption. Open weights lower the barrier to adoption, bringing more developers and enterprises into the chain of usage, fine-tuning and deployment. As the user base expands, inference, training, storage and network consumption are concentrated in cloud platform resource pools, with most revenue captured by the infrastructure layer. Alibaba Cloud Model Studio, or Bailian, is a typical example. It integrates models such as Qwen, GLM, MiniMax and DeepSeek, and charges based on input and output tokens. Its revenue covers model capabilities, inference computing power, data storage, network access and platform scheduling, rather than model capabilities alone.

Open-source large models can generate revenue. But under the current structure, it is difficult to sustain positive gross margins on model capabilities alone. Instead, APIs, private deployments, commercial licensing and fine-tuning are the main paths to commercialization. But pricing power for hosted inference APIs has been squeezed by subsidies for low-cost supply and cloud vendors. Commercial licensing is not a mainstream revenue source for China’s open-source models. Private deployments offer relatively better margins, but that window is narrowing. Fine-tuning and training services have weaker margins, while customers’ in-house capabilities are improving. As a result, open weights function more as a tool for customer acquisition and ecosystem expansion, while monetization is moving outward to stable usage, dedicated deployments and industry-specific project delivery.

Future competition will shift toward a system-level contest of “model capability x infrastructure capacity.” Model capability will still determine the upper limit for tasks such as complex reasoning, coding and multimodal applications. It will also influence developers’ willingness to try a product and the premium it can command. But the open-source ecosystem will narrow capability gaps in general-purpose tasks. Once enterprises complete their initial vendor selection, renewals and capacity expansion will depend more on service stability, unit call cost, response latency, system integration, compliance and long-term supply assurance. Model capability is the entry ticket, but not necessarily a durable moat.

Infrastructure capacity, by contrast, creates barriers in inference cost, high concurrency, network scheduling, computing power supply, low latency and stability. These barriers require years of development and billions of dollars in capital investment, making it hard for new entrants to catch up quickly. China’s token price war has already spotlighted this trend. As unit prices continue to fall, the focus of competition will shift from model capability to inference costs and scale efficiency. Vendors with self-developed chips and large-scale computing reserves will be better positioned to build structural cost advantages.

LeadLeo Research Institute is an original content platform for research on banks and companies and an innovative digital research service provider with nearly 100 senior analysts. You can contact the platform at CS@leadleo.com

This commentary is the views of the writer and does not necessarily reflect the views of Bamboo Works

To subscribe to Bamboo Works weekly free newsletter, click here

While investors focus on AI chips and servers, modified PPO resins are emerging as another major beneficiary of the AI infrastructure boom

By CLS Marketwatch

Global technology industries are navigating a new revolution led by AI computing. The ongoing emergence of a new generation of cutting-edge AI infrastructure has placed unprecedented requirements on the performance of underlying materials. Against that backdrop, polyphenylene oxide (PPO), an engineering plastic renowned for its exceptional performance, is rising as an indispensable core material for emerging industries such as computing infrastructure and advanced energy. Driven by domestic technological breakthroughs and strong demand for localization, China's PPO materials sector has entered a golden age in the last few years.

Chemistry of a computing revolution

As demand for AI computing hardware explodes, requirements for signal transmission speed and loss have risen sharply. This demand is spreading upstream through the electronics supply chain, from printed circuit boards (PCBs) and copper-clad laminates (CCLs) to the specialty resins used to manufacture them. Since electronic resins account for 20% to 25% of CCL production costs and play a critical role in determining performance, PPO applications have expanded beyond traditional home appliances into emerging sectors centered on AI computing infrastructure and advanced energy.

Notably, the market for high-end modified PPO resins is currently in a state of undersupply, as overseas production capacity has contracted and new domestic capacity has been slow to ramp up. That’s caused overall supply to tighten, leading to rising prices. In particular, the supply-demand gap for modified PPO resins used in M6-M8 grade CCLs has been widening steadily.

For example, PPO production lines at both Guangdong Silver Age Sci & Tech’s (300221.SZ) and Jinan Shengquan Group (605589. SH) are running at full capacity with full order books. Yet actual capacity is constrained by multiple technical bottlenecks. While Silver Age’s original design capacity was 300 tons per year, its actual output is less than that due to complex processes, non-standardized production lines, and a shortage of specialized talent. Meanwhile, Shengquan has been building on its existing capacity of about 1,500 tons per year, with an additional 2,000 tons under construction and expected to come online as early as the fourth quarter of 2026. Nevertheless, the long-standing dominance of the global electronic-grade PPO resin market by Sabic has not fundamentally changed, and Chinese domestic products are still in the early stages of development.

Staggering economics of modified PPO resins

PPO has long been hailed as the “crown jewel of plastics” due to its demanding manufacturing process and extremely high barriers to entry. “Modification” here is the pivotal step in achieving commercial applications for PPO, with the key being to address the shortcomings of insufficient heat resistance and poor adhesion in standard PPOs during PCB processing, which Chinese enterprises are working on. Domestic players have already succeeded in producing mid- to low-end products, and are expected to start taking share within the next few years.

The PPO resin independently developed by Tongyu Advanced Materials (301630. SZ) has been certified by select customers and is currently under commercial promotion. Similarly, following certification of resin-related products by its TYADMT subsidiary, both production and sales have grown rapidly. By 2025, sales reached 28,900 tons, representing a year-over-year increase of more than 6.3 times.

Shengquan can deliver a full range of resin products from M6 to M9, with its M10 resin currently being certified. The high-frequency, high-speed resins it produces notably go to both leading domestic and international CCL manufacturers.

At the same time, the premium effect for high-end products is remarkable. Premium PPO resins tailored for the AI computing sector command prices 20% to 30% higher, and certain specifications can even double the price. Boosted by AI servers and other applications, global demand for PPO in high-speed CCL is projected to reach approximately 8,000 tons in 2026, with prices potentially hitting 800,000 yuan ($118,160) per ton. This demonstrates the immense revenue-generating potential of advanced PPO products, which in turn points to a promising industry outlook and substantial gross margins.

A patient climb toward market acceptance

Despite technological strides among Chinese manufacturers, substitution in the premium market still presents major challenges. Electronic-grade PPO typically requires triple certification from CCL and PCB manufacturers, as well as end-users, and this lengthy verification cycle can pose a critical barrier to new entrants.

Modified PPO resins used in advanced CCLs are still largely imported for now, primarily due to reliance on established supply chains and proven trust. Since downstream customers in high-precision and cutting-edge fields are intensely sensitive to trial-and-error, domestically produced high-end materials require a longer validation period to build credibility. However, as domestic materials gain ground in the mid- to high-end segment, their cost-effectiveness and service advantages will gradually alter this landscape.

Looking ahead, the modified PPO resin industry has strong growth momentum and features multiple technological pathways developing in parallel. In addition to PPO itself, hydrocarbon resins have also become an important area for development. Yet in M9 and higher-grade high-speed CCL or substrate systems, PPO/OPE resins continue to serve as the primary material, with performance and processability balanced through blending with other hydrocarbon resins. In this regard, Sichuan EM Technology (601208. SH) has already achieved stable mass production of M9-grade hydrocarbon resins; Shengquan, meanwhile, is planning a 1,000-ton-per-year hydrocarbon resin production project, which is expected to be completed and gradually put into operation by the fourth quarter of 2026.

Overall, the modified PPO market has a promising outlook due to a widening supply-demand gap and rising prices. Although triple certification requirements and customer trust remain major obstacles for homegrown materials entering the high-end market, as new production capacity rolls out and downstream validation progresses, domestically produced electronic-grade PPO is set to break the current dependence on imported products in the coming years and provide more competitive material support for the global AI computing sector.

CLS Marketwatch provides insights and analysis on China’s industries. You can contact the author at liujingyi@cls.cn

To subscribe to Bamboo Works weekly free newsletter, click here

Zhipu and MiniMax have lost more than 40% of their market value in just two weeks, as investors reassess the true worth of China's large language model developers

By Lee Shih Ta

Investor enthusiasm for AI stocks was in overdrive earlier this year, when Knowledge Atlas Technology Joint Stock Co. Ltd. (2513.HK), better known as Zhipu, and MiniMax Group Holdings Ltd. (0100.HK) made their Hong Kong stock market debuts amid sentiment reminiscent of the most exuberant days of China's internet and electric vehicle booms.

As some of the first large language model (LLM) developers to reach public markets, both companies quickly became investor darlings. On May 29, Zhipu's shares surged to an intraday high of HK$1,993, more than 17 times their IPO price, giving the company a market value of more than HK$880 billion ($112 billion). MiniMax closed the same day at HK$840, more than four times its listing price, with a market value exceeding HK$260 billion.

Based on Zhipu's 2025 revenue of 724 million yuan ($107 million), the company was briefly valued at more than 1,000 times sales at its peak. Its market capitalization even surpassed that of some profitable technology companies, underscoring how investor optimism toward China's AI champions had stretched far beyond what conventional financial metrics could justify.

But sentiment reversed sharply just two weeks later. By June 12, Zhipu's shares had fallen to HK$1,097, down 44.9% from their peak, while MiniMax had dropped to HK$396, a decline of 52.9%. Together, the two companies lost more than HK$400 billion in market value.

At first glance, the selloff appears to be the result of upcoming lockup expirations. According to Hong Kong Stock Exchange filings, stock held by Zhipu's first batch of cornerstone investors will become eligible for sale on July 8, involving 25.68 million shares, equivalent to about 11.9% of its H-share capital. Given that only about 11.74 million shares are currently freely tradable, the company's free float will triple overnight. MiniMax faces a similar situation on July 9, when shares held by cornerstone investors, anchor investors and certain existing shareholders will be unlocked, significantly increasing the supply of stock available for public trading.

Yet attributing the recent plunge entirely to lockup expirations oversimplifies the story. In global capital markets, a company's first major lockup expiry six months after listing is hardly unusual. The more important question is whether investors believe existing shareholders have a compelling reason to sell once their lockup periods end.

If investors were chasing scarcity value when they bought shares of China's AI companies earlier this year, the recent pullback reflects a different mindset. The market is beginning to recalculate how much time and capital these companies will need to turn their technological advantages into sustainable commercial returns.

In fact, the lofty valuations given to Zhipu and MiniMax were never fully supported by their financial performance. In 2025, Zhipu generated revenue of 724 million yuan, up 132% year-on-year. But it still reported an adjusted loss of 3.18 billion yuan. MiniMax posted revenue of 543 million yuan, a 159% year-on-year increase, while recording a net loss of 1.75 billion yuan. By conventional valuation standards, both companies remain a long way from profitability.

A more important reason investors were willing to pay such rich valuations was scarcity. As UBS China internet analyst Xiong Wei has noted, there are very few publicly listed large language model companies available to investors globally. The relatively recent listings of Zhipu and MiniMax, combined with their limited free floats, further amplified both their scarcity and liquidity premiums.

Investors were buying not only the technology, but also the promise of China's AI future and the premium that comes from a lack of other comparable investment opportunities. Those premiums now appear to be fading.

The capital race begins

Even as investors worried about upcoming lockup expirations, the two companies moved almost simultaneously to pursue listings on China's domestic A-share markets in Shanghai and Shenzhen. MiniMax announced in late May that it was exploring a potential listing on Shanghai's STAR Market, while Zhipu unveiled plans in early June to seek an A-share listing and raise 15 billion yuan, without specifying a market. The fact that both companies pushed ahead with domestic fundraising plans while their share prices remained well above IPO levels suggests management may see far greater capital requirements in the future than investors currently expect.

This seemingly contradictory situation reflects a new stage in the development of China's AI industry. On one hand, early investors are beginning to consider their exit options. On the other, the companies themselves continue to require large amounts of fresh capital. The reason is straightforward. Compared with traditional internet companies, large language model developers are far more capital-intensive. Training models requires massive GPU computing resources, inference services demand continued investment in data centers, and fierce competition for talent keeps pushing up R&D costs. Even after going public, these companies will likely be unable to fund the next phase of competition through existing cash flow alone.

As investors reassess these companies' capital needs and prospects for profitability, the basis for valuation is also changing. The market is gradually moving from a first phase of valuing technological capabilities to a second phase focused on commercialization. In the first phase, investors cared most about who had the strongest model. In the second, they are increasingly focused on which companies can convert technology into revenue, control costs and ultimately build sustainable business models.

Viewed from this perspective, the recent pullback in Zhipu and MiniMax shares does not necessarily signal the end of the AI boom. Instead, it may mark the beginning of a transition away from scarcity-driven valuations toward a more mature stage of price setting.

Such a shift is a familiar part of the maturation process for emerging technologies. Be it the internet two decades ago, electric vehicles a decade ago, or biotechnology more recently, investors were initially willing to pay hefty premiums for technological breakthroughs and future potential. But as industries mature, attention inevitably returns to more fundamental questions: Are customers willing to pay for the product? Can revenue continue to grow? And when will the company become profitable?

China's large language model industry now stands at a crossroads. Over the past year, investors have focused on model performance, technological breakthroughs and which company might emerge as China's equivalent of OpenAI. In the years ahead, however, the competitive battleground is likely to shift toward who can achieve the most commercial revenue, enterprise customer scale and capital efficiency. Put simply, the contest among AI companies is evolving from a race in technology to a race in business execution.

Lee Shih Ta is an editor at Bamboo Works.

You can contact him at shihtalee@thebambooworks.com

To subscribe to Bamboo Works weekly free newsletter, click here

The beauty industry is starting to rethink how its services are delivered as AI moves further into consultation, operations and treatment devices

By CLS Marketwatch

The medical aesthetics industry has been growing briskly in recent years, driven by steady demand in China. But certain issues have also come to the fore, making it a high-risk sector. Against that backdrop, AI technology, with its strengths in data processing, process automation and decision-making support, is ideally suited to address many of the challenges. Accordingly, AI is gradually engaging in critical segments of China's medical aesthetics industry, covering areas such as marketing, process management and even equipment operation. Yet a significant gap in technology adoption across facilities remains, and issues regarding efficacy and data security also pose real problems.

Holistic supporter in modern aesthetics

The current application of AI in medical aesthetics is concentrated in non-medical areas such as marketing, customer service response, database organization, content generation and process management, where it is an effective tool for reducing labor costs and standardizing service. Vision algorithms allow AI to quickly capture clients’ key facial features and convert them into data parameters, or to analyze patient consultations and summarize findings as electronic medical records. A notable example is the MMGPT AI assistant, designed specifically for the medical aesthetics industry, which has announced its integration with DeepSeek. This enables it to generate conversations through semantic analysis, match user needs via algorithms, and automatically trigger follow-up care.

Beyond this, AI applications are also spreading across optoelectronic medical devices. Leaning on AI's precision and mechanical accuracy, devices equipped with intelligent ultrasound imaging and AI-controlled systems can automatically identify and annotate subcutaneous tissue layers and output real-time data on average skin thickness and specific fascia layer depth, empowering physicians to adjust treatment parameters accordingly. Similarly, another type of temperature-controlled therapeutic device uses AI algorithms to monitor temperature in real time and adjust radio frequency output power in milliseconds to ensure treatments remain within effective and safe temperature ranges. These functions play to the core value of AI’s abilities, and can greatly reduce treatment deviations caused by human error.

Why most clinics hesitate on AI

While AI has shown potential in certain areas, it still grapples with multiple practical limitations, leading most medical aesthetic institutions to adopt a cautious stance toward its use. In upstream R&D, the clinical success rate of AI-driven drug development has fallen short of expectations. Similarly, when AI is used to identify potential materials for medical aesthetics, the candidates it proposes often fail due to insufficient efficacy or unexpected toxicity. This is because while AI excels at identifying patterns in known data, the variability in human immune responses and metabolic variations in human bodies far exceed existing clinical data, and the gap between animal models and the human body is difficult to bridge through computational power alone.

More importantly, data has become the biggest bottleneck right now. The severe shortage of valid data, such as data on medical procedures, treatment processes and postsurgical outcomes, directly hampers AI’s potential to evolve from an assistive role to taking on more essential functions.

At the same time, since medical aesthetics deals in large volumes of physiological and health data, the industry is sensitive and prone to public scrutiny. To address this, some firms, like Bloomage Biotechnology (688363.SH) and Beauty Farm (2373.HK), have begun using encrypted data transmission, real-time monitoring and risk control mechanisms to intercept abnormal access and detect data risks. Other companies have employed a dual-review mechanism combining system filtering and manual verification, and use on-premises infrastructure to ensure that sensitive data remains within their internal networks.

It’s also worth noting that industry associations have partnered with internet giants to begin constructing a comprehensive database for the medical aesthetics sector.

Tomorrow's clinics: AI as future core

Despite the many challenges, insiders are generally optimistic about the prospects for AI in medical aesthetics. Most importantly, AI is expected to have a profound impact on all professions that profit from information gaps. There are already signs that a growing number of beauty seekers are shifting from traditional online media consultations to directly querying AI. And as AI continues to advance, new service models such as unmanned clinics and personal beauty consultants are expected to emerge, boosting efficiency for both upstream R&D and downstream clinics.

In terms of taking greater advantage of AI’s strengths, industry forecasts suggest a pivot from assisting with recommendations to semi-automated decision-making. For instance, So-Young International (SY.US) has already launched AI toolkits that provide end-to-end intelligent services, including medical aesthetics knowledge inquiries, clinic searches, appointment scheduling and reminders. Moreover, AI can also collaborate with cloud computing during treatment to streamline the entire patient journey from detection to consultation, treatment, and follow-up, and accumulate reusable data through post-treatment feedback.

Additionally, the integration of AI and devices is likely to emerge as a trend, with AI potentially used to manipulate devices or even work in conjunction with robotic arms to perform certain tasks. As this extends beyond simply commanding machines to working collaboratively with humans, AI is poised to play an even more central role in the future.

In summary, the application of AI in medical aesthetics has gradually expanded from non-core areas to equipment control and clinical workflows, laying a path for reducing risks associated with human error and improving service standardization. As clinical treatment data accumulates, AI will become better positioned to progress toward semi-automated decision-making, achieve deeper integration with devices, and facilitate new business models. The industry’s ability to pull off this transformation will also ultimately depend on development of its data governance framework.

CLS Marketwatch provides insights and analysis on China’s industries. You can contact the author at liujingyi@cls.cn

To subscribe to Bamboo Works weekly free newsletter, click here

SpaceX’s upcoming IPO has put the commercial space industry in the spotlight for investors, raising the question of how these companies should be valued

By Lee Shih Ta

If everything goes according to plan, June 12 could become a watershed moment for the global commercial space industry.

Market reports suggest that Elon Musk’s SpaceX space venture has entered the final stages of preparing for an IPO, with a valuation that could reach $1.75 trillion or more. Such a listing would not only rank among the largest tech IPOs of all time, but could also establish the first meaningful valuation benchmark for a commercial space company.

Commercial space has been officially incorporated into China’s strategy for developing “new quality productive forces” over the last two years. Both the central and local governments have rolled out supportive policies, while development of low-Earth-orbit (LEO) satellite projects such as the Spacesail and GW constellations have accelerated. Companies including LandSpace, GalaxySpace and MinoSpace have completed new financing rounds, and several are reportedly preparing for public listings. According to the China Center for Information Industry Development (CCID), the size of China’s commercial space industry has rocketed from about 800 billion yuan ($118 billion) in 2020 to nearly 3 trillion in 2025.

Yet there is still little consensus on how the industry should be valued. New energy vehicle (NEV) makers can be assessed by their sales volumes and market share. AI companies can be measured by model capabilities and user scale. But what should investors look for when valuing commercial space companies? The number of rocket launches? Number of satellites deployed? The size of their order backlog? Or something else?

In many ways, SpaceX’s IPO provides an ideal window to examine this question. Under traditional aerospace valuation models, SpaceX would clearly not come close to its current estimated valuation. While the company operates one of the world’s most successful commercial launch systems, its valuation far exceeds what could be justified by launch services alone.

Over the past decade, SpaceX’s greatest achievement may not be landing reusable rockets, but rather using lower launch costs to build the Starlink satellite network. For investors, the real attraction is not SpaceX’s actual launch capability, but rather a global communications network made up of thousands of satellites and the recurring revenue it could generate in the years ahead.

According to publicly available data, Starlink had deployed more than 7,000 LEO satellites by early 2026, with its subscriber base surpassing 5 million and annual revenue estimated at more than $10 billion. What investors are chasing is not the 7,000 satellites themselves, but the millions of paying subscribers behind them and their steadily growing stream of service revenue.

By comparison, the LandSpace Constellation project, led by Shanghai Spacecom Satellite Technology plans to deploy about 15,000 LEO satellites, a number that would eventually exceed Starlink’s current constellation of satellites in orbit. Yet the real concern for investors is not how many satellites ultimately get launched, but whether they can support a sustainable business model and generate stable cash flow.

From satellites to networks

China’s commercial space sector broadly encompasses three distinct groups. The first is companies focused on launches and manufacturing. These include commercial rocket makers such as LandSpace and iSpace, as well as satellite developers and manufacturers, whose competitive edge lies in their engineering expertise and execution capabilities. These businesses face high technological barriers to entry. But their revenue models remain largely project-based, making them more comparable to advanced manufacturing or defense contractors.

The second group is operations and services providers. Companies such as Shanghai Spacecom Satellite, which is building the Spacesail Constellation, and Geespace, backed by carmaker Geely, derive their long-term value not from the number of satellites they launch but from their ability to build and operate reliable satellite communications networks.

The third group is centered on data applications. Companies such as Piesat Information Technology (688066.SH) and Geovis Technology (688568.SH) have evolved beyond simply supplying satellite imagery. They now provide aerospace information services, digital Earth platforms and data-driven solutions. Rather than selling actual satellites, these companies provide the information and insights generated by those satellites.

Investment activity in China’s commercial space sector has increasingly shifted downstream. Beyond satellite internet infrastructure, investors are increasingly looking for companies that can provide remote-sensing data, geospatial services and aerospace data applications. Industries ranging from agriculture and logistics, to energy management, low-altitude aviation and autonomous driving, are increasingly relying on real-time spatial information to support their decision-making.

As a result, the value chain of the commercial space industry is gradually extending downstream. Investors have traditionally viewed aerospace companies through a manufacturing lens. But the greatest valuation upside in the future may come from companies providing data services, communications networks and the applications built on top of them.

Commercial space can be better understood as an industrial chain rather than as a single industry. From rocket launches and satellite manufacturing to satellite operations and data services, each segment serves different customers, addresses different markets and follows different business models. Investors may currently group them under the same theme, but valuation differences are likely to become increasingly apparent as the industry matures.

Once SpaceX goes public, the market may, for the first time, establish a valuation framework that can serve as a reference point for the commercial space industry.

For Chinese companies, the challenge is not simply sending more satellites into orbit, but turning those satellites into viable businesses. Spacesail may deploy 15,000 satellites and LandSpace may launch more rockets. But without recurring revenue, those assets ultimately remain cost centers rather than value creators.

Over the past decade, the companies that have commanded the highest valuations have generally not been those that merely control hardware, but those that control networks, users and data. Tesla is one case in point. Nvidia is another.

Whether commercial space will follow the same trajectory remains an open question. But as more Chinese commercial space companies move toward public listings, the key question for investors may ultimately be: Who will be able to charge the most for services in the future space economy?

Lee Shih Ta is an editor at Bamboo Works.

You can contact him at shihtalee@thebambooworks.com

To subscribe to Bamboo Works weekly free newsletter, click here

China’s humanoid robot boom is no longer just about futuristic demos — the real race is now shifting toward scale, cost and industrial deployment

By CLS Marketwatch

Humanoid robots are rapidly emerging as a new focal point in the global technology race. As AI, sophisticated manufacturing, new materials and other technologies increasingly come together, humanoid robots have made the leap from laboratories to diverse real-world scenarios. China, armed with a comprehensive manufacturing system, vibrant application market, and heavy R&D investment, has been laying the groundwork for a robotics industrial chain spanning from core components to assembled systems. Yet despite high market expectations, commercialization still faces big challenges related to efficiency, data infrastructure, and compatibility.

Robots take on diverse roles

China’s robotics industry is currently moving toward diversified development. Manufacturers are exploring different segments based on their respective strengths, ranging from entertainment to specialized industrial uses, with distinct differences in their offerings.

To begin with, low-cost robotic products are flooding the mass consumer market. A typical example is a smart AI-powered robot pet dog priced at 150 yuan ($22), capable of voice interaction, singing, dancing and doing somersaults. At the same time, high-value applications in medical settings have also made progress, reflected by hair transplant surgical robots now used at both private and public hospitals, priced at around 1.2 million yuan.

Meanwhile, industrial and commercial service scenarios are seeing more in-depth applications. In three representative settings of “unmanned factories,” “smart retail” and “smart kitchens,” robots are taking on specialized roles such as retrieving items, providing sales guidance and performing cooking tasks. Diving robots designed for underwater detection, marine scientific research, and water conservancy inspections are expected to achieve mass production in the second half of 2026, priced at around 600,000 yuan for a complete set.

Another noteworthy area is entertainment. A prime example is the humanoid robot developed by Unitree Robotics that can engage in boxing matches, which not only entertains audiences but also demonstrates advances in dynamic balance and real-time interactive control techniques. Interestingly, robot rental services are gaining traction. Take JiShiZu (Shanghai) Technology, the world’s first open robot rental platform, for example. It offers robot rental services to both corporate clients and individual users, with daily rents between 5,299 yuan and 11,900 yuan, effectively lowering the barrier to trying out cutting-edge robotics for the general public.

Scale benefits unlock upstream potential

Within the humanoid robot industrial chain, the upstream sector for core components has benefited most from the current boom. Orders for key components such as precision reducers, dexterous grippers, and servo drives are overflowing, with production capacity expanding at a rapid pace.

Leader Harmonious Drive Systems (688017. SH), a key player in the precision reducer sector, disclosed that it shipped 500,000 units in 2025, with a target of 800,000 for 2026. The company is continually expanding its capacity, and orders are already booked through 2027, reflecting strong demand from complete machine makers for upstream components.

As a critical end-effector for humanoid robots, dexterous hands are also increasing sales at lower prices. An example is Beijing Inspire Robots Technology’s latest anthropomorphic five-finger dexterous hand, featuring an active thumb force of 15N and static passive load capacity of 8 kilograms per finger. Its price of 18,000 yuan is half the price of similar products from the same period last year, primarily attributable to rising shipments.

At the same time, 3D printing technology is also infiltrating the robotic upstream. Given the manufacturing challenges underlying humanoid robots, such as slow R&D iteration, high costs for small-batch production and design constraints, 3D printing offers remarkable cost and efficiency advantages over traditional processes. Despite this, the industry’s overall production schedule remains tight. In particular, core component sectors are seeing surging shipments, soaring orders, favorable pricing and innovations in manufacturing procedures.

Real-world validation remains the ultimate test

IDC forecasts that global shipments of humanoid robots will exceed 510,000 units by 2030, with a compound annual growth rate (CAGR) nearing 95%, indicating a generally optimistic outlook. Still, industry insiders agree that three core challenges remain to robot commercialization: efficiency, data, and adaptability to real-world industrial environments.

Crucially, the gap in robotic operational efficiency remains significant, with overall productivity in real-life tasks still generally lagging far behind humans. At the same time, data, rather than hardware, has emerged as the primary obstacle to the industry’s advancement. Since the whole data ecosystem is still in its infancy, an immediate priority is to keep refining infrastructure through ongoing iteration by data users, alongside data generation and interaction, ultimately fostering a competitive advantage for the industry.

The deployment of embodied intelligence in industrial settings also still faces systemic adaptation challenges. In industrial settings, for example, simply porting laboratory algorithms to the factory floor isn’t enough; instead, architectures must be rebuilt from the ground up based on fundamental logic for each industrial operation. The most prominent challenges include the “impossible triangle” of cost, efficiency, and reliability; and exponential increases in system complexity resulting from the growing variety of tasks, work cycles and performance requirements when dozens or even hundreds of robots run simultaneously within a factory.

Looking ahead, humanoid robots should not be regarded merely as a product, but as a technological mix that will lead future scientific development. Their technological breakthroughs will generate powerful spillover effects, driving rapid progress in a range of advanced domains including AI, new materials, quantum computing, brain-computer interfaces, the internet of things and big data, among others. It is important to note, however, that the industry will need to keep validating robots’ value-creating potential through real-world practices.

CLS Marketwatch provides insights and analysis on China’s industries. You can contact the author at liujingyi@cls.cn

To subscribe to Bamboo Works weekly free newsletter, click here

Rapid advances in AI navigation, LiDAR and RTK positioning are helping manufacturers accelerate overseas, with Europe and North America in their sights

By LeadLeo Research Institute

Robotic lawnmowers are becoming the new standard in smart gardening. According to data from LeadLeo Research Institute, China’s intelligent robotic lawnmower exports reached $1.01 billion in the first quarter of 2025, up nearly 60% year-on-year. Some companies serving the European market even received an additional 30% in orders after the Lunar New Year holiday. As navigation, obstacle-avoidance and lithium battery technologies rapidly mature, Chinese companies are stepping up their efforts to rake in global market share.

Robotic lawnmowers are intelligent gardening devices that combine AI, navigation systems and cutting mechanisms to autonomously trim grass using preset programs. Compared with manual mowing, they improve efficiency, reduce labor costs and deliver more consistent mowing performance. The industry has entered a phase of rapid development, evolving from early “random collision” operating models to more advanced systems featuring route planning and active obstacle avoidance.

The upstream supply chain currently includes key components such as real-time kinematic (RTK) modules, LiDAR sensors, vision modules, system on chips (SoC), lithium batteries and drive motors. China has largely completed domestic substitution across much of the supply chain. Companies including Bynav Technology, Unicore Communications, Rockchip Electronics (603893.SH) and RoboSense (2498.HK) have already entered the mainstream OEM supply chain. As technologies mature and production scales up, component costs have fallen sharply. For example, LiDAR prices have dropped from 200,000 yuan ($29,400) to 300,000 yuan a few years ago to roughly $200 today, while RTK module prices have declined from around 2,000 yuan to about 300 yuan, accelerating the adoption of boundary-wire-free robotic lawnmowers.

Current mainstream technologies are broadly divided into wired and wire-free systems. Earlier products required boundary wires buried underground around lawns, resulting in higher installation costs and more complicated maintenance. New-generation products increasingly rely on RTK positioning, AI vision and LiDAR solutions, enabling them to create virtual boundaries and map routes directly, significantly improving intelligence levels. The industry widely believes wire-free systems will become the dominant technology in coming years, presenting a major opportunity for Chinese companies moving into the high-end market.

The competitive landscape is also changing. The global market was long dominated by Sweden’s Husqvarna (HUSQ-B.ST) and China’s Ninebot(689009.SH), using self-developed RTK navigation and vision systems, has become a leading wire-free robotic lawnmower brand with annual shipments exceeding 80,000 units. Greenworks (301260.SZ) reported that revenue related to its third-generation products surged 68.92% year-on-year after launching in Europe.

In addition to Ninebot and Greenworks, other Chinese companies including Ecovacs Robotics (603486.SH), Daye Garden (300879.SZ) and Zhongjian Technology (002779.SZ) are also aggressively expanding overseas. Emerging brands such as Mammotion, Dreame Technology, MOVA and Lymow Tech are focusing on RTK, LiDAR and AI-vision integrated solutions to target the mid- to high-end market. Some startups are even promoting fully AI-powered lawn management systems, hoping to expand their products beyond mowing tools into broader smart garden ecosystems.

Big potential in Europe, U.S.

Europe is currently the world’s largest market, boasting the highest robotic lawnmower penetration rate due to its high proportion of standalone homes, expensive labor costs and widespread lawn maintenance culture. Meanwhile, larger yard sizes are fueling rapidly growing demand for high-endurance and high-efficiency products in the U.S. Research indicates that overall penetration of robotic lawnmowers remains below 10% globally, leaving substantial room for future growth.

On the distribution side, offline sales still account for roughly 70% to 80% of the market. Since European and North American consumers place strong emphasis on after-sales service and installation support, many Chinese companies are aggressively building overseas channel networks. Ninebot has partnered with Orgill to enter the North American market, while Sanfun has cooperated with European retail giant Lidl. Dreame and Ecovacs are also accelerating their rollout of overseas offline retail channels.

LeadLeo Research Institute forecasts the global robotic lawnmower market will grow from $5.91 billion in 2025 to $22.2 billion by 2030, growing at an average rate of 30.34% annually over that time. Key growth drivers include environmental policies, falling upstream component costs and enormous global demand for lawn maintenance. The world currently has roughly 200 million lawns, while annual shipments of traditional lawnmowers total around 30 million units, all representing potential replacement demand.

The industry has evolved into a multi-player competitive structure. The first tier includes Husqvarna, Positec and Ninebot, while the second consists of companies such as Ecovacs, Greenworks and Mammotion. As technologies continue to mature, the market is expected to move toward greater concentration and differentiated competition, with companies introducing products tailored to different lawn sizes, terrains and price ranges. For Chinese firms, the robotic lawnmower export boom is no longer just a competition in home appliances or gardening tools, but increasingly a broader contest in AI, navigation and intelligent hardware capabilities.

LeadLeo Research Institute is an original content platform for research on banks and companies and an innovative digital research service provider with nearly 100 senior analysts. You can contact the platform at CS@leadleo.com

This commentary is the views of the writer and does not necessarily reflect the views of Bamboo Works

To subscribe to Bamboo Works weekly free newsletter, click here

AI toys are evolving from chatbots in plastic shells into intelligent playmates with memory, personality and emotional awareness, redefining how children interact, learn and connect

By CLS Marketwatch

As AI technology accelerates its penetration into myriad consumer sectors, the toy industry has emerged as one of the most imaginative fields for its application. As large language models (LLM) have matured and fallen in cost over the last two years, AI-powered toys have evolved from rudimentary products only capable of superficial interactions, such as “voice Q&A,” to sophisticated systems that integrate advanced technologies.

This new generation of toys includes elements such as agent architecture, multimodal interaction and mood-aware LLMs. It’s worth noting that mere technological stacking can no longer satisfy market demands. That’s leading the industry to undergo a notable transformation from feature-rich devices to context-aware intelligence, and driving a redefinition of the value proposition of AI-backed toys.

Rethinking AI-powered toys beyond cloning

The AI-powered toy market in China has gained some initial traction, yet still faces a common issue known as “LLM cloning.” This refers to products that simply transplant the conversational and Q&A capabilities of general-purpose LLMs directly into toy hardware, without a comprehensive workflow tailored to children's play scenarios. The interaction logic of such products is no different from that of ordinary LLMs, rendering them ineffective at providing in-depth guidance that combines education with entertainment, and failing to establish long-term emotional bonds with children.

That means that unless AI-driven toys address parents’ genuine concerns or meet children’s growth needs, they are essentially just short-lived electronic gadgets, and their limited interactive capabilities will struggle to retain users. Early cloning products have already exposed issues such as low user retention and lack of differentiation. Against this backdrop, the industry has begun experimenting with technologies like intelligent agents, multimodal recognition, and mood-aware LLMs to upgrade AI-powered toys from simple hardware + voice combos to smart companions featuring emotional interaction and scenario adaptation abilities.

Redefining toys through multi-agent systems

The key technological approach to achieving this “qualitative leap” is the introduction of AI-powered agents. Leading companies have begun to move away from the cookie-cutter logic and are instead adopting multi-agent systems. Such systems enable different agents to work in concert, each corresponding to separate systems for memory, personality, and user profiling, so that AI-driven toys can autonomously respond to complex scenarios and exhibit more human-like behavior. Technical barriers to building agents are not high. What’s more important is that excellent design must be grounded in extensive market research and in-depth analysis, which fundamentally determines the quality of task execution by the agents.

At the same time, in response to stringent requirements for applications targeting children, the new generation of AI-powered toys has undergone a comprehensive restructuring in terms of safety and adaptability. Most importantly, AI agents are positioned solely as communication interfaces and do not possess the authority to perform specific actions, thereby mitigating the risk of system takeover. Furthermore, all user data is stored in encrypted form to ensure end-to-end data safety.

The way to mainstream acceptance

The AI-backed toy sector is currently scaling from niche novelty to mass appeal, with market acceptance growing rapidly. As that happens, parents value the dual benefits of educational enrichment and emotional support, while children demonstrate far greater engagement with AI toys than traditional smart toys due to the interactive experience.

Given this trend, AI-backed toys in development need to fine-tune general-purpose LLMs specifically for toy-related scenarios, particularly to meet the high standards of content safety, conversational habits, and early learning required in child-oriented contexts. For example, the market has already seen the emergence of AI photo-learning cameras that integrate multimodal recognition technology to achieve a “snap-identify-educate” closed-loop; there are even AI-powered therapy dolls developed in collaboration with external partners that incorporate the “psychology-focused version DeepSeek” mood-aware LLM, further refining contextual applications.

According to the “White Paper on AI Toy Consumption Trends,” the global AI-driven toy market is expected to exceed 100 billion yuan ($15 billion) by 2030, growing over 50% annually. A typical product is Shifeng Cultural Development’s (002862.SZ) “Intelligent Flying Rabbit,” which is equipped with a customized AI model designed specifically for children. Building on its foundation as a safe AI-powered companion, the product has added functions such as child-friendly karaoke and story co-creation, making it a long-running bestseller.

Broadly speaking, the current AI toy market is exhibiting three key trends: First, a multi-player competitive landscape has formed among IP rights owners, innovative manufacturers, AI technology providers, and toy design and production companies, though no individual players have come to dominate the market. Second, products have shifted from a focus on “piling on features” to emphasizing “experiential value,” where users seek meaningful companionship and personalized experiences rather than basic conversation. Third, lack of differentiation, data security risks, and the technical and cost barriers associated with multimodal interaction remain common bottlenecks to development across the industry.

From functionality stacking to agent-driven interaction, China's AI-powered toy industry is witnessing a remarkable value leap. Users’ complex demands for emotional bonding, early learning, and contextual communication are forcing manufacturers to undertake disruptive transformations at the very foundation of their technical architectures. The incorporation of sophisticated capabilities, such as multi-agent systems, scenario adaptation, and end-to-end safeguards, marks the evolution of AI-backed toys into a new era as intelligent playmates. These next-generation AI-powered toys will undoubtedly open up new possibilities for children’s growth and development.

CLS Marketwatch provides insights and analysis on China’s industries. You can contact the author at liujingyi@cls.cn

To subscribe to Bamboo Works weekly free newsletter, click here

As China’s economy enters a new phase of slower growth, food and beverage brands are taking a more differentiated approach to reaching Chinese consumers

Key Takeaways:

• Foreign food and beverage companies are realizing that Chinese consumers prioritize health and emotional comfort over simple low prices
• Overseas students and global travelers remain a vital, organic pipeline for introducing foreign brands into the China market

By Doug Young and Brent Li

China’s food and beverage market is massive, but it’s also becoming starkly unforgiving as consumer caution grows and competition intensifies. For multinationals and local players alike, the formula for survival has shifted away from simple expansion, and increasingly relies on ruthless re-invention and hyper-localization, according to industry insiders speaking at the Food & Beverage Innovation Forum 2026, which featured 53,000 attendees, 650 exhibitors and more than 160 speakers last month in Hangzhou.

As China’s economy enters a new phase of slower growth, the standard playbook of flooding shelves with cheap consumer goods is losing its edge. Gary Hsi Chu, chairman of yogurt giant Yoplait Ltd., cautioned executives that navigating the current dairy sector requires a willingness to upend legacy revenue streams. Companies must never fear disrupting or cannibalizing their own core business, Chu noted.

Contrary to the widespread narrative of a massive downgrade by increasingly cautious consumers, Chu argued that people are not simply looking for the cheapest products. Instead, they are willing to pay a premium for healthier products. He pointed out that excessive competition is often the symptom of a deeper corporate failure — a lack of understanding of real consumer needs that results in undifferentiated products.

Eric Xia, general manager of Jiale Dairy, reinforced the need for focus over a one-size-fits-all approach. He noted that corporate R&D is often paralyzed by trying to chase every fleeting trend. Instead of reacting to every request, Xia said, companies must filter out the noise and only manufacture products with distinctive characteristics, using consumers' actual emotional reactions as a gage of demand.

This sentiment echoes across the beverage sector, where functional demands are increasingly overshadowed by the experience that people get when consuming a product. Chris Chen, retail general manager at Peet's Coffee, offered a sober assessment of China's highly competitive coffee market. Amid fierce price wars, the brand has focused on doing "difficult but correct" things, he said. Over the past three years, the U.S.-based coffee chain has pushed into more than 20 markets across China, pursuing a localized strategy with specialty outposts like the Xi'an Metropolitan store and a ski-friendly Songhua Lake Snow House store. By maintaining a small-batch roasting model, the company is attempting to sell a lifestyle rather than just caffeine.

Tapping a diaspora

While some global brands meticulously plan their China entry, others are discovering that their best point of entry comes via word of mouth from everyday travelers. Pavel Kislyakov, general manager of North Asia at German chocolatier Storck, highlighted an unconventional backdoor into the world's second-largest economy. Interestingly, the company’s Knoppers brand first entered China through overseas Chinese students purchasing products abroad and bringing them home, Kislyakov said.

He emphasized that this is an important way that some foreign brands enter China. Chinese outbound travelers regularly rank among the largest global tourist groups. As these travelers continue to broaden their world understanding, they increasingly seek the same high-quality products they discover abroad back home in China. Storck is now using this same organic momentum to introduce its Merci chocolate line, hoping the product's association with gratitude will resonate.

But getting a product on the shelf isn't the final hurdle. In an era of fractured attention spans, brand equity is practically nonexistent unless heavily managed. Donna Li, head of data insight and analytics at Kraft Heinz China (KHC.US), argued that brands need to constantly earn their loyalty. To bridge the cultural gap, the company localized its approach with campaigns like "Ketchup Meets Chinese Cuisine," which brings a traditionally Western condiment closer to Chinese consumers.

Tom Zhang, an associate partner at Prophet, added that the baseline for maintaining relevance has grown far more complex than it once was. Brands must continuously evolve to address many needs, deliver strong value-for-money propositions, provide emotional comfort, and activate deep interaction. This focus on emotional comfort is an emerging frontier in a high-stress market, suggesting that today’s winning products are ones offering consumers a psychological respite.

Even local retail heavyweights are restructuring their supply chains to meet these nuanced expectations. Paul She, chief merchandising officer at grocery store operator Yonghui (601933.SH), detailed a transformation shifting away from short-term, price-driven logic toward long-term quality evaluation. Meanwhile, Jonathan Shen, a vice president at online grocer Dingdong (DDL.US), explained how fresh produce is merely a high-frequency entry point to build user trust, allowing the platform to pivot into more individualized private-label products.

The shift away from one-size-fits-all to more personalized experiences is forcing an overhaul in product development. Sean Fu, a director at IDEO, warned that the industry's reliance on data-driven incremental innovation rarely opens new markets anymore. Pointing to a co-creation project with instant noodle giant Master Kong (0322.HK), Fu explained that true breakthroughs start by understanding real human needs. Rather than just upgrading instant noodles, Master Kong has redesigned its packaging with wooden chopsticks and a rounded lid to deliver a ritualistic "freshly cooked" experience.

Ultimately, simply having a globally recognized logo isn't sufficient anymore. The new reality is clear — companies aren't just selling calories. They're fighting to align with the shifting lifestyle and emotional needs of the modern Chinese consumer.

Doug Young is an editor at Bamboo Works.

You can contact him at dougyoung@thebambooworks.com

To subscribe to Bamboo Works weekly free newsletter, click here

AI’s cost structure is changing as it shifts from a tool to an autonomous executor, with both businesses and other users increasingly paying for services

By Lee Shih Ta

Over the past two years, the business model for AI has been built on what appeared to be a solid assumption: relatively low computing costs allowed companies to scale users quickly through a “free + subscription” model. Whether in chatbots or productivity tools, AI was typically packaged as an add-on feature to enhance product appeal, rather than as a core revenue generator. Users paid fixed monthly fees for unlimited use, a model that helped accelerate widespread adoption.

But that assumption is now starting to break down.

As AI shifts from being a simple Q&A tool to agent-based products capable of completing tasks autonomously, its cost structure is changing. AI is no longer just responding to prompts. It runs continuously, repeatedly calling models and tools, and often performs multiple rounds of reasoning and refinement within a single task. This shift is transforming AI from a “passive service” into an “active executor,” where the underlying consumption is no longer a one-off computation but a continuous accumulation of resources.

This shift is clearly reflected in changes in computing usage. According to media reports, daily token consumption in China’s AI model market has surged from about 100 billion in early 2024 to roughly 140 trillion in 2026. This growth is driven not only by an increase in users, but also by a fundamental change in usage patterns. AI is evolving from an occasional tool into a system embedded in an enterprise’s daily operations, with growth primarily driven by recurring process-based calls rather than one-off queries. As AI becomes a resource-consuming core capability, the traditional low-cost or free model becomes increasingly difficult to sustain.

This pressure is already showing up on the supply side. Between the second half of 2024 and 2026, Alibaba (9988.HK; BABA.US) and Baidu (9888.HK; BIDU.US) adjusted their pricing for certain cloud services, with increases ranging from 5% to 34%. As AI usage grows in both frequency and complexity, computing demand far exceeds that of traditional applications, pushing up cloud resource costs. These price adjustments reflect that rising pressure.

More importantly, the pricing model itself is evolving. In Alibaba Cloud’s large model services, companies no longer pay for a product, but are charged based on token usage, with input and output priced separately. Alibaba’s Ding Talk service, meanwhile, translates AI capabilities into usage counts, with different tiers offering different quotas. Whether measured in tokens or actions, the core idea is the same: AI is becoming a metered, consumable resource.

This shift is not limited to the infrastructure layer. As AI capabilities are integrated into software products, pricing structures at the application level are also rising. Kingsoft Office (3888.HK; 688111.SH) has introduced AI features such as writing, summarization and data analysis into its WPS product, similar to Microsoft Word, placing them within membership or premium tiers. Baidu has similarly bundled AI capabilities into its Wenku and cloud storage services, which now have over 40 million paying users. While these products have not universally raised their prices, they are creating new paywalls through feature tiering. For users, maintaining productivity gains often means spending more overall.

Take ByteDance’s AI assistant Doubao as an example. It recently began testing three subscription tiers on top of its free version, with advanced features focused on high-complexity use cases such as generating presentations, data analysis, and content creation. While the free version remains sufficient for everyday tasks, more advanced needs — such as long-form text processing, multi-step reasoning, or structured outputs — require upgrading to a paid plan. This reflects a broader shift in AI pricing logic, where charges are no longer based solely on feature differentiation, but directly tied to underlying computing and inference costs.

In effect, this is no different from a direct price increase.

Behind these changes is a shift in cost structure. Users are paying for the move from software to computing resources. More importantly, this shift in pricing logic is reshaping how value is distributed across the industry. As AI evolves from a product into infrastructure, value begins to concentrate at the lower layers. For cloud service providers and model developers, revenue can scale alongside rising demand for computing power. In contrast, application-layer companies may benefit from efficiency gains enabled by AI, but their growing reliance on underlying resources could compress their profit margins.

AI has not eliminated costs. Instead, it has transformed them into usage-based expenses. Under this model, traditional low-cost subscription strategies will come under pressure, as costs are now directly tied to computing consumption. In their place, hybrid pricing models are emerging, combining usage volume, frequency, and even task outcomes. In this process, AI is no longer a product that can be simply priced, but is increasingly becoming a fundamental resource, much like electricity or bandwidth.

At a deeper level, this shift is rewriting the entire industry’s dynamics. When computing costs become the key variable, competition is no longer defined solely by product features or user scale, but by who can secure computing power at lower cost and who can utilize model resources most efficiently. As a result, the industry’s power structure is shifting, with value increasingly concentrated among providers of computing power and models.

At the same time, the logic for business expansion is also changing. In the past, scale helped dilute costs. But in the AI era, every call incurs a real expense. Under this structure, model developers and AI service providers bear the costs, agent operators convert them into revenue, and these costs are ultimately passed along to end users, creating a new layer of costs.

Lee Shih Ta is an editor at Bamboo Works.

You can contact him at shihtalee@thebambooworks.com

To subscribe to Bamboo Works weekly free newsletter, click here

The country’s energy storage sector is racing toward bigger batteries. But as capacity climbs, so do questions over safety, costs and whether bigger is truly better

By CLS Marketwatch

Energy storage systems are accelerating toward larger-scale and longer-duration configurations, addressing a growing number of consumption scenarios challenging power supply stability. As the core component of such systems, energy storage cells are rapidly expanding in capacity. From the establishment of 314Ah as a mainstream configuration to the intensive rollout of cells exceeding 500Ah by multiple manufacturers, the technological iteration cycle continues to shorten. While debate persists over whether larger cell capacity is always better, it’s undeniable that leading players are racing to enter the large-capacity energy storage cell segment.

The race to higher capacities

While debate continues on whether energy storage battery cells need to keep evolving toward higher capacities, the technical roadmaps of leading enterprises show capacity upscaling has become a shared goal across the industry. The industry is currently transitioning from the mainstream 314Ah specification to 500Ah+ and even higher capacities. By the end of 2025, the penetration rate of 314Ah cells in the energy storage market had exceeded 70%, making it the benchmark configuration for mainstream projects. Notably, it took only about one year for this specification to go from introduction to market dominance, showing the pace of technological iteration in energy storage cells is far exceeding expectations.

Not only that, but the industry is continuing to upgrade to even higher-capacity products as reflected by the release of high-capacity cells covering capacities from 587Ah to 790Ah by mainstream manufacturers like Guangzhou Great Power Energy (300438. SZ), Ganfeng Lithium (002460. SZ) and Envision Energy.

Among those, Ganfeng Lithium’s 588Ah battery cell prototype has entered the testing and validation phase, with mass production expected to commence around May 2026. Envision Energy, meanwhile, announced that its 790Ah battery cells have officially entered mass production, simultaneously launching a 12.5MWh energy storage system equipped with this cell. As such, projections indicate that 500Ah+ battery cells will accelerate into large-scale application by the end of 2026 and are expected to become a mainstream specification in the market by 2027.

Forces driving cell upscaling

In fact, several factors are driving the movement into high-capacity battery cells. As energy storage projects trend toward larger-scale and longer-duration, and more high-power-consumption scenarios like AI computing centers emerge, buyers are imposing higher demands on the capacity and performance of individual cells. Large-format cells can deliver greater capacity within the same space, thereby saving customers on comprehensive costs such as land, infrastructure, and connectors.

At the same time, the shift toward a market-oriented economy is compelling enterprises to build core competitiveness through technological innovation and cost control. Therefore, while larger battery cells present greater challenges for thermal runaway and overall safety management, they can substantially enhance full-lifecycle economic efficiency.

Analysis of project model data for three long-cycle high-capacity battery cell products from CALB Group (3931.HK) reveal that, through the dual upgrade of capacity and cycle life, high-capacity cell technology drives a continuous reduction in levelized cost of storage (LCOS) via substantial increases in discharge capacity, under the premise of controllable initial investment. This provides a more cost-effective solution for scenarios such as integrated wind-solar storage systems and long-duration energy storage.

Furthermore, since only a handful of frontrunner manufacturers have achieved breakthroughs in the mass production of high-capacity battery cells, these firms are able to effectively attain competitive edges by developing dedicated technological platforms and accumulating core manufacturing expertise to secure first-mover advantage. Meanwhile, they are further strengthening their market dominance by expediting overseas expansion and adapting to standards and requirements in different regions.

From Ah numbers to system-level thinking

Despite the clear trend toward higher-capacity batteries, there is broad consensus within the industry that larger cell capacity is not necessarily better, as it faces multiple bottlenecks related to safety, energy density and market demand. Balancing safety and energy density is the core challenge. The amp-hour (Ah) rating of battery cells should not be the sole focus. Although technology allows for capacities exceeding even 5,000Ah, a relentless pursuit of capacity at the expense of thermal runaway management will pose serious safety risks.

Second, production line conversion and investment costs create practical constraints. A full-scale transition from existing production lines to high-capacity cells requires significant investment in equipment upgrades, process debugging, and capacity ramp-up, as well as considerable time. Therefore, a concentrated transition to high-capacity cells isn’t expected to begin until the second half of 2026 or 2027 at the earliest. Until then, 314Ah cells will remain the primary product being delivered.

Industry consensus holds that large-capacity cells will dominate the energy storage market over the next three years, with 500Ah+ products expected to achieve widespread adoption by 2027. Nonetheless, capacity upscaling has its limits. Given this, companies should define battery cells from a systemic perspective, determining cell capacity and specifications by working backward from end-user needs. At the same time, thermal runaway management for higher-capacity cells should be strengthened, which requires thorough validation across materials, structure, thermal dissipation and other aspects to avoid sacrificing safety in the pursuit of higher capacity.

Overall, the trend toward higher-capacity energy storage cells in China is being driven by a combination of evolving demand, cost optimization, and intense competition. However, there’s a limit to how much capacity can grow. Balancing safety and energy density, ensuring compatibility with market needs, and managing thermal runaway remain common obstacles that the industry needs to address collectively. High-capacity cells will gradually take over the market in the next three years, but a company’s core competitiveness will lie in finding the optimal balance between technology, cost-effectiveness, and safety.

CLS Marketwatch provides insights and analysis on China’s industries. You can contact the author at liujingyi@cls.cn

To subscribe to Bamboo Works weekly free newsletter, click here

Chinese platforms are stepping in to regulate mechanized mass production of online content aimed at viral hits more than truth or providing informational value

By Vivian Toh

A viral anecdote that recently circulated across Chinese social media suggested a couple had leveraged generative AI to produce WeChat articles at scale, amassing over 2 million yuan ($293,000) in annual income. The story was seductive in its simplicity: minimal effort, maximal yield and a seemingly frictionless path to money-spinning.

Within days, however, the narrative met a more sobering reality. Accounts tied to such practices were banned by Tencent’s (0700.HK) WeChat.

Rather than treating the incident as an isolated infraction, Tencent used it as a moral imperative. In a public statement delivered via a company channel, a Tencent spokesperson delineated a principle that is quickly becoming axiomatic within China’s content governance framework: AI may augment human creativity, but it must not supplant it.

A deep anxiety has been shared across content platforms that unregulated generative AI could precipitate a deluge of derivative, low-signal content, eroding both user trust and the economic viability of original creators.

Tencent’s intervention can be best interpreted as a preventive measure against online misuse. The company’s concern is less about the technology itself than about its propensity to facilitate what might be termed “content industrialization,” or the mechanized production of articles optimized for viral impact rather than veracity or insight.

This is not without precedent. China’s digital ecosystem has long contended with waves of attempts to monetize content, from clickbait to “pseudo-original” rewriting. Generative AI, however, introduces a qualitative shift. What once required coordinated human labor can now be executed by algorithms at a negligible cost.

The result is a potential glut of homogenous material: articles that differ in word but not substantively, saturating feeds with redundancy. In such an environment, differentiation becomes elusive, and the informational value of content declines rapidly.

Besides Tencent, the RedNote platform has concurrently adopted a more stringent stance toward AI-generated content, particularly in the context of commercial promotion.

RedNote has issued guidelines emphasizing authenticity and disclosure, proscribing the use of AI to fabricate narratives, such as product reviews, that simulate first-hand usage without any interaction. Enforcement has been neither perfunctory nor symbolic; non-compliant content has been removed and repeat offenders penalized.

What is notable here is the consistency of underlying logic across platforms. The objection is not to AI per se, but to its deployment as a vector for obfuscation, where automation masquerades as authorship and scale is achieved at the expense of substance.

In effect, Chinese platforms are converging on a regulatory point where AI is permissible as a tool of augmentation, but impermissible as an instrument of obfuscation.

Across the Pacific, the regulatory ethos appears more permissive — if not outright facilitative. Meta (FB.US) has embedded generative AI capabilities across its ecosystem, including Instagram, where AI-assisted content creation is increasingly normalized. Similarly, X has been positioned as a fertile ground for AI-native expression, from synthetic threads to algorithmically generated visuals.

This reflects an environment that is highly competitive and engagement-driven, where platforms are incentivized to lower creative barriers and accelerate content velocity. AI, in this scheme of things, is less a risk to be mitigated than a catalyst to be harnessed.

That said, Western platforms are not wholly indifferent to the attendant risks. Policies around labeling AI-generated content, particularly in political or advertising contexts, are gradually emerging. Yet enforcement remains comparatively diffuse, and the threshold for intervention is higher.

The divergence, then, is not one of awareness but of calibration. Chinese platforms have opted for preemptive constraints; Western platforms, for accommodation.

At its core, the debate over AI content moderation reduces to a question of signal integrity. What constitutes valuable content in an era where production is no longer a bottleneck?

Chinese platforms appear to privilege informational density and authentic authorship, seeking to curtail practices that dilute either. Western platforms, by contrast, are more tolerant of variability, prioritizing engagement metrics and creative plurality.

Neither model is without trade-offs. Excessive restriction may stifle innovation; excessive permissiveness may degrade quality. The challenge lies in finding a boundary that is neither overly prohibitive nor unduly lax.

From these divergent approaches, a set of provisional principles is beginning to crystallize.

For one thing, AI-native content is not inherently problematic. Its incorporation into content creation is both inevitable and, in many respects, desirable. It expands the creative toolkit and lowers barriers to entry. To categorically forbid AI-generated content would be both impractical and counterproductive.

Secondly, platforms should not suppress AI, but rather prevent its misuse. They bear a responsibility to curtail practices that exploit AI for content laundering. The objective is not to police creation, but to prevent the erosion of content quality through industrial-scale manipulation.

And thirdly, transparency and originality are non-negotiable. Encouraging AI-assisted creation must be contingent upon clear disclosure and a demonstrable contribution of human judgment. Labeling AI-generated content is not merely a compliance exercise; rather, it’s a mechanism for preserving trust between creators, platforms and users. Concurrently, platforms must continue to encourage original thought to ensure that human perspective remains the fulcrum of value creation.

Tencent’s recent intervention is, in this sense, less an isolated enforcement action than a bellwether. It signals an emergent consensus within parts of the industry that the future of content will be neither purely human nor purely machine, but a measured synthesis of both.

This commentary is the view of the writer and does not necessarily reflect the views of Bamboo Works

Vivian Toh is a Singapore-based journalist who writes about technology and co-founder of Tech Tech China, a technology media startup. You can reach her at Vivian_toh@techtechchina.com.

This article was originally published by the China Bizz Buzz.

To subscribe to Bamboo Works free weekly newsletter, click here

China’s home appliance giants are pushing into robots in their race to control the next gateway to the home — despite high costs and tech hurdles

By CLS Marketwatch

A sweeping transformation is unfolding in China's vast robotics market, led by home appliance makers. From cleaning to cooking, companionship to household chores, robots are emerging in ever-greater variety, with the household setting widely regarded as a primary target for embodied intelligence applications. The evolution of smart home appliances in China has gone beyond simple connection and control, moving into a new stage of embodied intelligence capable of autonomous action, environmental perception and understanding of human needs. Indeed, smart home appliances are entering a new development cycle.

Embodied AI finds its way home

According to data from the China National Light Industry Council, China has become the world’s largest robotics market, accounting for 55% of global robot production. China’s service robot output reached 18.58 million units last year, up 16.1% year-on-year. This immense market potential and a well-established industrial chain form the foundation for Chinese home appliance enterprises to advance into embodied intelligence.

A trend is becoming increasingly apparent in the current home appliance market, with majors such as Haier (6690.HK; 600690.SH), Hisense (0921.HK; 000921.SZ), Ecovacs (603486.SH), and TCL (1070.HK; 000100.SZ) all committed to developing embodied intelligent products for a wide range of household scenarios, including housework, cleaning, companionship and cooking. More importantly, smart home appliances are evolving from simple connection and control products to a new stage of perception and action.

Among these, the embodiment of home appliances and the integration of AI are emerging as two key trends. The AI integration at their core relates to “proactive intelligence,” meaning machines can anticipate user needs and automatically provide related services. The essence of such appliance embodiment lies in their spatial intelligence, which enables machines to move autonomously within the home setting.

How robots reshape daily home life

Home appliance companies are no longer limiting their robot products to single functions; instead, they are working to develop all-around home robots capable of handling complex household chores and delivering emotional companionship, catering to various household scenarios such as cleaning, emotional support and cooking.

In the field of housekeeping and cleaning robots, Ecovacs has launched Bajie, a home service robot equipped with a sliding base and gripping arm. It can perform tasks such as organizing toys, tidying tables, and retrieving and handing over objects. Powered by VLM and a family database, it can understand the spatial relationships and ownership of objects. Meanwhile, the brand also debuted Maotuaner, an AI-powered bionic companion robot with a Maltese dog appearance and able to support multi-sensory interaction, which is now officially on sale.

Haier Smart Home has also unveiled home service robots tailored to three specific segments: cleaning, companionship and household chores. Among them, the Haiwa robot can coordinate with home appliances throughout the house to perform tasks such as transporting groceries and sorting them into storage; the cleaning robot can use its grippers to put items like toys back in their proper places before planning a route for cleaning an entire house; and the companion robot supports medication reminders, fall detection, and emergency alerts.

When it comes to companionship and caregiving, emotional interaction and safety have topped the list of priorities in product design. For example, TCL’s AiMe is the world’s first modular AI companion robot, capable of multimodal natural interaction and adapting to family members’ habits through daily learning. Similarly, Hisense’s Moii companion robot is designed specifically for child and elderly care, while its Savvy butler robot can seamlessly control home appliances and handle household chores.

In the kitchen setting, Fotile’s world-first kitchen robot has taken culinary automation to the next level. Equipped with high-precision arms, this kitchen robot achieves fully automated operation throughout the entire process – from pouring oil and stir-frying to serving dishes and post-meal cleanup.

In fact, one of the major advantages traditional home appliance makers have in entering the embodied intelligence market is their supply chain strength and accumulated expertise in applying large language models. Take Ecovacs as an example. The firm has completed its investment and begun operating five manufacturing bases in Nanxun, Zhejiang. Its battery business module achieved 250 million yuan ($36 million) in sales during the first quarter, up 50% year-on-year. On completion of its capacity expansion, its monthly capacity will double. Such supply chain capabilities can be fully replicated for its robotics business, creating a cost-efficiency advantage.

Long road to robotic homes continues

While groundbreaking robotic products continue to emerge, observers are generally cautious on the large-scale entry of embodied AI robots into households, with estimates suggesting it will still take three to five years.

A series of challenges remain to be addressed. For instance, actions that humans take for granted, such as pouring water or tying shoelaces, involve multiple intricate technologies for robots to master, like high-dimensional joint control, force feedback, and visual reasoning. Conversely, robots can easily handle tasks that humans find difficult, such as complex calculations and playing chess. Furthermore, the complex and ever-changing household environment places higher demands on robots' adaptability.

The alignment between price and value is another critical threshold for penetration. Current prices for embodied intelligence products range from 10,000 yuan to 300,000 yuan, far higher than those for traditional home appliances. Therefore, only when prices match the value of the products they represent can robots transition from being a novelty to mass adoption. Furthermore, while Chinese consumers have a basic recognition of such products, the percentage of those with in-depth understanding of smart home robots still remains low.

Given these challenges, industry insiders generally believe that 2026 will mark the transition of humanoid robots from technical validation to mass production. However, supply chain development and technological maturity will likely still require another three to five years. Notably, the advancement is expected to evolve in four key directions: from single- to multi-function integration; from specific scenarios to serving as a central hub for whole-home smart systems; from purely utility-oriented tools to devices capable of emotional interaction; and a gradual decline in prices.

The overall trend is increasingly clear: robots are emerging as the next generation of smart devices for home use. While mass adoption will still require some time, home appliance manufacturers are steadily bringing the vision of smart home robots entering households closer to reality.

CLS Marketwatch provides insights and analysis on China’s industries. You can contact the author at liujingyi@cls.cn

To subscribe to Bamboo Works weekly free newsletter, click here

As AI shifts from generating content to taking action, it is no longer just a technology story — but a governance challenge

By Vivian Toh

China’s annual “Two Sessions” gathering has long served as a showcase for growth priorities, industrial strategy and technological ambition. This year, however, one theme is becoming harder to ignore: artificial intelligence is increasingly being treated as a governance problem.

That shift matters. For much of the past decade, China’s AI story was framed around scale — who had the most data, the strongest engineering talent, the deepest industrial base and the fastest path to commercialization. Now the debate is changing because the technology itself is changing. The rise of AI agents — systems designed to take actions across apps, devices and services — is forcing policymakers to confront a more complicated set of legal, economic and social questions.

That challenge is especially urgent as Beijing promotes its broader “AI+” strategy to accelerate adoption across the economy. The state wants AI to raise efficiency, strengthen industrial upgrading and support growth. But the more widely these systems are deployed, the more complicated the governance questions become.

The urgency is already visible in the OpenClaw frenzy that has swept China in recent weeks. In Shenzhen, crowds lined up outside Tencent’s office for help installing the viral open-source agent. Local governments in Shenzhen and Wuxi moved to subsidize OpenClaw-related projects.

Meanwhile, a U.S. federal judge recently issued a preliminary injunction in the case of Amazon vs. Perplexity AI, ordering the company to stop its Comet browser AI agent from accessing password-protected Amazon accounts. The court signaled that user permission alone may not be enough for AI agents to operate on third-party platforms, suggesting that both user consent and platform authorization could be required. This "dual authorization" principle may mark the beginning of a new legal framework for agentic AI.

Traditional chatbots mostly stayed within a single interface. They summarized documents or generated text, for instance. AI agents promise something more consequential. They are being built to compare products, organize calendars, book trips, summarize meetings, manage files and interact with multiple applications with limited human supervision.

That may sound like a natural evolution of the digital assistant. But it fundamentally changes the risk profile. Once an AI system can move across apps, access permissions, read screens and trigger actions, it becomes an operational layer between the user and the digital economy.

AI agents raise more difficult policy issues

How much data should an AI assistant be allowed to access? What counts as meaningful user consent when permissions are bundled into complex app ecosystems? If an AI agent makes a purchase, cancels a booking, mishandles sensitive data or gives a flawed recommendation that causes financial harm, who is legally responsible — the developer, the platform, the device maker or the user?

Over the past several years, Chinese regulators have already developed a visible framework for emerging digital technologies. Rules on recommendation algorithms, deep synthesis and generative AI services have established a broader pattern: Beijing generally allows innovation to move forward, but under a clear structure of state supervision, cybersecurity compliance and content-related responsibility.

Balancing innovation and control

That model reflects a familiar balancing act. China sees AI as strategically important for robotics, manufacturing, semiconductors, consumer electronics and long-term productivity growth. It wants domestic firms to compete globally while deploying quickly at home. But powerful digital systems can create risks if commercialization moves too far ahead of governance. The result is a regulatory philosophy that tries to combine rapid adoption with political and institutional control.

That balancing act makes China’s AI debate important beyond China itself. Around the world, governments are still struggling to answer the same question: What exactly should AI regulation regulate? Europe has moved furthest with the EU AI Act, using a risk-based framework. The United States has taken a more fragmented route, relying on executive actions, agency intervention and sector-specific enforcement. China is hardly alone in facing that challenge, but the scale of its digital economy gives the issue unusual weight.

Major governance challenges emerging in China

The first challenge is data governance. AI agents are only as useful as the information they can access. But that same logic also raises the risk of overcollection, weak consent and misuse. In the agent era, privacy is about what AI systems can infer, combine and act upon.

The second pressure point is market structure. If AI assistants become the main interface through which users discover products, compare services, make bookings or manage daily tasks, then the power dynamics of China’s platform economy could shift again. Search, e-commerce, payments and local services could all be reshaped by whoever controls the assistant layer. That would raise competition questions about access, ranking, self-preferencing and the distribution of traffic between large platforms and smaller developers.

The third is liability. The more autonomy AI systems gain, the harder it becomes to assign responsibility when something goes wrong. Minor recommendation errors are one thing. Financial losses, privacy breaches or security failures are another. Regulators need to define clearer boundaries of accountability, especially as AI systems begin making decisions in more sensitive commercial or industrial settings.

Security is another concern. AI systems can be manipulated by malicious prompts, compromised through poisoned data or exploited through software vulnerabilities. As they spread into enterprise tools, connected devices and industrial environments, the consequences of those weaknesses become much greater. An unreliable chatbot is inconvenient. A compromised AI agent embedded in logistics, finance or critical systems is far more serious.

The larger point is that China’s AI debate is no longer just about innovation. It is about institutional readiness. The rise of increasingly autonomous systems is testing whether current legal and regulatory frameworks are robust enough for the next phase of the technology cycle.

That is why the conversation now unfolding matters. As these systems begin to act, not just generate, policymakers are being forced to confront a harder reality: the countries that lead in AI should be the ones that learn fastest how to govern them.

This commentary is the view of the writer and does not necessarily reflect the views of Bamboo Works

Vivian Toh is a Singapore-based journalist who writes about technology and co-founder of Tech Tech China, a technology media startup. You can reach her at Vivian_toh@techtechchina.com.

To subscribe to Bamboo Works free weekly newsletter, click here

Booming AI demand is driving a structural shift in the multi-layer ceramic capacitor industry, with Chinese manufacturers moving upmarket to challenge the global supply chain

By CLS Marketwatch

Multi-layer ceramic capacitors (MLCC) have long served as a cornerstone component of the electronics industry due to their massive usage and widespread applications. The sector has endured a prolonged downturn over the past two years. But the market has shown signs of recovery since the second half of last year, with spot prices rebounding, capacity utilization rates improving, and orders increasing. This resurgence essentially stems from the explosive growth of AI computing infrastructure, which has opened up fresh opportunities in the high-end market. In this context, Chinese MLCC manufacturers are entering a golden era where they are transitioning from followers to pioneers.

MLCC market turns corner

After a two-year downturn, the MLCC industry finally reached a critical turning point in the second half of 2025. Starting in October, spot prices began to rebound. The uptick covered a wide range of specifications, with spot prices for premium MLCCs rising by 10% to 15% and by 5% to 8% for mid- to low-specification products. For example, the 0402 model’s price has jumped over 10% compared to November 2025. Similarly, common models such as 0603 and 0402 have also seen price hikes of 10% to 15%.

Meanwhile, a rapid rebound in capacity utilization by manufacturers is another key indicator of market recovery. Leading manufacturer Fenghua Advanced Technology (000636. SZ) has seen its capacity utilization climb back to a high level of 80% to 90%. Such high rates are the result of strong demand, with a renowned mid- to high-end MLCC manufacturer in Hunan reporting order backlogs stretching over six months.

The current recovery is most evident in the spot market. Since major manufacturers usually enter into long-term agreements with downstream clients, many companies, including Fenghuaand Chaozhou Three-Circle (300408. SZ), have yet to formally announce price hikes. Nevertheless, the possibility that Japanese manufacturer Murata Manufacturing (6981.T) may soon raise prices due to its current short supply of high-end products means Chinese domestic manufacturers may be preparing to follow suit with their own adjustments. As such, the higher price trend is set to extend further into the contract market.

Recovery follows dual logic paths

Notably, no single factor is driving this recovery cycle, which has resulted in a distinct “dual-track” revival pattern across the higher- and lower-end markets.

Price hikes in the high-end market are primarily the result of booming demand. As AI moves from cloud computing to edge devices, demand for high-performance MLCCs in end-use applications such as AI servers, autonomous vehicles, robots, and smart glasses is growing exponentially. Data reveals that a single Nvidia GB300 server uses approximately 30,000 MLCCs – 30 times the quantity used in a typical smartphone and three times that of a single automobile. Such massive demand has pushed Japanese and Korean makers of high-end products to operate at nearly full capacity, with even some mainstream products experiencing tight supply.

Strong demand related to AI data centers is expected to keep propping up demand for advanced MLCCs. What’s more, emerging devices like lightweight smart glasses are also making extensive use of miniature MLCCs, with each piece requiring 150 to 200 units, further boosting the high-end market. Fueled by such robust demand, Chinese manufacturers of AI computing-related products could see year-on-year order increases of more than 50%.

Unlike the demand-driven premium market, rising prices for mid- to low-end MLCCs is more the result of cost pressures and supply-side self-adjustment. Since late 2024, prices of metals essential for MLCC production, such as tin, nickel, copper, and silver, have been climbing steadily, raising manufacturing costs. In the face of these inflationary pressures, previously assertive major downstream clients in sectors like home appliances have begun to accept modest price bumps of around 5% to 8%.

During the downturn of the past two years, prices for mid- to low-end MLCC products remained persistently below their costs, prompting many manufacturers to halt production to stem losses. Before 2025, most stakeholders could only return to profitability through proactive supply contraction and price negotiations. Critically, we should note that since lower-end manufacturing cannot be reallocated to higher-end production, redundant capacity cannot be repurposed. Therefore, price increases will not occur due to higher-end capacity crowding out lower-end capacity; instead, the future market is likely to become even more segmented.

Local producers ride AI wave to high-end

Immense opportunities brought by this AI-led wave is causing Chinese MLCC manufacturers to accelerate their move into the premium market. The global MLCC supply chain has long been dominated by overseas players, with the top five global suppliers collectively holding over 80% of the market.

Responding to the new boom, numerous domestic enterprises have already initiated forward-looking deployments. Fenghua and Three-Circle Group are two examples, directing capital raised through public offerings toward high-end capacitor projects since 2021. As of now, the first phase of Fenghua’s project has reached its planned capacity, while capacity at the third phase continues to ramp up and is delivering products to clients on a batch-by-batch basis. High-end MLCCs now account for over 30% of the firm's profits. Likewise, Three-Circle’s MLCC products now cover various models required for AI servers.

At the same time, Chinese manufacturers are accelerating their efforts to catch up with international standards in product R&D. Viiyong Electronic Technology achieved full coverage of automotive-grade MLCCs by 2025, and the second phase of its smart factory is scheduled to commence production in 2026, with annual capacity projected to exceed 900 billion units. Suzhou Gyz Electronic’s (688260.SH) high-capacitance MLCCs have also completed full-performance testing and IEC international standard certification, officially entering mass production. These show that Chinese manufacturers are striving for technological breakthroughs to fill gaps left by Japanese and Korean firms in certain segments of the high-end market.

The big story is that China's MLCC sector has started a new era of structural growth spurred by demand for AI computing power. The premium market is benefiting from ongoing expansion of AI infrastructure, with demand steadily growing. Meanwhile, the lower-end market has emerged from its price trough through negotiation and supply optimization. In this industrywide transformation, Chinese manufacturers are positioning themselves to become dominant players in the high-end market through forward-looking capacity planning and technological breakthroughs. Looking ahead, this crucial foundational component of the electronics industry is gaining new economic significance in the computing power era.

CLS Marketwatch provides insights and analysis on China’s industries. You can contact the author at liujingyi@cls.cn

To subscribe to Bamboo Works weekly free newsletter, click here

China’s MCN operators are moving from a focus on key influencers to content efficiency, with overseas expansion and AI reshaping the sector’s growth

LeadLeo Research Institute

China’s livestreaming e-commerce market surpassed 5 trillion yuan ($726 billion) in 2024, placing multi-channel networks (MCN) – online talent agencies that are essential links between content production and commercial monetization – at a new crossroads.

The number of MCNs has surged from hundreds to tens of thousands in recent years, while content formats have evolved from text and images to short videos and livestreaming. Influencers, typically content creators with a certain level of followers and platform influence, are also being incubated at a much faster pace.

However, this rapid expansion has brought mounting pressure, including rising traffic acquisition costs, increasing content homogenization, and intensifying competition for talent, leaving some MCNs facing the need to adjust their business and profitability models.

What exactly defines an MCN?

An MCN is an operating model that brings together multiple content creators. They provide services such as content production, distribution, marketing, and monetization, while serving as a bridge between platforms and brands. In China, the development of MCNs has been closely tied to the rise of short video and livestreaming, gradually evolving into different models, including e-commerce-focused, general content, and marketing-driven MCNs.

MCNs can be broadly understood as talent agencies for influencers. But beyond managing creators, they are also responsible for content production, traffic operations, and commercial monetization.

In practice, MCNs handle content planning and traffic distribution, while also taking on the role of monetization. For example, Joy Media manages more than 1,000 influencers, with operations spanning livestreaming e-commerce, brand partnerships, and content management. Among Joy Media’s total sales of between 100 million and 250 million yuan in any given month, the agency’s top influencer duo can generate over 100 million yuan some months, while another could contribute between 75 million and 100 million yuan, highlighting the critical role of top-tier talent and content in driving commercial conversion.

At the same time, the industry remains highly fragmented. By 2022, China had more than 24,000 MCN agencies and nearly 140 million streamer accounts. Demand continues to expand, with 87.8% of online content consumers regularly watching short videos or livestreams. However, under pressure from rising traffic, content, and talent costs, about 90% of MCNs remain unprofitable, indicating the industry is still searching for more sustainable business models.

Battle for top influencers

Competition in the industry has long centered on top influencers. A single top creator can often contribute more than 80% of an MCN’s e-commerce sales. However, their strong bargaining power and mobility also create instability for individual MCNs. In recent years, mid-tier influencers have become increasingly important. Since the second half of 2020, about 50% of the 50 fastest-growing accounts on Douyin, Kuaishou (1024.HK), and Taobao have been signed by MCNs, making mid-tier creators a key source of stable growth.

At the same time, the role of MCNs is evolving. Moving beyond influencer management and advertising placements, they are expanding into integrated marketing and long-term operations, offering brands a more comprehensive service chain. The focus of competition is shifting from “who has more influencers” to “who can consistently produce effective content and deliver conversion.”

From 2018 to 2023, China’s MCN market expanded from 25.2 billion yuan to 41.8 billion yuan, representing a 10.65% annual growth. The market is expected to reach 162 billion yuan by 2028, growing 31.12% annually between 2024 and 2028.

As competition in the domestic market intensifies, MCNs are accelerating their overseas expansion, with Southeast Asia emerging as a key destination. Data shows that in Indonesia, daily livestreaming gross merchandise value (GMV) on TikTok has reached between $3 million and $5 million, while during the 2022 “Double 11” Nov. 11 shopping festival, livestreaming GMV surged by 408%, with orders increasing by 169%.

Chinese-backed MCNs have already gained a foothold in these markets, with several ranking among top sellers. Platform subsidies and traffic incentives are attracting more players, including Be Friends Holding (1450.HK) and Make Wonder MCN. For Be Friends, influencer marketing accounts for 70% of its overseas business, while distribution-based sales contribute 10% and livestreaming operations account for 20%. However, expanding overseas is not simply a matter of replicating the domestic model. It requires building localized supply chains, developing local influencer networks, and adapting to different consumer behaviors.

A new logic of content production

Technology is reshaping the underlying operating model of MCNs. The use of AI and related tools has lowered the barriers to content production while significantly improving efficiency. For example, an evaluation report for a 1.2 million-word novel can now be completed in 15 to 30 minutes, compared with about a week of manual work, greatly accelerating content screening and planning.

At the same time, virtual humans and digital content are developing rapidly. By 2030, China’s virtual human market is expected to reach 270 billion yuan, with virtual idols, e-commerce assistants, and digital content hosts becoming increasingly mature, providing new formats and monetization channels for MCNs.

These changes are reshaping the industry’s underlying logic. As content production becomes more efficient and tools become more widely adopted, MCNs are gradually reducing their reliance on individual top influencers, with content production shifting toward standardization and scalability. The focus of competition is moving from traffic acquisition to content production capabilities and conversion efficiency.

As traditional e-commerce accelerates its shift toward content-driven commerce, demand from brands for influencer marketing is no longer limited to one-off sales but is extending to brand building and long-term operations. Leading MCNs are expanding into overseas markets, deepening their presence in vertical segments, and adopting technology to improve content efficiency, evolving from talent agencies into integrated marketing service providers.

The MCN business used to be about who could capture traffic the fastest. Now, it is increasingly about who can turn content into a consistent and scalable capability. Top influencers still matter, but they are no longer the only answer. As platform-driven growth stabilizes and content supply continues to expand, the gap between MCNs increasingly comes down to content organization, operational execution, and the ability to drive commercial conversion.

LeadLeo Research Institute is an original content platform for research on banks and companies and an innovative digital research service provider with nearly 100 senior analysts. You can contact the platform at CS@leadleo.com

This commentary is the views of the writer and does not necessarily reflect the views of Bamboo Works

To subscribe to Bamboo Works weekly free newsletter, click here

New AI video tools led by ByteDance’s Seedance 2.0 are accelerating the commercialization of content such as AI-generated anime across China’s digital media industry

By CLS Marketwatch

Early 2026 marked a historic moment for AI-generated content (AIGC) with the splashy arrival of Seedance 2.0, ByteDance's next-generation AI video generation model. The new version can simultaneously generate synchronized audio and video with cinematic quality, swiftly igniting a global AI-generated video storm with its groundbreaking capabilities. This new technology has spawned a wave of “AI-generated anime,” an emerging digital content industry with immense economic potential, which has captured the market’s attention. Driven by both technological advancements and market demand, AI-generated animes are rapidly moving from concept towards large-scale commercial application.

Dawn of a new era of AI-generated video

The AI-generated video industry has been plagued by a numerous limitations, resulting in low usability rates for those creators struggling to make videos that can meet commercial standards. Against that backdrop, the emergence of Seedance 2.0 looks like a major breakthrough. Through revolutionary capability across multiple dimensions, including better character-scene consistency, enhanced authenticity of complex physical dynamics, and improved continuity and rationality of camera movement, it provides technical support for users who demand high-quality video experiences, and shatters many previous application barriers in AI video generation.

In terms of production efficiency, Seedance 2.0 has also achieved a major leap forward. According to estimates, the usability rate of a 15-second video generated by this model can reach as high as 90%, a stark contrast to the previous industry average of 20%. This translates to substantial savings in labor and time costs during creative testing and production phases, while significantly shortening the path from concept to finished product. Taking a 90-minute long-form video project as an example, production costs could potentially be slashed from over 10,000 yuan ($1,458) via traditional methods to around just 2,000 yuan. Such major cost reduction lays the foundation for the explosion of AI-generated video content and commercial exploration of emerging content forms derived from it, such as AI-generated anime.

Powering the next generation of content

Notably, the immense potential unleashed by Seedance 2.0 is rapidly triggering a transformation across China’s digital content industry. From content production platforms to traditional publishing houses and media outlets, and further to film and television production firms, the entire industrial chain is embracing AI video generation with open arms, jockeying for position in the upcoming reshuffle.

IReader Technology's (603533.SH) “PopoMint” anime production platform has improved its production efficiency by over 100% and reduced costs by 70% compared to using traditional methods. It is currently integrating Seedance 2.0 and is expected to launch for use in the near future. Similarly, Funshion Online, a subsidiary of Shenzhen MTC (002429.SZ), has signed a collaborative agreement with Volcano Engine. Its “Chengxing Dream Factory” AI creation platform will fully integrate key capabilities of the Doubao LLM, including image and video generation. The aim is to build a full-process intelligent production loop covering scriptwriting, storyboarding, and audio-visual synthesis to create a product not easily imitated. Traditional film enterprises like Shanghai Film Co. (601595.SH) have also formed strategic partnerships with Dreamina AI to jointly develop AI-powered animes. The collaboration will also explore end-to-end innovation in “AI + film/drama” production, paving the way for a commercialized closed-loop system for AI content creation.

Regional powerhouses for AI-generated anime

Building on this foundation, a broader landscape is starting to emerge. Digital cultural and creative hubs, represented by cities like Chengdu, are leveraging their unique resources to lay the groundwork for the growth of AI-driven digital content industries such as AI-generated anime. Data shows that Chengdu alone generated more than 400 billion yuan in revenue from its core digital cultural and creative industries in 2025, marking an 8.3% year-on-year increase. New first-tier cities like Chengdu enjoy abundant talent and labor costs as much as 30% to 35% below traditional hubs, affording greater room for trial and adjustment for early-stage AI-powered content production enterprises. With Seedance 2.0, the computing cost per minute for AI videos can now be controlled to within just 100 yuan.

A typical example is Chengdu IGCAI Digital Technology, whose core business is AI-generated anime. The company's AI-generated anime segment is currently showing strong momentum. It plans to develop online training courses to cultivate skilled professionals in the AI-generated anime field, while simultaneously focusing on AI-powered film and drama production, commercial advertising creation, and the launch of AI-generated anime themed around parent-child education, family, and promoting culture going global. These developments show a comprehensive AI-generated anime industry chain is gradually taking shape.

Another example is Chengdu Perfect Park, which has made the animation industry one of its key priorities. By the end of 2025, the park had attracted over 130 enterprises, including Perfect World Animation and Chengdu Huachi Technology, with a cumulative total output value exceeding 2 billion yuan. Among their output, multiple animated works from Huachi Technology have surpassed 100 billion views across the internet. More notably, the park registered 1,000 animated copyrighted works in 2025, concluding copyright transactions worth 20 million yuan.

For now, China's AI-generated anime industry is developing quickly. While the concept only formally emerged in the spring of 2025, the industry had already entered a period of explosive growth by the third quarter. Production costs per title have dropped sharply, daily capacity has soared to hundreds of episodes, and platform traffic, revenue, and user bases have all skyrocketed. Following the Oct. 1 National Day holiday in 2025, intensive capital inflows and accelerated construction of animation production houses has signaled the formation of a full industrial chain connecting upstream funding, content IPs, and streaming platforms with downstream production houses and distribution channels. With the launch of Seedance 2.0, costs are expected to keep falling substantially, and production stability will markedly improve, laying a solid foundation for the AI-powered animation boom.

In summary, the AI-generated anime industry currently holds tremendous potential. The emergence of Seedance 2.0 has greatly lowered barriers to content creation, while cost optimization is reshaping the industry's business model. As massive capital pours in and regional ecosystems take shape, the AI-generated anime sector is accelerating from a nascent stage toward rapid expansion. As technology continues to evolve and the industrial ecosystem further matures, AI-generated anime is poised to become a key growth engine for China's digital content industry, opening new frontiers for cultural dissemination and commercialization.

CLS Marketwatch provides insights and analysis on China’s industries. You can contact the author at liujingyi@cls.cn

To subscribe to Bamboo Works weekly free newsletter, click here

From drone logistics to flying cars, the low-altitude economy is seen as a key industry in China’s 15th Five-Year Plan, though the path to commercialization remains uncertain

By Lee Shih Ta

New industrial concepts are constantly emerging in China, but only a few manage to generate excitement in both policy circles and capital markets simultaneously. The “low-altitude economy” is one of those. From drone logistics to industrial inspection, flying cars and urban air mobility, this sector — sometimes described by investors as the “electric vehicle industry of the skies” — is increasingly portrayed as the next trillion-yuan market.

In the draft outline of the 15th Five-Year Plan disclosed so far, authorities propose to “promote the healthy and orderly development of the low-altitude economy,” while encouraging exploration of applications in areas such as delivery logistics, emergency response, urban governance and cultural tourism.

In recent years, central government agencies have also released a series of policy documents, including statistical classifications and technical guidelines for the low-altitude economy, seeking to establish regulatory frameworks and technical standards for this still nascent sector. In a sense, the government is attempting to create an “industrial order” for the low-altitude economy, encouraging more companies to participate while maintaining safety and regulatory control.

However, any vision of flying taxis speeding through cities still seems difficult to imagine anytime soon. If this truly is a trillion-yuan market, where will it take off first? And what form will its development ultimately take?

Capital market reaction reflects these uncertainties to some extent. Shares of companies focused on unmanned aerial vehicles and urban air mobility have retreated in recent years after an initial wave of enthusiasm. For example, EHang Holdings (EH.US) — a company developing autonomous passenger drones — once attracted strong investor interest after gaining an airworthiness certification for its unmanned aircraft. But the stock later pulled back. This suggests investors are beginning to realize that commercialization of the low-altitude economy may take longer than originally expected.

The first challenge is airspace management. In China, low-altitude airspace has long been subject to relatively strict regulation. A new airspace management system will be needed for large-scale low-altitude flight operations to become routine, including designated air routes, flight approval mechanisms, and low-altitude traffic management systems. The repeated emphasis on airspace management and standardization frameworks in policy documents is intended precisely to address this issue.

Another challenge is infrastructure. Even if airspace restrictions are gradually eased, low-altitude flights will struggle to scale without supporting infrastructure such as takeoff and landing sites, communication and navigation systems, and regulatory platforms. Cities may need to establish large numbers of vertiports and digital monitoring systems in the future — similar to road networks and traffic signal systems in ground transportation. Such infrastructure will form the operational foundation of the low-altitude economy.

Starting from the ground

Based on recent pilot programs, the first commercial applications in the low-altitude economy are unlikely to be flying taxis, but rather simpler, more practical use cases. In Shenzhen, for example, drone deliveries have already begun operating on a routine basis. Meituan (3690.HK) and SF Express (6936.HK; 002352.SZ) — through its Fengyi Technology drone subsidiary —have launched low-altitude delivery routes in certain areas, allowing takeout meals or medical supplies to be delivered within just over 10 minutes. These delivery routes typically cover short distances with fixed flight paths, making regulatory oversight and safety management easier. As a result, they are considered among the earliest commercial applications of the low-altitude economy in China.

Beyond deliveries, energy and infrastructure inspection is another important application for low-altitude flight. Drones are gradually replacing manual inspections across China’s extensive networks of power transmission lines and oil and gas pipelines. State Grid and China Southern Power Grid have increasingly deployed drones to inspect transmission lines in recent years, reducing the risks associated with high-altitude work for people while improving efficiency. In the mining and surveying sectors, drones are also being used to monitor open-pit mines and conduct geographic mapping.

Low-altitude technologies are also beginning to play a role in public safety. Some cities have tested firefighting drones for tasks such as reconnaissance in high-rise fires or for monitoring wildfires. In agriculture, crop-protection drones are widely used for seeding and pesticide spraying, with companies such as XAG currently leading China’s agricultural drone market.

At the same time, several companies are pushing forward with passenger aviation applications. EHang is exploring urban air mobility and low-altitude tourism routes, while XPeng (9868.HK; XPEV.US) — through its flying car subsidiary XPeng AeroHT — is developing a flying vehicle that it plans to bring into mass production in the coming years. Meanwhile, aerospace equipment manufacturers such as AVIC Aerospace (2357.HK) and drone maker Aerospace CH UAV (002389.SZ) are also widely viewed as companies that could benefit from the development of the low-altitude economy.

Beyond aircraft manufacturers, the industry’s potential supply chain also includes makers of navigation and communication systems, flight-control software, sensors, power semiconductors, and data dispatch platforms. As low-altitude flight activity increases, air traffic management systems and cloud-based operational platforms will become increasingly important. In many ways, the low-altitude economy concept seeks to integrate these previously fragmented application scenarios into a unified industrial framework while also promoting new business models such as urban logistics and aerial transportation.

From an industry development perspective, commercialization of the low-altitude economy will likely unfold in stages. In the near term, applications such as delivery logistics, industrial inspection, and public services may generate the most stable initial demand. In the medium term, sectors such as tourism and short-distance passenger aviation may begin to emerge. Large-scale urban flying taxis, meanwhile, will likely require much longer to materialize.

The low-altitude economy may indeed become a new trillion-yuan industry one day. But its development path will likely be more complex than investors initially imagined. Before low-altitude aircraft can truly fill city skies, this highly anticipated sector still needs time to evolve from a concept into a fully developed industry.

This is part 5 in a 5 part series. To read previous parts, click on the links below:

15th Five-Year Plan: Solar and property wait for the next policy tide

15th Five-Year Plan: Opportunities and trade-offs under technological self-reliance

15th Five-Year Plan: Who is leading offshore listings?

15th Five-Year Plan: Inside China’s push to become a financial powerhouse

Lee Shih Ta is an editor at Bamboo Works.

You can contact him at shihtalee@thebambooworks.com

To subscribe to Bamboo Works weekly free newsletter, click here

The narrowing of subsidies has yet to suppress buying, and instead consumers are gravitating towards products that offer more value-added features like AI capabilities

By CLS Marketwatch

China's national consumer subsidy policy is heading into a new cycle in 2026. An initial allocation of 62.5 billion yuan ($9.1 billion) was smoothly channeled to retail terminals across the country at the start of this year, though coverage under the policy has narrowed notably. Nevertheless, during the peak Lunar New Year consumption season, purchases of 3C electronics and major home appliances remained active, with a noticeable shift in consumer preferences towards mid- to high-end products.

At the same time, supply-side pressures have mounted as rising raw material costs led to price adjustments for several home appliance categories. The combined dynamics of policy adjustments and evolving consumer behavior are adding notable new trends to China's 3C electronics and home appliance markets.

From broad stimulus to selective upgrades

The first round of subsidies for 2026 was distributed ahead of the Lunar New Year holiday. Notably, this round broadly covered a diverse range of 3C products, including mobile phones, tablets, and smart wearables from mainstream brands such as Huawei, Xiaomi (1810.HK) and Apple (AAPL.US). The policy adjusted the subsidy rate to 15% with a per-item cap of 500 yuan, and applied only to products priced below 6,000 yuan. It also imposed quantity and monetary limits on subsidies for certain home appliances, while removing multiple kitchen appliance subcategories. This narrowing of covered categories aims to create a buffer for smoother future phase-outs.

The narrower coverage hasn’t cooled market enthusiasm. In the first month after the new policy’s rollout, Suning.com, China's premier electronics and home appliance retailer, saw a remarkable jump in purchasing of major appliances. Sales of air conditioners, refrigerators, washing machines and TVs all surged by over 90% month-on-month. Mid- to high-end smartphones emerged as the primary consumption focus within the 3C category, while purchasing power for emerging tech products like tablets and smart wristbands rose sharply. These figures show the policy adjustments have not dampened demand, but rather spurred more concentrated purchasing within subsidy parameters.

Chasing quality

Crucially, the ongoing rewriting of policy incentives is reshaping consumer logic. In China's 3C and home appliance markets, consumers are accelerating their shift from basic renewal to proactive upgrades for superior quality. Consumers are showing growing interest in larger appliances, with ultra-large refrigerators exceeding 550 liters and giant-screen TVs over 85 inches becoming top choices for replacing older models. Meanwhile, demand is surging for appliances and 3C products that enhance efficiency and assist in scenarios like reducing household chores, hosting gatherings, and tech gifting. Among these, certain categories like voice-controlled air conditioners, microwave-steam oven combinations, and superior AI imaging smartphones are particularly popular.

It’s interesting to note that the subsidy cap has actually spurred new consumer behavior, with buyers seeking the most powerful configurations possible within the 6,000 yuan limit. This mindset has fueled a substantial rise in sales of mid- to high-end models, with devices priced between 3,500 and 6,000 yuan now accounting for over 70% of sales, bringing them into the mainstream. The Huawei Mate 80 standard edition, precisely positioned within the subsidy range, has been consistently sold out, while the higher-tier Pro series, excluded from subsidies due to its price, has experienced a starkly different sales trajectory.

But strong demand hasn’t addressed underlying concerns on the supply side. Since 2025, prices for commodities like copper and aluminum, as well as memory chips, have continued to climb, with this cost pressure being passed down to the retail end.

The market has responded with comprehensive price adjustments, with minor price tweaks already appearing in certain kitchen appliance categories such as ovens. Similarly, Aux Electric (2580.HK), a leading brand in air conditioning, has confirmed a 6% to 10% price hike starting in March. Similarly, the Xiaomi 17 Ultra, released in late 2025, saw its launch price raised due to rising memory costs. To balance costs and profits, top manufacturers are increasingly channeling R&D and production resources toward high-performance, high-margin products. AI-powered features, for instance, are being prioritized for mid- to high-end models.

Enduring mark of temporary policy

The deeper significance of the ongoing national subsidy policy lies not only in stimulating short-term sales, but also in creating momentum for industrial upgrades. Data indicates consumer willingness to pay for cutting-edge technologies is gradually moving beyond mere subsidy dependency. A growing number of consumers are seeking full-feature functions and opting for top-tier home appliances in a single purchase.

Since implementation of the subsidy program’s latest round, sales of home appliances equipped with on-device AI capabilities have approached 60% of the total. Highly intelligent products, such as TVs powered by large language models (LLMs), washing machines capable of identifying different fabric types, air conditioners with active temperature control, and refrigerators featuring automatic freshness preservation, are becoming increasingly prominent and are becoming sales leaders within their categories.

However, it’s worth noting that as subsidy policies become more routine, their stimulating effects will inevitably diminish. As this round of home appliance and electronics renewals nears its end, the underlying slump in consumption will inevitably resurface. Therefore, the true path forward for enterprises lies in proactive industrial upgrades. This involves identifying and targeting specific market segments, and developing sophisticated operational strategies to secure user loyalty. Within this context, a pathway with good potential is leveraging an "AI + home appliance" strategy to drive product innovation to deliver greater application value across a diverse range of consumer scenarios.

Looking forward, the national subsidy policy will remain the most direct tool for lifting consumer spending in the short term. The narrowing of subsidies has yet to suppress spending, and instead consumers are gravitating towards products within program guidelines that offer higher specifications, longer lifecycles, and better solutions to pain points. This provides a key signal that Chinese consumers are increasingly willing to pay for advanced technology.

Nonetheless, the marginal benefits of the national subsidy policy are diminishing. Therefore, an immediate priority for businesses should be shifting their focus from promotional discounts to developing higher-value products and refining their operational strategies. National subsidies will eventually be phased out. But they have played a powerful role in driving a technology-led value reconstruction within China's 3C and home appliance markets, and in pushing consumer mindsets toward larger products with more advanced features.

CLS Marketwatch provides insights and analysis on China's industries. You can contact the author at liujingyi@cls.cn

To subscribe to Bamboo Works weekly free newsletter, click here

A debut of the term ‘financial powerhouse’ in China’s new five-year plan will push the financial system to the front lines of industrial transformation

By Lee Shih Ta

China’s 15th Five-Year Plan is not just another five-year roadmap, but instead represents a critical starting point toward achieving basic modernization by 2035. Against a backdrop of rising geopolitical tensions, intensifying technology competition and a slowing domestic economy, policy priorities are shifting from stabilizing growth to reshaping the development model. Stock markets in Shanghai, Shenzhen and Hong Kong have all rebounded over the last year, with authorities repeatedly signaling support for market stability. Investors, however, are increasingly focused on what role the financial system will play in the latest transition.

China’s “Recommendations for the 15th Five-Year Plan,” released late last year, included the goal of building a “financial powerhouse” for the first time. The document also called for increasing the share of direct financing, strengthening the multi-tier capital market system, and developing technology and digital finance. This echoes President Xi Jinping’s repeated calls for finance to serve the real economy and avoid drifting away from productive activities, signaling a shift from a system characterized by short-term policy adjustments toward a broader restructuring of the financial system itself.

In fact, “financial support for the real economy” is nothing new in China. It has been repeatedly stressed since the Central Financial Work Conference in 2017, mainly through tools such as reserve requirement cuts, relending programs and targeted credit to stabilize growth and contain risks. Yet the overall financial structure has never undergone a fundamental change.

Tian Xuan, vice dean of Tsinghua University’s PBC School of Finance, noted that as of June 2025, direct financing accounted for only about 31.1% of China’s total social financing, while bank assets still represented more than 90% of the financial system’s total assets. In some regions, the market-based fundraising share of government-guided funds is even below 20%.

In his view, this shows that China remains a highly credit-driven financial system, with limited capacity for markets to truly bear risk and price innovation.

The power of markets

As AI, semiconductors, advanced equipment and new materials become policy priorities, capital needs are clearly changing. In the 15th Five-Year Plan recommendations, references to “technology,” “innovation” and “new quality productive forces” appear frequently, alongside the launch of an “AI+” initiative. This places technological upgrading at the core of national strategy and shifts finance from simply supplying funds to building capital mechanisms for long-cycle innovation, signaling that market mechanisms and regulatory frameworks will take on greater importance, rather than relying solely on administrative guidance.

So, what changes may lie ahead?

First, with increasing the share of direct financing now an explicit policy direction, new capital is more likely to flow into the real economy through the equity and bond markets. This should improve the equity financing environment and allow capital markets to play a more central role in economic transformation.

Next comes a shift in valuation logic. As “patient capital” and improved venture investment and M&A mechanisms move onto the policy agenda, markets may place greater emphasis on technological barriers and scaling capabilities when assessing companies. Short-term profits will no longer be the sole benchmark, while industry positioning and alignment with policy priorities will increasingly become positive factors. These changes may not immediately show up in share prices, but they are likely to shape medium- and long-term capital allocation.

At the same time, refinancing and secondary share offerings may gradually become the norm. Equity markets are being tasked with funding industrial upgrading, and rights issues and private placements with clearly defined investment purposes are expected to receive greater regulatory support.

For investors, markets may increasingly accept a model in which companies grow while raising capital, with evaluation shifting from short-term earnings per share to whether fundraising aligns with broader industry priorities.

On the bond side, demand from technology firms for convertible bonds and corporate credit is expected to rise, while sectors tied to new quality productive forces may benefit from cheap financing. Credit pricing is likely to become increasingly linked to industry characteristics, rather than relying solely on financial leverage.

What bears watching is that as technology and new quality productive forces move to the center of policy priorities, financial markets in Shanghai, Shenzhen and Hong Kong are likely to become more accommodating toward hard-tech companies, including those not yet profitable but with technological moats and commercialization potential. Listing standards may not be broadly loosened, but approval signals could become clearer for sectors such as AI, advanced equipment, semiconductor tools and new materials.

Regulatory changes also warrant close attention

The 15th Five-Year Plan recommendations also call for “comprehensively strengthening financial regulation,” meaning that as capital markets expand, requirements for disclosure, use of funds and internal controls are likely to tighten in parallel. Going forward, how listed companies use the capital they raise may matter more than whether they can raise it, helping curb past practices where some firms strayed from their core businesses or left funds idle or diverted them into non-productive areas.

Financial regulators have also repeatedly stated that industrial M&A should serve as a key tool for improving the efficiency of direct financing and driving structural adjustment, encouraging listed companies to pursue consolidation around their core businesses. Behind this push lies the reality of difficult venture capital exits and fractured financing chains for technology firms.

Against this backdrop, developing “patient capital” and improving M&A restructuring mechanisms are seen as two sides of the same institutional framework — the former supplying long-term funding, and the latter opening exit channels. If supporting measures are implemented, capital may move beyond venture investment and IPOs to form a fuller cycle to support venture funding, public listing, expansion through refinancing, and exit via consolidation.

Still, becoming a financial powerhouse is no cure-all. Investment, consumption and exports remain under pressure, and IMF Managing Director Kristalina Georgieva has noted that China needs to accelerate its shift toward domestic demand and consumption, underscoring that financial reform must be coordinated with industrial and demand-side policies.

From a broader perspective, the drive toward becoming a “financial powerhouse” represents a long-term effort to realign the financial system with China’s industrial structure, with capital markets set to take on an expanded role in the process.

This is part 4 in a 5 part series. To read previous parts, click on the links below:

15th Five-Year Plan: Solar and property wait for the next policy tide

15th Five-Year Plan: Opportunities and trade-offs under technological self-reliance

15th Five-Year Plan: Who is leading offshore listings?

Lee Shih Ta is an editor at Bamboo Works.

You can contact him at shihtalee@thebambooworks.com

To subscribe to Bamboo Works weekly free newsletter, click here

As the outlines of China’s 15th Five-Year Plan take shape, offshore listing trends are becoming more structured, with certain industries moving ahead of others

By Lee Shih Ta

After topping the world for fundraising last year, Hong Kong’s IPO market has extended its momentum into early 2026. According to data compiled by Bloomberg, IPO fundraising in January alone reached $5 billion, with 13 companies listing on the city’s stock exchange. The new listings spanned AI chip designers and other technology-driven firms. As China prepares to enter its 15th Five-Year Plan policy cycle, this burst of activity offers a telling preview of where capital markets may be heading.

Even before the new cycle kicks off, the broad industrial direction is increasingly clear. Technological self-reliance, energy transition, advanced manufacturing and new forms of productivity have repeatedly been emphasized by both policy signals and capital allocation. Against this backdrop, offshore listings are increasingly being viewed as a capital strategy aligned with industrial development rhythms — particularly for sectors characterized by front-loaded funding needs, intense competition and a high degree of globalization.

New energy: Well positioned for offshore listings

Among all sectors, new energy remains the most established candidate for offshore listings. Over the past year, Hong Kong’s market has repeatedly demonstrated its capacity to absorb battery manufacturers, energy equipment suppliers and system-level players. Large fundraising amounts, strong participation from international investors and relatively stable post-listing trading have created clear market benchmarks.

Among domestically listed companies yet to complete offshore listings, power and energy storage battery maker EVE Energy (300014.SZ) and energy storage system provider HyperStrong (688411.SH) stand out as strong Hong Kong listing candidates. Their order books and capital needs are already highly internationalized, while offshore markets can offer longer funding cycles and a more stable investor base to support overseas expansion.

AI and computing power: Fast-track to listings

AI and computing power represent two of the most capital-intensive focus areas under the 15th Five-Year framework. Chip development, computing architecture and software-hardware integration require substantial capital expenditure by companies within compressed timeframes.

AI accelerator developer Enflame Technology and edge-AI chipmaker Aixin Intelligence have both entered the Hong Kong IPO pipeline. GPU and general-purpose computing firms Moore Threads (688795.SH) and MetaX (688802.SH) have already completed A-share listings in Shanghai. Domestic investors have been willing to price in policy certainty, treating indigenous computing power as a strategic asset. However, as these companies expand their ecosystems and customer bases internationally, offshore listings may evolve as mid- to long-term strategic options — serving less as financing tools and more as gateways to overseas investors and industrial partners.

Semiconductors and advanced manufacturing: Valuations taking shape

Conditions for offshore listings in semiconductors and advanced manufacturing are gradually improving. As more technology-oriented companies have raised funds in recent years, investor understanding of these sectors has deepened and valuation frameworks have become clearer.

Among firms yet to pursue Hong Kong listings, automotive-grade and high-performance processor developer Siengine Technology, along with foundry and specialty process platform United Nova Technology (688469.SH), are widely seen as meeting the prerequisites for overseas capital markets. Their customer bases are already integrated into global supply chains, and offshore listings could help lower long-term capital costs while enhancing international visibility.

Intelligent driving and robotics: Preparing for mass production

The intelligent driving and robotics sectors are approaching a critical commercial inflection point. As more companies complete offshore listings, valuation frameworks across the supply chain are becoming clearer.

Among companies yet to list overseas, autonomous driving solution provider DeepRoute.ai and industrial robot and motion control manufacturer Estun Automation (002747.SZ) are regarded as likely candidates for offshore listings. DeepRoute.ai underwent several equity changes in the second half of last year, widely seen as preparatory steps toward an overseas listing. Estun, which filed for a Hong Kong listing last June and renewed its application in January, is seeking funds to expand its global capacity, pursue acquisitions and reduce debt.

A recent report by China Merchants Securities International forecasts that China’s humanoid robotics sector will begin accelerating in 2026, supported by improvements in large-model generalization, hardware maturity and falling costs. As mass production and overseas deliveries increase, companies in this space are expected to face rising demands for scale and stable funding — further strengthening the case for offshore listings.

Spin-offs: An extension strategy for large conglomerates

Under the 15th Five-Year cycle, spinoff listings are increasingly emerging as a strategic extension for large technology groups — particularly where core technology units align closely with national priorities but have growth trajectories and capital needs divergent from their parents. The aim is to give computing, chip and critical technology platforms clearer industrial identities and connect them directly with long-term capital and specialist investors.

A concrete early example comes from Baidu (BIDU.US; 9888.HK), whose Kunlunxin AI chip unit has already been spun off and is widely viewed as possessing independent capitalization potential. This reflects a broader trend among major technology groups to pave independent development paths for core technology assets. Within the same context, Alibaba (BABA.US; 9988.HK) is also reportedly considering an independent listing for its T-Head chip arm.

As the 15th Five-Year Plan continues to emphasize technological self-reliance and the role of finance in serving the real economy, such spinoffs not only help unlock hidden value within conglomerates but also establish funding and valuation structures better aligned with the development stages of critical technology businesses.

Overall, sectors such as new energy, AI computing power and semiconductors, intelligent driving and robotics are showing strong alignment between their funding requirements, technological trajectories and global market exposure. As the contours of the 15th Five-Year Plan become clearer, the appeal of offshore listings remains robust. Amid strong trading activity and buoyant fundraising in Hong Kong, an increasing number of companies are reassessing offshore listing opportunities — a trend not confined to any single sector. That said, industries most closely aligned with priorities in the 15th Five-Year Plan will continue to enjoy stronger valuations and greater access to capital, making them especially compelling to investors.

This is part 3 in a 5 part series. To read previous parts, click on the links below:

15th Five-Year Plan: Solar and property wait for the next policy tide

15th Five-Year Plan: Opportunities and trade-offs under technological self-reliance

Lee Shih Ta is an editor at Bamboo Works.

You can contact him at shihtalee@thebambooworks.com

To subscribe to Bamboo Works weekly free newsletter, click here

As AI and semiconductors become national priorities, markets are already positioning for how China’s 15th Five-Year Plan will reshape its industrial landscape

By Lee Shih Ta

Not every industry filled prioritized for development in China ultimately survives tensions between policy support and commercial reality. Some sectors gradually lose momentum due to weakening demand, rising costs, or flawed business models, eventually fading from both policy narratives and capital allocation priorities.

With such a reality as backdrop, a key question is not only which industries will be written into policy documents as attention shifts to China’s new 15th Five-Year Plan, but which ones are actually capable of making it to the finish line.

AI self-reliance: Computing power and chips take the lead

In the current policy environment, AI is no longer being positioned as a standalone growth industry. Instead, it is repeatedly framed as a capability that must be deeply integrated with manufacturing, energy systems and the real economy, with an emphasis on developing industry- and industrial-grade large models.

This language effectively elevates demand for computing power from a discretionary corporate choice to an institutionalized form of foundational demand.

Optical communications have emerged as the most representative beneficiary. Over the past year, demand from AI data centers for 800G and 1.6T high-speed optical modules has driven leading companies in the segment to outperform the broader market.

Shares of Zhongji Innolight (300308.SZ) have surged 389.7% over the past year, with its market capitalization approaching 700 billion yuan ($100 billion), while Eoptolink Technology (300502.SZ) jumped 320%. That’s made optical components one of the strongest-performing segments in the AI supply chain among companies listed on China’s domestic markets in Shanghai and Shenzhen.

In semiconductors, policy priorities have similarly shifted from “comprehensive substitution” toward “deployable capability.”

Demand for mature-node processes as well as industrial and automotive chips has put SMIC (0981.HK; 688981.SH) back onto investor radars. Its shares have risen about 80% over the past year, keeping it among the most valuable semiconductor names in China.

The AI segment, by contrast, has shown far sharper divergence. Advanced chipmaker Cambricon Technologies (688256.SH) has gained more than 110% over the past year, driven by demand for industry-specific models and inference workloads — an indication that capital markets are increasingly pricing in the strategic role of computing power.

Biren Technology (6082.HK), recently listed in Hong Kong as China’s first domestically developed GPU company to go public, saw its IPO oversubscribed by about 2,363 times. Its shares are up roughly 7% since the listing, with investors remaining broadly optimistic about its prospects.

Large models as platforms: Valuation divergence

As model capabilities gradually converge, market attention has shifted from asking which model is the strongest to identifying who can offer a scalable AI platform. In China, competition in large models is moving away from startups toward major internet platforms.

Baidu (BIDU.US; 9888.HK), with its Ernie model, and Alibaba (BABA.US; 9988.HK), with Tongyi Qianwen, continue to enjoy high technical visibility, but investors are clearly less impressed. Baidu’s share price and valuation multiples remain depressed, reflecting investor skepticism over whether AI can meaningfully lift its core advertising and cloud businesses. Alibaba’s shares have rebounded, but its valuation recovery remains constrained by limited visibility into cloud profitability.

By contrast, Tencent (0700.HK) has delivered the most stable market performance. Rather than emphasizing model rankings, Tencent has embedded AI capabilities across its gaming, advertising and enterprise services. Its shares have risen more than 50% over the past year, maintaining its position as China’s most valuable technology company.

This suggests investors are more willing to pay for AI narratives that clearly enhance cash flow quality.

The rise of AI agents: From deployment to monetization

More telling than the models themselves is the gradual emergence of AI agents. Unlike general-purpose conversational models, agents are designed to operate continuously within specific scenarios, proactively handling research, analysis, content generation and workflow coordination. This marks a shift from AI as a support tool to AI as an embedded component of real-world workflows.

This trend has already begun to materialize in the Chinese market.

ByteDance’s Doubao, for example, is being integrated into content creation, data organization, and enterprise use cases. The goal is to internalize AI capabilities into reusable work modules, serving as foundational tools to boost organizational and content production efficiency.

This contrast also highlights differing AI development paths between China and the U.S. American tech giants continue to double down on frontier model capabilities, while China places greater emphasis on rapid deployment and monetization.

In late December, Meta (META.US) announced a deal to acquire Manus, a Chinese AI agent company. Such a move underscores the practical advantages of China’s AI ecosystem, which prioritizes rapid deployment and early monetization, while also opening up new possibilities for the future development of domestic AI agent companies.

Robotics and power systems: Efficiency as a structural necessity

If AI computing power and large models form the foundation of the digital economy, then robotics and power systems are increasingly becoming the infrastructure that supports the current wave of upgrades. Unlike earlier phases that emphasized technological demonstrations, both policymakers and markets are now focused on which systems can generate measurable and replicable productivity gains in real-world applications.

In the robotics sector, a key development has been the rise of the robotics-as-a-service (RaaS) model. By adopting leasing arrangements or usage-based pricing, RaaS converts large upfront capital expenditures into predictable operating expenses, significantly lowering barriers to adoption for enterprises.

According to Counterpoint Research, global installations of humanoid robots are expected to reach approximately 16,000 units in 2025, with Chinese companies occupying the top three positions by market share — namely Zhiyuan Robotics, Unitree Robotics, and Ubtech (9880.HK). Cumulative installations are projected to exceed 100,000 units by 2027, with primary applications in logistics, manufacturing and the automotive sector — scenarios where RaaS models are most likely to generate cash flow.

Among these companies, Ubtech’s stock price has risen by more than 135% over the past year. Other players, such as industrial automation firms Inovance Technology (300124.SH) and Estun Automation (002747.SZ), have also gained roughly 30% over the same period.

Some market observers caution that valuation bubble risks in robotics and embodied intelligence could emerge even faster than in AI itself. This serves as a reminder that beyond high-growth expectations, investors should pay close attention to companies’ actual commercialization progress and cash flow performance, in order to avoid potential corrections driven by overly inflated valuations.

Power and energy systems are being repriced alongside robotics. The expansion of AI computing, data centers and automated equipment means that electricity is no longer defined solely by a company’s ability a to provide more volume, but increasingly by requirements for stability, flexibility, and energy efficiency. This shift has brought a new group of privately owned power equipment and energy storage companies back onto investor radars.

Over the past year, share prices for power electronics and energy storage companies have diverged significantly, though the overall trend has improved compared with 2023. Shares of Sungrow Power Supply (300274.SZ) more than double at one point, supported by a recovery in demand for energy storage inverters and large-scale power plant solutions, prompting the market to reassess its transition from a solar equipment manufacturer to an energy management platform provider.

Other inverter manufacturers, including Ginlong Technologies (300763.SZ) and GoodWe Technologies (688390.SH), have also seen gradual valuation recoveries as penetration of overseas energy storage and commercial and industrial applications increases, with share gains ranging from 40% to 80%.

This is part 2 in a 5 part series. To read previous parts, click on the links below:

15th Five-Year Plan: Solar and property wait for the next policy tide

Lee Shih Ta is an editor at Bamboo Works.

You can contact him at shihtalee@thebambooworks.com

To subscribe to Bamboo Works weekly free newsletter, click here

As China shifts from the 14th to the 15th Five-Year Plan, solar tells a story of swelling under policy momentum before a pullback, while property experienced a forced, sharp slowdown

By Lee Shih Ta

In China’s policy framework, five-year plans are less operating manuals for companies and function more as a continually recalibrated outlays for the pace for national development. They reset priorities and resource allocation at each stage, but don’t determine corporate success or failure. Even companies that align perfectly with policy direction can face outcomes far beyond their expectation as regulation, capital flows and competitive dynamics continually shift.

During the recently concluded 14th Five-Year Plan period from 2021 to 2025, solar energy and real estate became the clearest examples of industries that “got the beginning right, but misjudged the ending.” Solar surged under China’s dual-carbon agenda, only to fall into overcapacity and price wars within a few years. Property, meanwhile, cooled rapidly under policies forcing deleveraging and risk-control, dragging the entire industry into a prolonged adjustment. In both cases, the industries’ trajectories were fundamentally altered.

Solar: From acceleration to stall

At the start of the 14th Five-Year Plan, solar power was assigned a clear policy mission: advancing the energy transition, lowering power generation costs, and building a secure, self-sufficient renewable supply chain. Under policy language such as “accelerated development,” “large-scale deployment” and “build whenever feasible,” installed capacity became the core performance metric.

As a result, China’s solar installations surged. New capacity expanded from 48 GW in 2020 to 216 GW in 2023, and exceeded 260 GW in 2024 — a more than fivefold increase in just four years. China cemented its position as the world’s largest solar market, effectively completing the 14th Five-Year Plan’s core task of scaling up capacity.

Yet this explosion in installations failed to translate into sustained industry returns. Capacity across polysilicon, cells and modules was released almost simultaneously. Polysilicon prices collapsed from over 300 yuan ($42.91) per kilogram at their 2022 peak to below 70 yuan by the end of 2024, a decline of more than 70%, with cell and module prices following the same trajectory. Even growing global demand proved insufficient to absorb the simultaneous release of so much capacity, pushing the industry into a “volume-for-price” competition trap.

The deterioration in profitability quickly showed up in company stocks. Over the past five years, Longi Green Energy (601012.SH) has lost about 65% of its market value, Tongwei (600438.SH) nearly 50%, JinkoSolar (688223.SH; JKS.US) more than 55%, and Trina Solar (688599.SH) around 25%. Despite generally rising shipments and revenue, valuation multiples for industry leaders have fallen sharply, reflecting investors’ reassessment of whether solar can still deliver outsized returns after completing its policy-driven expansion.

This shift is also evident in changing policy language. While the 14th Five-Year Plan emphasized “scale,” forward-looking signals for the 15th Five-Year Plan increasingly stress “high-quality development,” “orderly construction,” “grid integration,” and “market mechanisms.” Solar is moving from a phase of mandatory expansion to one of selective consolidation.

Property: Deleveraging hits the entire chain

If solar’s slowdown was a byproduct of encouragement, property’s reversal was the result of a clampdown on risks in a concentrated burst. Early in the 14th Five-Year Plan, the rollout of the “three red lines,” tighter financing rules and other deleveraging measures — compounded by the pandemic — sent core indicators into rapid decline.

According to official data, national housing sales area fell from 1.79 billion square meters in 2021 to about 973 million square meters in 2024, a drop of roughly 45%. In the first 11 months of 2025, sales slipped further to around 787 million square meters. Property investment and new construction starts collapsed in tandem and have yet to bottom out.

Under pressure from tightening funding and falling sales, major developers faced similar outcomes of different magnitude — Vanke (2202.HK; 000002.SZ) entered a period of financial strain, while Evergrande and Country Garden (2007.HK) fell into full-blown crises. Evergrande was forced to delist, and its founder Hui Ka Yan was detained. By some estimates, around 77 Chinese developers have defaulted over the past five years. Surviving firms have been forced to prioritize debt repayment and project delivery, sharply cutting land purchases and new investment — undermining demand across the entire supply chain.

Upstream building materials were hit first. Cement and flat glass output has declined for three consecutive years since 2021. Cement production fell from 2.36 billion tons in 2021 to 1.83 billion tons in 2024, with July 2024 output at just 146 million tons — the lowest monthly figure since 2009. Prices weakened alongside volumes, and even relatively strong players such as China National Building Material (3323.HK) struggled to defend their margins.

Demand also cooled for midstream firms engaged in construction and decoration. Slower residential construction directly affected order flows at China Lesso (2128.HK), while companies like Kuka Home (603816.SH), once buoyed by the “post-property cycle plus consumption upgrade” narrative, lost key growth drivers as new housing supply slowed, compounded by weaker consumer demand and overseas expansion challenges. Over the past five years, Kuka Home’s shares have fallen more than 41%, while China Lesso is down nearly 62%.

Even the relatively defensive property management sector wasn’t immune. As developer deliveries declined, growth in newly managed floor area slowed sharply. Country Garden Services (6098.HK), heavily reliant on its developer parent’s pipeline, saw its scale-driven growth model questioned, with valuation logic shifting from expansion to survival. Its shares have collapsed by more than 90% over five years.

When policy sets the rhythm

What solar and property ultimately share is not a strategic misjudgment, but their position within a tightly choreographed policy system. When that system demands speed, scale and stability — and expects stage-specific objectives to be met within a fixed timeframe — companies have little choice but to keep pace. Industries elevated to center stage in one cycle may be required to slow, or even step back, in the next. Being lifted by the wave, and then left adjusting as it recedes, is part of the system’s rhythm. For solar and property, the task now is simply to regain balance and wait for the next turn of the policy tide.

Looking ahead to the 15th Five-Year Plan, the solar sector is unlikely to be tasked with simply expanding scale, with policy attention shifting instead toward grid integration, power market reform and energy-storage deployment, bringing competition back to cost and efficiency. As for property, the policy focus is more likely to remain on risk containment and order restoration rather than reigniting investment and sales growth, pointing to a new normal of a smaller, lower-leverage and slower-turnover industry. For markets, the key question during the 15th Five-Year Plan period is not whether a new round of policy stimulus will emerge, but how institutional constraints will redefine the boundaries of survival for these two sectors.

This is part 1 in a 5 part series

Lee Shih Ta is an editor at Bamboo Works.

You can contact him at shihtalee@thebambooworks.com

To subscribe to Bamboo Works weekly free newsletter, click here

After years of controlling the apps on their products, smartphone brands are on notice that China’s market regulator won’t tolerate unfair practices like app blocking and traffic hijacking

By Vivian Toh

In China’s huge yet semi-walled smartphone market, a long-simmering feud between handset makers and software developers is finally boiling over.

For years, domestic Android phone brands have leveraged their control of devices and app distribution to tilt the playing field in their favor. They’ve built their own app stores and services on top of Android – effectively replicating Apple’s closed ecosystem model – and positioned themselves as gatekeepers between developers and China’s hundreds of millions of users. This entrenched hardware-versus-software conflict stayed largely out of the spotlight, but now Chinese regulators are stepping in, signaling that the era of unchecked platform power may be nearing its end.

Hardware-software feud: From “hardcore alliance” to 50% app tax

As Android’s global services (like Google Play) never officially took root in China, domestic manufacturers filled the void with their own app marketplaces, pre-installed apps, and custom interfaces. This arrangement gave phone makers leverage well beyond the hardware itself. Control over operating systems – from default apps and permissions to distribution channels – became a powerful commercial tool. Accessing China’s massive Android user base often meant software firms had to accept onerous terms set by the phone makers. Over time, that leverage solidified into what developers describe as a de facto “hardcore alliance” of top phone makers jointly dictating app market economics.

One outcome was sky-high commission rates. Domestic Android app stores steadily drove up their revenue share on app sales and in-app purchases, in some cases demanding 50% of transaction value – far above the 30% cut that Apple and Google take globally. At the same time, developers accused handset makers of using technical tactics to favor their own services and hobble rivals’ apps. In essence, China’s Android giants turned the smartphone into a gatekeeper walled garden – one that they collectively profited from, at developers’ expense.

The consequences have been tangible. NetEase (NTES.US; 9899.HK), one of China’s biggest game publishers, has openly rebelled against these practices. After voicing discontent with app store commissions, NetEase began withdrawing its hit games from certain Android app stores in protest. Tencent (0700.HK) made a similar move this year: in June, it yanked the highly anticipated Dungeon & Fighter Mobile off Huawei, Oppo and Vivo’s stores after revenue-sharing talks broke down. These acts of defiance by industry giants underscore how contentious the 50% cut and restrictive policies had become.

Phone manufacturers haven’t relied only on high commissions to lock in their advantage. They’ve also engineered their systems to actively discourage users from downloading apps via any outside source. Attempts to install an app from a third-party website or alternative app store trigger a gauntlet of warnings and roadblocks. Meanwhile, installing that very same app through the official store is seamless. It’s a psychological and technical barrier explicitly designed to deter deviation.

Regulators step in: New rules of fair play

It was only a matter of time before Beijing’s regulators took notice of these unfair tactics. In late November, China’s State Administration for Market Regulation (SAMR) convened major smartphone companies in Shenzhen to issue compliance guidance under the Anti-Unfair Competition Law. The meeting – notable for its stern tone – called out specific behaviors that officials deemed problematic. Rather than speaking in generalities, SAMR officials explicitly denounced “irrational competition” in the mobile sector, highlighting traffic hijacking, forced redirects, and malicious incompatibility as practices that disrupt market order and harm consumers’ rights. In regulatory terms, Beijing is drawing a bright red line: tactics long used by Chinese Android vendors to sideline competitors are now being framed as illegal anti-competitive conduct, not acceptable industry norms.

Crucially, the regulators are focusing on conduct over market share. Unlike a traditional monopoly scenario, China’s Android ecosystem is fragmented among several big players rather than dominated by one. SAMR’s guidance makes clear that even without a single firm controlling the market, certain exclusionary behaviors can still violate the law. In March, China’s Supreme Court underscored this point by issuing a judicial interpretation that using technical means to obstruct a rival’s software or services constitutes unfair competition, regardless of the offender’s size. The message: platform power comes with responsibility, and abuse of that power won’t be excused simply because no one company has a 90% market share.

New boundaries for China’s tech giants – and what’s next

China’s mobile ecosystem may be unique, but the implications of SAMR’s move are clear: the country’s smartphone heavyweights will have to adjust. Tactics that Chinese handset makers long treated as standard business practice – from aggressive app store commissions to deeply embedded software hurdles – are now explicitly on notice. By naming and shaming these behaviors, regulators have effectively turned them into compliance risks. For the likes of Huawei, Xiaomi (1810.HK), Oppo, Vivo and others, this likely means re-evaluating their playbooks. App stores and pre-loaded services have been lucrative, especially with games yielding up to 50% commission, but those revenue models may need to be tempered under stricter scrutiny.

For software developers and content providers, the regulator’s intervention is a hopeful sign – albeit not a cure-all. The new guidance by itself won’t dismantle the dominance of the built-in app stores overnight; the mobile giants’ grip on distribution remains firm for now. However, developers finally have the weight of official policy on their side of the argument.

Ultimately, what’s at stake in this clash is not just a few percentage points of app commission, but the future structure of China’s digital marketplace. The hardware-software tug-of-war in China’s mobile sector will not resolve overnight – the incentives (and profits) that drive manufacturers to tighten their ecosystems remain powerful. Yet the ground rules are undeniably shifting. Regulators have made it clear that unchecked, system-level leverage by hardware firms is no longer tenable. For now, one thing is certain: the rules of engagement in China’s mobile industry have been redrawn, and everyone – from phone makers to app developers – is on notice.

This commentary is the view of the writer and does not necessarily reflect the views of Bamboo Works

Vivian Toh is a Singapore-based journalist who writes about technology and co-founder of Tech Tech China, a technology media startup. You can reach her at Vivian_toh@techtechchina.com.

To subscribe to Bamboo Works free weekly newsletter, click here

The low-altitude economy’s launch is opening a practical pathway for AI drones as Hong Kong enters a new phase marked by ageing buildings and challenges of managing a vertical city

By Gary Ng

Today’s Hong Kong is entering a pivotal moment in its urban evolution as the low-altitude economy (LAE) shifts from concept to becoming critical infrastructure. As cities worldwide explore new aerial corridors for logistics, inspection, and public safety applications, Hong Kong stands out due to its tightly packed skyline, ageing building stock and rising demand for safer urban operations.

With buildings averaging an age of 34.3 years and 19.2% of those built more than 50 years ago, the city faces increasing pressure to modernize how it inspects, repairs, and manages its vertical infrastructure. This is where the LAE begins to show its transformative potential that supports smarter, faster and safer city management.

Reimagining façade maintenance

For years, façade inspection has relied on such traditional methods as the use of scaffolding, gondolas, and manual visual assessments. While effective, these approaches are increasingly insufficient for a city ageing at scale.

These methods expose workers to fall risks, demand significant time and struggle to keep pace with the inspection frequency required across thousands of towers. Under the Mandatory Building Inspection Subsidy Scheme, every building older than 30 years is required to undergo a mandatory inspection every 10 years. This further increases the complexities of manual façade inspection measures.

AI drones, an initiative under the LAE, is now reshaping this challenge. Their ability to conduct high-resolution façade scans across entire elevations allows engineering teams to detect subtle defects long before they become safety hazards.

For example, in many deployments, autonomous drones can capture high-resolution façade scans of an entire block within a single morning, depending on site constraints and weather conditions. The drone’s onboard edge AI can help identify early-stage cracks around window ledges and areas of moisture seepage that have not yet manifested visibly.

The engineering team can then schedule targeted repairs within days, preventing potential structural deterioration cycles that could escalate into public-safety risks.

Hong Kong’s irregular building geometry creates tight spaces where conventional inspection methods struggle between towers. In the city’s North Point district, for instance, engineers are now overseeing a 25-storey residential redevelopment using AI-driven drones to examine a tight rear elevation that was nearly impossible to access by gondola.

Navigating with AI precision

Such integrated AI technologies in the drones rely on AI-based spatial mapping instead of GPS, using real-time 3D reconstruction to navigate between two buildings, which might be only four metres apart. It can capture façade footage from angles that were previously inaccessible due to human risk constraints.

This level of autonomy is often essential for building a more complete structural record. Continuous data over several flights helps engineers confirm the recurring cracks on the walls from thermal stress, enabling the project team to adapt their material selection for long-term performance.

It demonstrates how the LAE is enabling routine operations that were previously dependent on manual labour and partial visibility.

In dense cities, the future of drone operations is not about flying higher or farther; it is about enabling AI to interpret structural conditions, predict risk patterns, and interact natively with the urban digital ecosystem.

The value of modern technologies like aerial inspections is extending across the built-environment ecosystem. On several ongoing construction sites, contractors are deploying automated drone flight paths to capture weekly progress models. The drones map concrete pours, excavation volumes, and tower-crane zones with precision, feeding data into BIM models that help project managers reconcile discrepancies between design schedules and on-site realities.

This not only reduces industrial disputes but also accelerates measurement workflows and provides objective visuals for safety planning.

Utility operators are also experimenting with drone-based asset-condition tracking, where AI drones can conduct thermal and visual assessments of rooftop transformer units, detecting early overheating in a secondary coil. The issue might have remained unnoticed during standard manual inspections due to its intermittent nature.

Digital twins are another area seeing rapid integration. When drone-captured façade data is integrated into a building’s digital-twin platform, it automatically creates a temporal record to forecast repair budgets over a 10-year cycle. Engineers can now simulate water-seepage progression, thermal expansion stress, and material fatigue, making it easier to plan long-term maintenance rather than relying on reactive repair cycles.

Regulatory momentum drives scalable operations

Underlying these innovations is Hong Kong’s evolving LAE regulatory sandbox established in March 2025. It facilitated the testing of drone applications beyond visual-line-of-sight (BVLOS) flights under the permission of the Civil Aviation Department. For example, Traffic Control Technology (Hong Kong) Company Limited and MTR Corporation are using drones aided with AI technologies for inspecting areas such as railway tracks, stations and buildings.

As the ecosystem matures, new expectations are emerging around data governance, cybersecurity, and interoperability as well. Firms are moving toward encrypted, on-device processing to minimise the movement of raw visual data, with the ability of face and body anonymisation to protect the privacy of workers and residents.

Meanwhile, city-wide drone-ready infrastructure, such as rooftop landing pads and distributed charging nodes, is being explored to support future autonomous fleets that may eventually conduct routine inspections without human pilots.

The advancement of Hong Kong’s LAE is not about replacing traditional inspection techniques but elevating them with enhanced accuracy, repeatability, and operational safety. AI drones provide comprehensive visibility across not only mandatory façade inspections but also in utility assets and construction site safety.

This advanced form of monitoring is faster, more objective, and most essentially less hazardous for frontline workers. These changes signal a broader transformation where aerial intelligence becomes embedded in everyday city management.

Gary Ng is the CEO and co-founder of viAct, with over 10 years of experience driving AI- and automation-led innovation in the construction and industrial sectors. You can reach him at: gary@viact.ai

This commentary is the view of the writer and does not necessarily reflect the views of Bamboo Works

To subscribe to Bamboo Works weekly free newsletter, click here

With its significant gap in adoption, China has big potential to join growing global weight management, gut health and chronic disease prevention movements

LeadLeo Research Institute

As the concepts of “sugar, fat and calorie control” gain traction, dietary fiber has shifted from a passive nutritional component to a functional ingredient that many consumers are proactively choosing. It’s now widely used in sugar-free beverages, breakfast cereals, meal replacement powders and dietary supplements, becoming an important ingredient for food innovation and functional product pricing.

Industry experts note that dietary fiber – recognized by nutritional science as the so-called “seventh nutrient” – is increasingly supported by science for its benefits in weight management, blood sugar regulation, and gut microbe maintenance. Its water-absorbing and swelling properties help induce feelings of satiety, while certain soluble fibers ferment into short-chain fatty acids, which support a balanced intestinal environment. These mechanisms have driven a sharp rise in demand as metabolic health gains public attention.

The global dietary fiber market reached $514 million in 2023 and is projected to grow to $1 billion by 2030, representing 10% annual growth, according to industry data. Market growth is mainly driven by the rising need for chronic disease prevention, faster innovation in functional foods and changing consumer preferences in Asia – especially in China.

China is now the largest dietary fiber consumption market in the world. The Chinese Nutrition Society recommends 25 grams to 30 grams of daily fiber intake for adults, but the average remains significantly below that level, creating a clear intake gap. A large population and growing health awareness have made dietary fiber a crucial differentiating ingredient for food manufacturers.

China’s production of dietary fiber surpassed North America in 2019, ranking first globally. Several upstream enterprises are also accelerating their expansion of capacity, including a newly commissioned 30,000-ton annual soluble dietary fiber project, signaling expectations of sustained market growth.

From an industry chain perspective, upstream raw materials – including starch, sucrose, glucose and their processed derivatives – are strongly influenced by global corn and sugarcane prices. The sector’s midstream is the core production stage, where technological upgrades may reduce costs and improve particle quality. Downstream applications span food, beverages, dairy, baked products and nutraceuticals.

Soluble dietary fiber gains traction

Soluble fibers such as inulin, resistant dextrin and β-glucan are favored by beverage and dairy producers due to their better palatability and formulation compatibility. They serve as nutrient sources for probiotics in fermented dairy, enhance mouthfeel, improve stability and dispersibility in high-fiber beverages, and extend shelf life and product structure in baked foods.

Major dairy companies including Yili (600887.SH) and Mengniu (2319.HK) have launched products containing prebiotic dietary fiber, while meal-replacement and light-food brands have also made “fiber-enriched” a prominent value proposition, indicating that functional positioning is redefining formulation logic and pricing power.

Industry forecasts suggest that dietary fiber will evolve from a basic nutritional supplement toward a more systematic health management tool, with clearer functions in weight and blood sugar control, gut microbe balance and immune support. As precision nutrition, aging-focused nutrition, and chronic disease prevention continue to advance, “Dietary Fiber + Functional Foods” is expected to become a new competitive focus in the food sector.

Analysts note that the dietary fiber segment shows strong long-term fundamentals: big likelihood for strong demand, rapid expansion of application scenarios, and cost reduction driven by industrial upgrading, though challenges remain in regulatory standardization and intensifying competition.

As consumer health awareness increasingly shifts toward “prevention over treatment,” dietary fiber looks like one of the most promising functional ingredients following protein and probiotics, and is becoming a key growth engine in the food industry’s path toward a multi-billion-dollar market.

LeadLeo Research Institute is an original content platform for research on banks and companies and an innovative digital research service provider with nearly 100 senior analysts. You can contact the platform at CS@leadleo.com.

This commentary is the views of the writer and does not necessarily reflect the views of Bamboo Works

To subscribe to Bamboo Works weekly free newsletter, click here

Artificial intelligence is learning to ‘understand’ people to create a new wave of AI companionship that’s quietly reshaping how people connect with technology and themselves

By Lee Shih Ta

In the fast-diversifying world of generative AI, emotional companionship has emerged as one of the fastest-growing applications. A Harvard Business Review survey found that “emotional and psychological companionship” has become one of the most common use cases for generative AI — surpassing text generation and translation — signaling that expectations for technology are shifting from “help me work” to “understand me, be with me.”

Imagine this: late at night, the person chatting or watching a show with you may not be human, but an AI creation that understands you better than anyone else.

In the U.S., AI companionship is dominated by “virtual personalities.” Early pioneer Replika has attracted over 30 million users, nearly half of whom have set their AI as a “romantic partner.” Its successor Character.AI lets users create their own characters and talk to celebrity-like AIs, reaching over 100 million interactions in a single year. The latest sensation, AI Bro, positions itself as a “digital friend” that browses Tinder or social media alongside users, commenting in real time. Within weeks of the launch, it landed in the top 10 on the U.S. Apple App Store with over 70,000 daily downloads.

In China, the AI companionship market has taken on a more grounded and practical form. The viral app DouDou Companion combines “AI gameplay + chat.” Users can summon a voice-based AI teammate for the popular “Honor of Kings” online game, or talk with it at night for stories and casual conversation. The company told media outlets that its monthly active users has surpassed 2 million and that it plans to expand overseas. Its model is built on MiniMax technology, fine-tuned with Alibaba’s (BABA.US; 9988.HK) Tongyi Qianwen (Qwen) large language model.

Often dubbed China’s “AI personality factory,” MiniMax is the developer behind the popular voice-chat app Talkie, which lets users create custom AI characters for emotional interaction. The platform now has over 100 million global users, including 11 million monthly active users, with females making up about 48%. Founded by former SenseTime executive Yan Junjie, MiniMax has raised funds in multiple rounds, with backers including Tencent (0700.HK), Alibaba and HongShan, formerly known as Sequoia China. The Wall Street Journal reported that the company has confidentially filed for a Hong Kong IPO, targeting a listing later this year.

Paying to “be understood”

China’s AI companion market moves quickly — from model development to app launch, the typical cycle averages just three months. According to QuestMobile, user activity peaks at night between 10 p.m. and 2 a.m., driven mainly by young adults aged 18 to 35. They may not all be lonely, but they’re willing to pay for an AI that “understands them.”

A report by Business Research Insights estimated the global AI companion market will be worth $366.7 billion this year and projected it will rise to $972.1 billion by 2035, representing 36.6% annual growth.

Such rapid growth has raised ethical and regulatory concerns. In late 2024, China’s Ministry of Industry and Information Technology (MIIT) and the Cyberspace Administration of China (CAC) jointly issued guidelines that, for the first time, explicitly included “virtual characters and emotional companionship” under their supervision. The guidelines state platforms must prevent underage usage, limit hyper-realistic voice synthesis, and strengthen data security.

Yet regulation has not dampened innovation. Instead, it’s steering the industry toward longer-term applications. Some AI companies are collaborating with mental health platforms to launch “emotional-support AI coaches,” while others plan to integrate AI companions into smart speakers, home robots, and in-car voice systems, bringing companionship from screens into real life.

As humans grow accustomed to conversing, confiding, and coexisting with AI, companionship is becoming a new frontier in technology. This emerging intersection of algorithms and emotion is redefining what it means to “understand.” Perhaps it doesn’t mean machines know us better, but rather that, through them, we are learning more about ourselves.

Lee Shih Ta is an editor at Bamboo Works.

You can contact him at shihtalee@thebambooworks.com

To subscribe to Bamboo Works weekly free newsletter, click here

The country’s ready-to-drink coffee market has grown rapidly in recent years, transforming from a simple pick-me-up product into a lifestyle statement

LeadLeo Research Institute

China’s ready-to-drink (RTD) packaged coffee market has been expanding rapidly, evolving from a simple energy booster into a symbol of lifestyle and personal identity. Driven by convenience stores, supermarkets, and e-commerce platforms, RTD coffee has become a “grab-and-go” daily choice for many young consumers — shifting from a purely functional alertness booster to expressions of health and individuality. Brands are integrating low-sugar, low-fat, and functional ingredients to meet growing health awareness.

At the same time, the trend towards premium products is accelerating. Offerings are expanding from single flavors to diversified categories such as cold brew, latte, and oat milk coffee, with more refined packaging that appeals to consumer demand for quality and personalization. China’s RTD coffee market is thus moving from expansion in quantity to competition in quality.

RTD coffee — pre-packaged beverages ready for immediate consumption — now stands alongside instant and freshly brewed coffee as one of the three major categories. Its explosive growth reflects a deep shift in consumer behavior: from a social beverage to a daily necessity, and now a lifestyle symbol, mirroring the diversification of consumption patterns in China.

Industry data shows that China’s RTD coffee retail revenue grew from 12 billion yuan ($1.68 billion) in 2020 to 28 billion yuan in 2024, representing 18% annual growth, outpacing the overall beverage industry. Online channels have become the main growth driver, with their market share rising from 9% to 35%.

New generation redefining coffee market

According to data from Douyin, China’s domestic version of TikTok, coffee category sales in China grew 110% year-on-year in 2024, while the number of participating brands and listed products surged 142.9% and 380.3%, respectively. Livestreaming e-commerce has lowered barriers to entry, on-demand retail has enhanced convenience, and content-driven marketing has stimulated consumer interest — combining to power explosive online growth.

Competition is intensifying as the market shifts from international dominance to a mix of global and domestic players. The first tier includes Nestlé (NESN.ZU) and Starbucks (SBUX.US), which together hold about 60% of the market. The second tier — comprising Nongfu Spring (9633.HK), Uni-President (0220.HK), and Coca-Cola (KO.US) — is carving out space using health positioning and value-for-money strategies, while emerging brands are rising rapidly using innovative products and digital marketing.

Channel strategies are becoming increasingly differentiated. International brands rely heavily on supermarket and hotel channels, while local brands focus on e-commerce and membership-based retailing channels. For example, Tasogare Coffee’s 1L RTD coffee gets over 60% of its sales online, while Nongfu Spring thrives in Sam’s Club stores, reflecting a reshaping of consumption scenarios.

Health functionality has become the main direction for product upgrades. With growing consumer awareness, brands are adding functional ingredients and launching specialized formulations to meet segmented needs. These precision-nutrition strategies not only enhance premium pricing power but also create clear differentiation from traditional brewed coffee.

Localization through regional flavor innovation highlights the deepening maturation of China’s RTD coffee market. Brands are developing products inspired by local taste preferences — such as Yunnan small-bean coffee blended with local fruits, Sichuan pepper coffee, and frozen pear coffee — turning beverages into emotional and cultural connectors. The share of regional-flavor RTD coffee is expected to rise from 15% to 30%, becoming a key driver of market growth.

Packaging formats and technological innovation continue to push the boundaries of the RTD coffee experience. Beyond traditional bottles and cans, coffee powders, capsules, and concentrates are emerging, offering greater variety and flexibility.

Sugar-free and low-sugar options have shifted from niche differentiation to standard expectation. Data from leading grocery delivery company Dingdong Maicai shows that 87% of large-bottle coffee sales now come from sugar-free versions, compared with around 50% three years ago. This “health-as-default” trend is reshaping the market and pushing sweet, milk-heavy coffee drinks out of the mainstream.

Ingredient and formulation innovation underscore the market’s move toward professionalism and more premium products. Brands are moving beyond the simple coffee-sugar-milk formula, embracing specialty coffee principles that emphasize bean origin, roast profile, and extraction methods. At the same time, formulations are becoming more scientific — using natural sweeteners instead of artificial ones and adding dietary fiber and other nutrients.

Propelled by the dual drivers of health and quality, China’s RTD coffee market is entering a stage of high-quality competition. Only brands that can balance wellness attributes with superior sensory experiences will stand out in this new phase of the industry’s evolution.

LeadLeo Research Institute is an original content platform for research on banks and companies and an innovative digital research service provider with nearly 100 senior analysts. You can contact the platform at CS@leadleo.com.

This commentary is the views of the writer and does not necessarily reflect the views of Bamboo Works

To subscribe to Bamboo Works weekly free newsletter, click here

A recent ‘scandal’ involving a major restaurant chain is raising the question of whether China’s proud food culture can digest an industrialized dining revolution

By Lee Shih Ta

Whenever a dish arrives too quickly in today’s Chinese restaurants, diners can’t help wondering if it was made from scratch, or more possibly pre-made somewhere else.

Many well-known chains are facing criticism over pre-made meals. Northwestern-cuisine chain Xibei Youmian Village was recently accused by a famous blogger of using semi-finished food prepared in centralized kitchens, triggering a nationwide backlash. Overnight, the long-praised ideal of restaurant standardization suddenly turned into a crisis of trust. Many Chinese diners react almost instinctively to the notion of pre-made food, believing mass-produced, industrial cooking insults the country’s culinary heritage.

Ironically, the government feels just the opposite. Since its issue of a national document proposing the “cultivation and development of the pre-made-dish industry” in 2023, such food has been framed as a new form of “advanced productive force” that can modernize agriculture, expand cold-chain logistics and support rural revitalization.

Guangdong province began implementing a “General Requirements for the Labeling of Cantonese Pre-Made-Dish Packaging” rule this year, mandating disclosure of production methods and reheating instructions, while planning multiple industrial parks. To policymakers, pre-made food epitomizes a form of supply-chain upgrade — from farm to table — shortening processes, stabilizing quality, reducing waste and creating new jobs and export potential.

Yet the policy push doesn’t quite suit the public’s taste.

Efficiency tastes like betrayal

Also under fire is the once-trendy Tai Er Sauerkraut Fish chain. Some diners complain that three dishes can appear within seven minutes, suspecting the chain of using pre-sliced fish and ready-made broth rather than freshly killing each fish to order.

Tai Er’s parent, Jiumaojiu (9922.HK), reported a 10% drop in revenue and a 16% decline in net profit in the first half of this year, with table turnover falling to 2.2 times per day. Same-store sales at Tai Er’s directly operated restaurants fell 19% year-on-year. The company’s stock has dropped 40% since the beginning of the year.

Both Jiumaojiu and Xibei have denied using pre-made food — Tai Er insists its fish are delivered daily and marinated in-store, while Xibei claims its central kitchen only handles “pre-processing.”

This is the industry’s subtle gray zone: Under the official definition, only dishes that are fully cooked, packaged, and the later reheated for consumption count as pre-made. But in consumers’ eyes, anything not stir-fried by an on-site chef already fits their own pre-made definition. That perception gap turns standardization itself into an offense.

According to Wind data and a media report, 31 companies in China’s domestic A-share market are classified as pre-made food concept stocks, and about 60% have seen declining business over the past year. In 2024, net profits fell year-on-year for 22 of these companies. Weizhixiang (605089.SH), known as “China’s first pre-made food stock,” has seen its shares drop about 73% from their peak.

Why, then, does China persist in advancing this controversial sector? The answer is simple: as labor costs rise, urbanization accelerates, and food-safety rules tighten, pre-made food offers efficiency, traceability, and stable supplies.

For policymakers, it’s part of a modern food-manufacturing system — akin to electric vehicles or smart factories — representing a productivity upgrade. The problem is that Chinese dining culture is built on the immediacy of using a wok — creating dishes that are freshly cooked, smoky and alive. When the state openly promotes “pre-made dishes on the table,” ordinary diners hear a subtext: that bowl of noodles you’re eating might have been made in a factory far away. Such “de-localization” of cooking only heightens psychological resistance.

To be sure, Chinese consumers are not universally opposed to pre-made food. Many embrace it in the context of fast-food chains like McDonald’s or KFC, where industrial efficiency is expected and even celebrated. But when it comes to restaurants that market themselves as serving up Chinese culinary heritage — those emphasizing craftsmanship, freshness and the “wok spirit” — the use of pre-made dishes feels like a violation of cultural authenticity.

And that is the most delicate paradox of pre-made food. It marks the inevitable modernization of China’s dining industry, yet also crosses the deepest emotional boundary for many Chinese consumers. If one day restaurants can make people forget the words “pre-made” while tasting what they’re eating, then — and only then — will pre-made dishes truly find a place in the hearts and stomachs of Chinese diners.

Lee Shih Ta is an editor at Bamboo Works.

You can contact him at shihtalee@thebambooworks.com

To subscribe to Bamboo Works weekly free newsletter, click here

Beyond the financial figures, real determinants of success lie in the details of an operator’s unmanned driving systems and the scale of their deployments

By Yang Zhe

Since their listings on the Nasdaq at the end of last year, the regular stream of quarterly earnings reports from Pony AI (PONY.US) and WeRide (WRD.US) have offered a new lens onto robotaxi operators and their broader sector.

First and foremost, robotaxi companies won’t be profitable anytime soon. Even leading players can only hope to turn positive in terms of their unit economics — a widely recognized industry consensus. That means traditional valuation tools such as P/E ratios are irrelevant. With annual revenues of only a few hundred million yuan, or tens of millions of dollars, P/S ratios are also not meaningful. The sector resembles both robotics and innovative drugs: what investors care about is the size of the future market and each company’s probability of success.

Just as humanoid robots won’t be taking care of your retirement in the near term, investing in robotics companies is about their understanding of AI, not how many robots they send to perform at China’s Lunar New Year Spring Festival Gala on TV. While the market potential for robotaxis is widely acknowledged, profitability is still a future proposition. What really matters is the probability of future earnings, not short-term sales of vehicles to clients for one-off revenue. If valuations were based on car sales, autonomous driving companies would have to be valued like traditional automakers. But for a company that sells only a few hundred or thousand cars per year, its fair value would be close to zero.

Therefore, for L4 companies making completely autonomous vehicles — whether robotaxis or robotrucks — earnings reports, along with regular news flow, should be used to assess long-term plans, not quarterly profits. After all, there are no profits yet, just revenue.

In this race, investors must ask two key questions: First, do you accept the scenarios each company has painted? Second, after the promises are made, can you actually track the company’s progress? Tracking matters. Otherwise, you risk being misled. In the robotaxi sector, assume everyone is exaggerating, and focus on verifiable metrics that can’t be dressed up.

The real benchmark

So, how do you verify progress in robotaxi companies? Two metrics stand out: unmanned operation and scalability, in that order. A company must first demonstrate true unmanned capability, then expand the fleet at scale under unmanned conditions.

Fleet scale should not be measured simply by the number of cars owned as assets, but rather by how easily users in different locations can hail an unmanned vehicle, and whether these cars can operate under extreme scenarios, during rush hours and in adverse weather. The more driverless vehicles on the road and the greater the daily unmanned mileage, the higher the safety risks. Each company knows the internal accident probability per average kilometers without a safety driver, though such data is hard for outsiders to obtain. Instead, it manifests in observable variables: whether cars can operate without safety drivers, whether they can accept trips during rush hours and in crowded districts, and how many unmanned cars are actively available for on-demand rides each day.

Companies lacking confidence in their technical capabilities or safety records will try to limit operations in complex scenarios, and restrict operating hours or road segments to minimize risk exposure.

Thus, while robotaxis may appear similar on the surface, there’s a huge gap between a fleet where you can only hail cars with safety drivers, whether in the front or passenger seat, and one where you can hail 10 or even 100 driverless cars in public service. Each jump represents an order-of-magnitude increase in risk exposure, and an order-of-magnitude higher bar for technology and safety. Likewise, companies that allow rides during peak hours and in busy districts are fundamentally different from those that only operate in quiet times and locations. Accident rates in complex scenarios are significantly higher, even for human drivers.

Once safety indicators and technical requirements diverge by an order of magnitude, the technology gap is real. Lagging companies may never cross into the next tier. As L4 robotaxis emerge as the first large-scale commercial embodiment of intelligent AI agents, the bar for R&D is exceptionally high — and not every team can deliver.

Autonomous driving is, at its core, a game of probabilities. Without statistical rigor, neither autonomous driving nor investment in the sector is possible. Through earnings reports, public filings, and the user experience visible to the public, investors can infer the technical gaps between companies. And when gaps exist, so too do differences in each company’s probability of ultimately reaching the robotaxi “endgame.”

Yang Zhe is a senior director at Pony AI with nearly 10 years of experience in the autonomous driving industry. You can reach him at: mail@yangzhe1991.org

This commentary is the view of the writer and does not necessarily reflect the views of Bamboo Works

To subscribe to Bamboo Works weekly free newsletter, click here

K-pop superstars are once again staging blockbuster concerts in Macao, while online music platforms are testing the waters against an unspoken ban in China

By Lee Shih Ta

Strictly speaking, a so-called “Hallyu ban,” barring Korean pop culture from China, never appeared in any official government document. Beijing never formally announced such restrictions, and thus could never explicitly lift them either. Yet for the past nine years, the reality has been clear: Korean dramas were pulled from Chinese TV screens, K-pop stars were edited out of programs, and idol tours came to a halt. The once-vibrant Korean Wave in China cooled overnight, turning from a warm current into a cold tide. Now, in 2025, the ice may finally be thawing.

According to Korean customs data, Korean audiovisual exports to China in January and February reached $9.25 million, 3.5 times higher than the same period a year earlier. In the first half of the year, cumulative exports exceeded $31.52 million, more than doubling year-on-year. These figures suggest that demand for Korean content in China has not disappeared. Once policy restrictions ease, even slightly, new consumer spending could quickly follow.

Offline activities tell a similar story. In recent years, the number of K-pop fan meetings in China has increased noticeably, from newcomer idols to veteran group members, gradually restoring closer interaction with Mainland Chinese fans. Although the scale is not yet comparable to the frenzy of the 2010s, the growing frequency of such events across an increasing number of cities signal that cultural exchanges are resuming.

An even more symbolic scene is playing out in Macao, which is part of China but enjoys a high degree of autonomy. Recent K-pop mega-concerts there have sold out and drawn large crowds of Mainland fans from across the border. Macao has thus become a window into a thaw of the “Hallyu ban,” showing that enthusiasm for K-pop among Chinese audiences never truly faded, only lay dormant.

Those most sensitive to this comeback include online music platforms. Tencent Music (TME.US) is a prime example. In May, the company invested 12.9 billion yuan ($1.8 billion) for a stake in South Korea’s SM Entertainment (041510.KQ), becoming its second-largest shareholder. Known for producing multiple generations of idols such as Super Junior, EXO, and aespa, many believe SM’s capital tie-up with a Chinese internet giant will systematically channel its K-pop resources into China.

Soon after, Tencent Music also announced partnerships with The Black Label and H MUSIC, further importing K-pop albums, music videos, and exclusive content. More significantly, Tencent introduced “Bubble,” a private fan-interaction community that has been hugely popular in South Korea. For 28 yuan, or about $4, fans can join an idol’s inner circle, receive exclusive messages, and even gain priority ticket access. This “emotional economy” is markedly boosting user stickiness and subscription rates.

The numbers speak for themselves. In the second quarter of 2025, Tencent Music reported revenue of 8.44 billion yuan ($1.17 billion), up 17.9% year-on-year, with net profit surging more than 30%. Its paying users surpassed 124 million, including 15 million super members. Analysts widely agree that the expansion of K-pop content and concert-related services were key drivers of this performance. By contrast, NetEase Cloud Music (9899.HK), which relies more heavily on independent musicians and community engagement, managed to turn a profit but still saw revenues decline in the first half of the year — underscoring the unique value of K-pop in driving platform loyalty.

But will you still love me?

K-pop’s apparent comeback does not necessarily mean a full revival. First, China’s entertainment landscape has undergone structural changes since K-pop’s untimely exit. Domestic films and idols have gained significant appeal in recent years. One such example is box office phenomenon “Ne Zha 2,” which showed that local culture can also spark nationwide excitement. K-pop’s re-entry into the Chinese market may serve more as a subcultural supplement rather than a dominant mainstream force.

Second, nearly a decade of restrictions has reshaped the cultural preferences of younger audiences. While policies may be loosening, some consumers now prefer to support domestic idols and cultural products. This psychological shift could be an invisible barrier to a full K-pop revival.

Finally, any platform’s reliance on external intellectual property (IP) carries inherent risks. Should policies tighten again, or should fan enthusiasm fade, platforms may find themselves facing another round of “content withdrawal.” For Tencent Music, K-pop is both a commercial wager and a delicate experiment in emotional economics.

From a business perspective, betting on K-pop to amplify emotional engagement is still worth the gamble. But when the lights of K-pop shine once more on China’s stages, the screams in the crowd may carry not only excitement at reunion, but also the complex emotions of uncertainty about the future.

Lee Shih Ta is an editor at Bamboo Works.

You can contact him at shihtalee@thebambooworks.com

To subscribe to Bamboo Works weekly free newsletter, click here

AI competition is shifting from models to applications, creating new opportunities in consumer and enterprise markets while reshaping industries

LeadLeo Research Institute

As the rollout of new AI model technologies marches on, the focus of innovation is moving from the models themselves toward applications. A model-driven application revolution is now transforming business paradigms and user experiences across countless industries.

As AI capabilities extend from the cloud to real-world scenarios, the pressing question for all market participants is how to precisely capture user needs, open viable commercialization pathways, and harness emerging traffic patterns. LeadLeo Research Institute recently released its 2025 China Foundation Model Application Market Insight White Paper, which analyzes the market by focusing on consumer- and enterprise-level applications, exploring core scenarios, traffic structures, differentiated demands and competitive strategies.

Application ecosystem taking shape

The report points out that in the consumer-facing market, application-layer products of large models can be divided into three categories by business maturity: more mature embedded applications, developing native AI applications, and early-stage explorations of smart hardware. Embedded applications enhance existing software to achieve more efficient monetization and are the most maturity. Native AI applications create entirely new services around the model itself, though their business models are still being tested. Smart hardware combines physical interaction with AI, holding huge potential but also facing significant cost, as well as technical and adoption challenges.

Each type follows a different value realization path: embedded applications rely on their user base for rapid monetization; native AI must prove itself through competition; while smart hardware must overcome technological, cost and user-acceptance hurdles.

Globally, consumer-facing web-based AI applications show a strong concentration effect. ChatGPT leads with nearly 4.7 billion monthly visits, followed by Microsoft’s New Bing at 1.53 billion. Together they form the first tier, while DeepSeek, Gemini, Perplexity, Character.ai, and Claude make up a second tier that still lags significantly behind the leaders.

The report states that AI chat assistants and AI search engines are the two dominant categories for web-based AI applications, accounting for over 80% of traffic. This is because information retrieval and interactive Q&A are the most frequent and fundamental online needs, and these applications efficiently fulfill them. Their broad applicability and large user base, combined with the first-mover advantage and integration with existing entry points, have helped the leading products accumulate massive users and validate their core value.

The differences between mobile apps and web-based applications largely lie in interaction and feature integration. Web applications require no installation, work across platforms, and are well-suited for quick information access and text-based interaction. By contrast, mobile apps provide smoother, more personalized interfaces and can deeply leverage hardware resources such as cameras and microphones, enabling richer integration into mobile use cases like photo editing and real-time translation.

Industry landscape reshaped

These differences between web- and app-based platforms significantly affect the global mobile AI application market. Although AI chat assistants continue to dominate with nearly 70% of MAU share, their lead and user scale are weaker than on the web. Meanwhile, AI search engine share has declined relative to web use, while AI image editing applications have surged to nearly 10% share thanks to their strong fit with mobile photography and imaging scenarios, highlighting distinct mobile user preferences.

On the consumer side, AI applications have shown strong penetration in assistants and office tools, while AI-driven creative and entertainment apps struggle with stickiness and slowing growth. Assistant-type applications are already central, with deep voice interaction and smart assistant products growing rapidly and becoming industry benchmarks. Office and creative applications such as WPS AI and AiPPT.cn have also shown strong growth, tightly integrating with productivity needs. In contrast, lifestyle and entertainment AI apps, though capable of short-term hype, often lose users once the novelty fades.

For enterprise applications, the key to successful deployment lies in aligning capabilities with needs, ensuring quantifiable ROI, and securing adequate data and computing resources. Only with these three factors in sync can enterprises unlock efficiency and sustainable value. Industries such as finance and healthcare already show great potential, sharing traits of high-value proprietary data, sufficient budgets, and clear demands, enabling meaningful benefits in automation, prediction, and decision support.

Currently, Chinese enterprises are realizing returns from AI foundation model applications mainly through improved operational efficiency. Evidence across the finance, manufacturing, and retail sectors shows impact in task automation, process optimization, and accelerated R&D. As technologies mature and scenarios expand, the value of large models is expected to shift from short-term cost reduction toward strategic growth and revenue expansion — evolving from pilot projects to deeply embedded, vertically integrated innovations at the core of business.

Nevertheless, significant challenges remain in ensuring precise alignment of models with enterprise needs. Surveys show that as many as 87% of enterprises believe current models still fall short when handling highly complex logical tasks. 62% report a lack of standardized criteria for model selection. 50% indicate a mismatch between model capabilities and business needs. 39% say that limited fine-tuning and deployment tools affect efficiency. Overall, achieving large-scale, sustainable adoption will require further progress in accuracy, standardization, industry alignment, and tool chain support.

LeadLeo Research Institute is an original content platform for research on banks and companies and an innovative digital research service provider with nearly 100 senior analysts. You can contact the platform at CS@leadleo.com.

This commentary is the views of the writer and does not necessarily reflect the views of Bamboo Works

To subscribe to Bamboo Works weekly free newsletter, click here

China’s centralized bulk procurement policy has driven dental implant prices down by half, easing costs for patients but shaking the profit model of dental providers

By Lead Leo Research Institute

As China’s healthcare reform has deepened in recent years, the country’s centralized bulk procurement (CBP) policy has expanded from pharmaceuticals and medical consumables to dental implants. CBP involves large-scale, unified government tenders to lower medical product prices by exchanging volume for price reductions. First piloted in the pharmaceutical sector in 2018, the policy was later extended to high-value consumables such as coronary stents and artificial joints, and has now reached dental implants.

In the past, a single dental implant could cost over 10,000 yuan ($1,390), a heavy burden for many middle-aged and elderly patients. With the implementation of CBP, prices have dropped significantly, greatly easing the financial pressure on patients. But for dental institutions long reliant on the high-margin implant business, the policy represents a structural shock.

According to Lu Jiarui, a medical industry analyst at Lead Leo Research Institute, CBP is transforming the dental implant market by dismantling the long-standing high-price model, compressing profit margins across the supply chain into a more transparent range.

Current dental implant prices generally range from 5,000 yuan to 7,000 yuan, a drop of more than 50%. This has sharply lowered the threshold for patients, theoretically stimulating demand. However, the sudden price cut is painful for many dental chains. Implant services used to be one of their profit mainstays with gross margins of 60% to 70%, which may now be slashed in half. In the short term, many small and medium-sized clinics may be forced out due to cash flow pressure. Leading chains can partially offset the shock through economies of scale and diversified services, but an overall margin decline appears inevitable.

Upstream integration and service innovation

For dental chains, this is both a crisis and an opportunity for transformation. Leading companies are building competitive advantages in three ways: moving upstream in the supply chain, such as by establishing processing centers or partnering with domestic manufacturers to reduce costs; adopting digital transformation through AI implant planning and 3D printing to improve efficiency; and innovating service models, such as offering full-cycle care packages combining implants, periodontal treatment, and regular maintenance.

The market is becoming more segmented — high-end clinics are shifting toward complex implants and premium services like ALL-ON-4 to maintain pricing power, while the mass market is pursuing scale through standardized procedures.

At a deeper level, the reform highlights structural contradictions in China’s dental care sector, addressing the public’s complaints about expensive dental care while ensuring the sustainability of medical institutions. Future policies may need to establish differentiated payment systems (basic CBP-covered options plus premium self-paid options), or reform personal medical insurance accounts to improve payment capacity. The era of relying solely on policy dividends is over; companies must build long-term competitiveness centered on medical quality and supported by patient experience.

Lu Jiarui predicts several trends for the dental clinic sector. First, the industry will accelerate toward a dual-track structure of “high-end specialty clinics + community-based affordable clinics.” The high-end market will focus on complex implants and orthodontics, while community clinics will handle basic treatment and preventive care, achieving scale through standardized services and insurance coverage.

Second, growth momentum will shift toward efficiency gains from technological innovation, new value created through services, and incremental demand from market penetration into lower-tier areas. Two niche segments — early intervention in children’s oral health and elderly oral care — may become new growth drivers.

On a broader level, the evolution of dental clinics reflects a transformation in China’s medical service system — from “commercialized medicine” toward “medicalized commerce.” Where the past emphasized rapid expansion and high profits, the focus must now return to the essence of medicine, seeking sustainable models.

For chain operators, the biggest strategic challenge is balancing economies of scale with medical quality. Some brands have shown a “McDonaldization” tendency during expansion, where excessive standardization leads to declining quality; others have remained “small but beautiful” but struggle to break growth bottlenecks. A potential solution is the “flagship + satellite” model: establishing flagship institutions with teaching and research capacity in core cities, supported by standardized satellite clinics in surrounding areas, with remote consultations and technology sharing to maximize resource efficiency.

LeadLeo Research Institute is an original content platform for research on banks and companies and an innovative digital research service provider with nearly 100 senior analysts. You can contact the platform at CS@leadleo.com.

This article was originally published on Blue Whale Finance by Tu Jun.

This commentary is the views of the writer and does not necessarily reflect the views of Bamboo Works

To subscribe to Bamboo Works weekly free newsletter, click here

As Hong Kong’s Stablecoin Bill gets set to take effect, the RWA boom is arriving ahead of schedule — turning the city into a live experiment in on-chain finance

By Lee Shih Ta

As Hong Kong’s Stablecoin Bill gets set to take effect in August, the global spotlight is returning to the Asian financial hub. The legislation formalizes rules for issuing stablecoin and opens a regulatory gateway for Real-World Assets (RWA) to access mainstream financial markets.

RWA refers to tangible, deliverable physical assets, like commodities, real estate, and commercial papers, that have been verified, digitized, and tokenized via blockchain technology. Unlike purely virtual assets, RWA enhances transparency, liquidity, and access to financing within the real economy. It bridges the gap between traditional finance and crypto markets, while offering investors exposure to assets with real-world value.

The significance of the Stablecoin Bill for RWA lies in its legal foundation for digital settlement and clearing systems. Without stablecoins, there would be no programmable, real-time, on-chain trading environment for RWA to scale meaningfully. Conversely, stablecoins without asset-backed use cases or tangible economic anchoring struggle to gain traction or trust. RWA thus provides both the backing and the application context that can enhance stablecoins’ credibility and utility as a medium of settlement.

Since June 2023, the Hong Kong Monetary Authority and the city’s Financial Services and the Treasury Bureau have supported multiple pilot projects for stablecoins and RWA, with participants including the Chinese Gold and Silver Exchange, Bank of China (Hong Kong), Hong Kong Exchanges, HashKey and Ant Digital Technologies, the technology arm of Alibaba’s Ant Group.

In addition, several export-oriented enterprises in key trading Mainland Chinese provinces such as Zhejiang and Shandong have begun tokenizing assets like aluminum ingots, copper plates, and sugar through partnered platforms to facilitate financing and advance settlement for overseas orders.

According to estimates by Boston Consulting Group, the global tokenization market could reach $16 trillion by 2030 — roughly three times the current market value of gold.

Market heats up

Capital markets have responded swiftly. Hong Kong-listed firms such as Synagistics (9993.HK), GL-Carlink Technology (8279.HK), and DL Holdings (1709.HK) have all recently announced plans to enter the stablecoin and RWA space. Be it applying for licenses, signing MOUs, or simply stating “interest in exploring opportunities,” such moves have triggered surges in company stock prices.

Some experts have warned that current “stablecoin concept stocks” have significantly deviated from their fundamentals. Driven by short-term speculative capital from Mainland China, shares of many companies are being aggressively bid up based only on partnership intentions or unconfirmed rumors. Without concrete project implementation, these stocks are likely to face corrections once the new regulatory policies officially come into effect in August.

In reality, RWA — as the foundational infrastructure connecting asset verification with on-chain finance — serves as a true test of whether companies possess real industrial use cases and transaction demand. Enterprises engaged in physical goods trading, warehousing and logistics, cross-border settlement, and blockchain deployment are expected to have a stronger competitive edge in this transformation.

Meanwhile, fintech platforms can leverage their existing user bases and technical capabilities to build tokenization and settlement systems. For example, Finloop Finance Technology, the fintech arm of Fosun International (0656.HK), recently launched an integrated RWA platform called FRP. According to Finloop, after tokenizing money market funds, the next wave will likely include bonds, higher-rated funds, and commercial papers — with physical assets following later. In the next two to three years, standardized financial products are expected to be the preferred RWA category in the market.

With the Stablecoin Bill taking effect in August, Hong Kong may be on the cusp of a new era — one defined not by hype but by asset-backed, on-chain financial innovation grounded in the real economy. But in this race to redefine digital finance, the winners will not be those who run the fastest, but those who build with purpose and substance.

Lee Shih Ta is an editor at Bamboo Works.

You can contact him at shihtalee@thebambooworks.com

To subscribe to Bamboo Works weekly free newsletter, click here

Key hurdles have shifted from purely technological breakthroughs to feasible business models and implementation of policy and regulatory frameworks

By Lead Leo Research Institute

In 2025, China’s robotaxi industry is entering a critical phase of scaled operations. Reducing hardware costs and optimizing algorithms are becoming essential prerequisites for commercialization. Tesla’s (TSLA.US) Cybercab — featuring a steering wheel-free design and a “boxless” manufacturing process — is expected to bring the cost per unit down to $25,000 by 2026, representing a 70% reduction compared to Waymo’s models.

Chinese players are also progressing: Baidu’s (BIDU.US; 9888.HK) Apollo RT6 uses mass-produced LiDAR to cut costs to $29,000, and WeRide’s (WRD.US) seventh-generation system reduces the number of sensors by 30% while improving perception accuracy.

At the policy level, over 10 cities, including Beijing and Shanghai, now permit fully driverless robotaxi services with fare collection. Shenzhen has further clarified that liability in accidents falls on automakers and system suppliers, accelerating the industry’s expansion.

Key barriers to national expansion

With robotaxi services increasingly appearing in daily life, what obstacles remain to nationwide deployment in China? Challenges include unclear profit models, fragmented regulation, and reliance on human safety operators. Yet with falling costs and growing policy support, robotaxis are moving closer to becoming commercially viable.

While technological progress remains crucial, commercial viability also depends on non-technical factors. A closed business loop is essential for scale. Robotaxis aim to replace traditional ride-hailing services with a more cost-effective solution by reducing labor costs and operating around the clock. In the short term, however, manufacturers face high upfront costs, including for vehicle production, sensors, maintenance, marketing, and retrofitting. In 2023, estimates placed these at 100,000 yuan ($13,950) for the base vehicle, 80,000 yuan per vehicle for sensors, another 80,000 for annual operations, and 10,000 yuan for retrofitting. The first three of those account for 37%, 29.6%, and 29.6% of total costs, respectively.

Regulatory alignment is another key factor, covering safety, ethics, and public acceptance. The full commercialization of robotaxis requires policy support — specifically, allowing them to operate freely on open roads like traditional ride-hailing vehicles.

When formulating comprehensive open-access policies, governments must take into account factors such as technology maturity, public acceptance, and potential safety and ethical concerns. Given these considerations, authorities have taken a cautious stance toward fully opening up robotaxi services, aiming to balance technological innovation with the need to ensure public safety and social stability.

Robotaxis boast high revenue potential — Waymo estimates six times the mileage of regular cars. But current limitations prevent them from operating in fully open environments, and their technological maturity remains under scrutiny. Substantial marketing subsidies, continuous R&D, daily maintenance, human backup drivers, and expensive custom vehicles remain heavy financial burdens for operators.

Still, the core manufacturing cost of robotaxis is increasingly coming under control, improving their labor replacement potential. As market awareness grows and regulation matures, pilot zones are expanding. Hitting a fleet of 1,000 vehicles is generally considered the breakeven point for most operators.

LeadLeo Research Institute is an original content platform for research on banks and companies and an innovative digital research service provider with nearly 100 senior analysts. You can contact the platform at CS@leadleo.com.

This commentary is the views of the writer and does not necessarily reflect the views of Bamboo Works

To subscribe to Bamboo Works weekly free newsletter, click here

As the U.S. moves swiftly to set the rules for on-chain finance, an increasingly anxious China is seeking entry to the virtual currency world via Hong Kong

By Lee Shih Ta

Stablecoins have played an increasingly pivotal role in the virtual asset market in recent years due to their blockchain-based payment and settlement capabilities. While the U.S., under President Donald Trump, is actively advancing stablecoin legislation to establish regulatory dominance, China has taken a more indirect approach due to its own ban on cryptocurrencies on the Mainland. It has responded through institutional reforms in Hong Kong and deployments by Mainland-backed brokerages — extending the global currency-clearing battleground into the digital asset space.

In April 2025, the U.S. House Financial Services Committee passed the Stable Act, requiring stablecoin issuers to hold banking licenses and submit to federal oversight. Meanwhile, the more stringent Genius Act was recently approved by the Senate. The U.S. strategy is clear: incorporate U.S. dollar-pegged stablecoins like USDC and USDT into its formal financial system to reinforce its dominance in on-chain finance. Driven by this momentum, Circle (CRCL.US), the issuer of USDC, has seen its stock surge 3.8 times since its early June listing, strengthening market expectations that the dollar will ultimately rule supreme in blockchain-based settlements.

Stablecoins are cryptocurrencies pegged to fiat currencies, most commonly the U.S. dollar, and are favored in scenarios like DeFi, cross-border payments, and tokenized assets due to their price stability and redeemability — features not shared by highly volatile cryptocurrencies like bitcoin or ethereum.

The significance of U.S. stablecoin legislation goes beyond consumer protection and anti-money laundering. It aims to integrate stablecoins into the official financial clearing system, enabling cross-border payments and asset settlements without reliance on the widely used SWIFT clearing system or commercial bank networks. Once achieved, this would embed the dollar into every major blockchain ecosystem worldwide.

90% of stablecoins pegged to the U.S. dollar

The global value of circulating stablecoins surpassed $140 billion by the end of 2024, with over 90% of those pegged to the dollar, according to data provider CoinMetrics. Even in decentralized Web3 ecosystems, the dollar maintains its influence through stablecoins.

This presents a serious challenge for China, which wants to enhance the global standing of its own currency, the yuan or renminbi (RMB). While Beijing bans cryptocurrency trading and mining domestically, it has aggressively promoted its central bank digital currency (e-CNY). Yet, e-CNY’s closed-loop design hinders its global adoption. Against this backdrop, Chinese state media have frequently criticized U.S. stablecoin legislation this year, warning of "digital dollar hegemony."

But China hasn’t stopped with simple rhetoric. In June, Hong Kong – which is part of China but has its own separate financial system – passed its own stablecoin licensing regime, which will take effect on Aug. 1, marking China’s first tangible step into the global digital asset clearing framework.

At the same time, Guotai Junan International (1788.HK) became the first Mainland-backed brokerage approved to offer virtual asset trading and custody services in Hong Kong — a cautious but symbolic entry of Chinese capital into blockchain finance. While not directly issuing stablecoins yet, China’s regulatory framework in Hong Kong could pave the way for yuan-pegged stablecoins in the future.

Other major Chinese brokerages such as CSC Financial and Haitong International are also rumored to be exploring potential participation, building virtual asset platforms and risk control models. This two-pronged strategy — at the policy and commercial levels —suggests that China may be using Hong Kong as a testbed for yuan-based blockchain clearing systems.

Unlike the U.S. legislative model, Hong Kong uses an administrative guideline and licensing regime. Still, it offers a comprehensive framework on reserve backing, disclosure and transparency. And, crucially, it allows the issuance of stablecoins pegged to its own currency, the Hongo Kong dollar, and also the yuan — opening the door for more Chinese financial institutions to participate in cryptocurrency development.

For now, it remains unlikely that Mainland China will ease its restrictions on cryptocurrencies. Without a liberalized digital asset ecosystem, catching up to U.S. dollar-backed stablecoins will be a formidable task. However, further relaxation in Hong Kong or limited cross-border use cases are possible. As a controlled financial hub, Hong Kong may ultimately become China’s launchpad for developing yuan-based stablecoins — ensuring China and its currency won’t be absent from the coming competition for global clearing dominance.

Lee Shih Ta is an editor at Bamboo Works.

You can contact him at shihtalee@thebambooworks.com

To subscribe to Bamboo Works weekly free newsletter, click here

Content creators may feel the most profound shift and play a more important role as data becomes a strategic asset in the AI era

By Lee Shih Ta

As the global AI race heats up, it’s becoming clear that data doesn’t lose its value once large models reach the reasoning stage. On the contrary, it’s even more critical due to the need for dynamic knowledge. The so-called “last mile” of high-quality datasets often determines a model’s ultimate performance.

That is likely why Facebook parent Meta Platforms (META.US) made a $14.3 billion strategic investment in Scale AI, a company focused on data labeling and cleaning for AI training.

Scale AI provides structured, high-quality datasets to OpenAI, Meta, Google and other tech giants by combining the output of massive human labor with automated pipelines. Its data labeling process involves tagging images, texts or audio with meaningful metadata — such as identifying pedestrians in a photo or labeling the point of an article. Data cleaning eliminates errors, duplicates or irrelevant material to ensure consistency and accuracy.

Another example of the growing value of quality data is a recent licensing deal between The New York Times and Amazon (AMZN.US), which allows fact-checked editorial content to be used for training AI models. A similar agreement between the Associated Press and OpenAI has also been signed.

Though these arrangements are described as content licensing, they reflect a deeper shift: content has become data, and data has become a service. These deals highlight how media organizations are reassessing the value of their content, while AI developers continue to pursue high-quality material with growing urgency.

In contrast, the Chinese-language AI ecosystem faces unique challenges, such as a shortage of publicly available data, lack of large-scale professional annotation and difficulty digitizing classical and cultural texts at scale. Such obstacles highlight the challenges facing development of localized large AI models.

Chinese-language materials are relatively scarce

A white paper published by Alibaba Research Institute notes that English accounts for 59.8% of all crawlable web text, while Chinese represents just 1.3%. Wikipedia, a commonly used open resource, has over 7 million English articles, whereas there are only 1.5 million Chinese — less than a quarter of the volume.

This imbalance creates a major disadvantage. Without sufficient publicly available Chinese material, local large language models in Chinese may fall far behind their English-language counterparts in natural understanding and text generation — potentially leading to culturally mismatched outputs and a sense that these models have “consumed too much foreign ink.”

Chinese authorities have long recognized this gap and have taken steps to address it. Platforms such as People’s Daily and Xinhua are actively constructing curated, high-quality materials, consisting of vetted news, commentary and policy interpretation, designed to ensure alignment with official values and to support AI safety from a moral and ideological standpoint.

Initiatives like the "Cyber Research Large Language Model" further concentrate on integrating data from legal and policy documents, state media and other publications, reinforcing alignment with Chinese values.

In China, such value alignment has become a basic requirement for any domestic AI system. While China has yet to produce a company of Scale AI’s size, several local firms, including Aishu Technology, Testin, iFlytek (002230.SZ) and Haitai Ruisheng (688787.SH), are building up their capabilities in large-scale data annotation and cleaning. The Shanghai AI Lab is also developing a platform-based material processing system in partnership with policy and academic resources, laying the foundation for a “Chinese version of Scale AI.”

According to market research firm IDC, the value of China’s AI training data market was estimated at $260 million in 2023, and is expected to grow to approximately $2.32 billion by 2032, representing a compound annual growth rate of 27.4%.

Ultimately, the performance of any AI model depends on the content it consumes. In the AI era, content creators — especially those in journalism — must recognize that they are no longer merely material providers. They are now an integral part of the data services supply chain.

When news stories, commentary, academic papers and cultural archives are structured, semantically labeled and integrated into AI training pipelines, their value shifts from real-time information to durable data assets. Content creators who proactively organize and annotate their materials, and pursue licensing partnerships with AI developers, may find themselves unlocking new revenue opportunities.

It’s time for content to be seen not just as narrative, but also as infrastructure.

Lee Shih Ta is an editor at Bamboo Works.

You can contact him at shihtalee@thebambooworks.com

To subscribe to Bamboo Works weekly free newsletter, click here

The storytelling revolution of short, three-minute dramas on e-commerce platforms has seamlessly merged ‘watching’ and ‘shopping’

By Lee Shih Ta

“She wore shabby clothes and was dumped by her boyfriend. But she was determined to turn her life around and looked stunning as she made an amazing comeback at his wedding three months later.”

If this storyline sounds familiar, congratulations — you’ve become a regular in the universe of storytelling content on e-commerce.

Over the past two years, a growing number of e-commerce platforms have launched “short drama” sections, investing real capital to support creators. By integrating products into plots, they allow users to “watch and shop” simultaneously — often making purchases without realizing it. What began as a marketing gimmick has evolved into a structural shift in how attention and consumption decisions are captured.

Driving this shift is the explosive rise of Douyin E-commerce, the Chinese version of TikTok. In 2024, Douyin’s annual GMV exceeded 3.5 trillion yuan ($486.4 billion), making it China’s third-largest e-commerce platform. It proved that content-driven commerce — "video-based seeding + livestream conversion" — can disrupt the traditional “search–compare–checkout” model. Traditional platforms realized that relying solely on static listings and text might soon render them invisible.

Short dramas have a natural stickiness: each episode runs about three minutes, with melodramatic plots, extreme characters and exaggerated twists. More importantly, short drama marketing goes beyond embedded ads — brands can custom-produce episodes, integrate with livestreams, and use the platform’s algorithms to track user preferences during viewing.

That’s why e-commerce platforms are now producing dramas themselves.

Alibaba (BABA.US; 9988.HK) launched “Guangguang Theater” on Taobao at the end of 2023, focusing on co-branded dramas and merchant-involved scriptwriting, integrating with Taobao Live and product pages to enable “watch and buy” functionality. Douyin pushed its “drama + shopping” model in the second half of 2023, embedding shopping carts and character-based shops. Meanwhile, PDD Holdings (PDD.US) added a short drama section to its “Duoduo Video” in April 2024, focusing on low-cost “good value dramas” for smaller, less-affluent cities. This deep fusion of storytelling and commerce is redefining the shopping experience.

Different genres, different products

Different genres of drama often pair with specific products. The “revenge glow-up” narrative —where a protagonist undergoes a striking transformation after being mocked or underestimated — is particularly effective for promoting fashion, beauty products, and self-improvement services such as online courses. “Emotional family dramas,” involving marital or household conflict, are suited for home goods, kitchenware, and seasonal gifts. Meanwhile, “twist and reconciliation” storylines, which build emotional tension before ending on a heartfelt note, are ideal for enhancing the sentimental value of accessories, keepsakes, or romantic gifts.

While these storylines seem repetitive, they’re highly efficient. The goal of short dramas isn’t originality, but familiarity and rhythm. Viewers don’t need to understand backstories — they just need to expect revenge, redemption, or tears. This is when platforms guide users toward purchases, creating a golden moment for brand conversion.

Meanwhile, an increasing number of drama clips are being repurposed and distributed across content platforms like Douyin and RedNote. E-commerce platforms have become content producers, while content platforms act as amplifiers and distributors.

Don’t underestimate the impact of these short dramas. According to the “China Internet Audio-Visual Development Research Report (2025),” the micro-drama market was worth 50.44 billion yuan in 2024, surpassing China’s total box office revenue of 42.5 billion yuan. The micro-drama market is expected to reach 100 billion yuan by 2027.

However, this content-commerce experiment isn’t without its risks and challenges.

First, there’s content fatigue. Over-reliance on formulaic plots may eventually bore viewers, leading to user loss despite short-term success. Second, brand identity is often diluted — viewers remember the drama, not necessarily the brand, making long-term loyalty hard to build. Third, cross-platform conversion is hard to track. Even if a drama goes viral, platforms and merchants often lack visibility into how much traffic gets converted, and where.

But these challenges don’t spell the end of the short drama e-commerce model. To the contrary, they may be paving the way for an entirely new form of content-driven commerce — one that builds a multi-layered ecosystem with content at its core. Viewers binge-watch short dramas while shopping, and platforms continuously iterate to find stories that make people emotionally — and financially — invest.

In a way, e-commerce is starting to look like a drama that never ends.

Lee Shih Ta is an editor at Bamboo Works.

You can contact him at shihtalee@thebambooworks.com

To subscribe to Bamboo Works weekly free newsletter, click here