Chip valuations return to rationality; T-Head misses prime IPO window Alibaba Group recently announced plans to spin off T-Head, its chip-design subsidiary, for an independent public listing in Hong Kong Key Takeaways:  Alibaba submitted its application to the Hong Kong Stock Exchange via a confidential filing  The market estimates T-Head will raise between $2 billion and $3 billion By Lau Chi Hang The Chinese name of T-Head, “PingTouGe,” is a popular moniker for the honey badger, recognized by Guinness World Records as the world's most fearless animal. Despite its diminutive size, the creature boasts a robust physical build and a distinctively fierce temperament. It battles massive predators without a trace of fear, and its extraordinary immunity to venom prompts even deadly snakes to give it a wide berth. After Alibaba Group Holding Ltd. (9988.HK; BABA.US) successfully acquired integrated-circuit design firm C-SKY Microsystems in 2018, it merged the company with the chip team at its DAMO Academy to form a new semiconductor entity. At the time, the team intended to name it “Hummingbird.” However, Jack Ma had just returned from a trip to Africa, where the honey badger's tenacious character had left a lasting impression on him. Overriding all other opinions, he had the final word and christened the new company T-Head. According to Alibaba's announcement, it submitted a confidential application to the Hong Kong Stock Exchange in September this year to spin off and list T-Head (Shanghai) Technology Co. Ltd.. The company stated that the spinoff would better reflect T-Head's intrinsic value and allow investors to more easily assess Alibaba's core businesses. Concurrently, T-Head could attract investors specifically focused on the sector and enhance its overall value, which would facilitate more streamlined and convenient financing efforts should the need arise in the future. T-Head's product portfolio is primarily divided into three categories: First, the Zhenwu series of AI chips, which are used for artificial intelligence model training and inference; second, the Yitian series of server CPUs, primarily designed for use in cloud servers and data centers; and third, Xuantie, which mainly provides RISC-V processor core designs and related IP. Annualized revenue exceeds 10 billion yuan Because Alibaba filed its application confidentially, it did not disclose specific operational or financial data. However, during an analyst call early this year, CEO Eddie Wu Yongming offered a glimpse into the unit's operational status. He noted that as of the end of February this year, cumulative deliveries of chips produced by T-Head had topped 470,000 units, pushing its annualized revenue above 10 billion yuan ($1.45 billion). By May, Gao Hui, vice president of T-Head Semiconductor, further disclosed that as of the end of April, cumulative shipments of the Zhenwu series AI chips had reached 560,000 units, serving more than 400 clients across over 20 industries. By August, the Zhenwu series chips had achieved commercial application via Alibaba Cloud among 650 clients in more than 20 sectors — including autonomous driving, the internet, and finance. According to media reports, T-Head is planning a Hong Kong IPO next year and could raise between $2 billion and $3 billion. However, Alibaba has not yet disclosed the specific offering size or listing timeline. Unlocking valuation and facilitating financing Although Alibaba's share price dipped slightly instead of rising following the announcement of the T-Head spinoff, the move should bring more pros than cons for the group over the long term. First and foremost, it will unlock the value of its semiconductor business. Previously, T-Head was largely viewed as just one piece of the Alibaba Cloud puzzle, making it challenging for the market to evaluate. Investors typically appraised it through the traditional valuation framework applied to the parent company, leaving the chip unit significantly undervalued. After the spinoff, based on current market valuations for AI chip firms — even considering the recent sharp pullback from their peaks — T-Head's valuation will still command a premium over that of a traditional internet platform. With Alibaba retaining a majority stake, unlocking T-Head's massive valuation through an IPO will undoubtedly serve the parent company's best interests. Moreover, going public will pave the way for T-Head's future financing endeavors. After all, the research, development, and iterative upgrades of microchips require staggering amounts of capital. Having a publicly listed platform will allow the company to tap into both equity and debt markets and secure bank credit facilities, all of which will streamline its fundraising and borrowing processes. Remaining tucked under Alibaba's wing would force the parent group to continually inject capital, eventually turning T-Head into a financial burden. Strengthening market image In the past, T-Head's chips were designed exclusively to serve Alibaba. As its operations have grown, it has started catering to external clients. Nevertheless, market watchers continue to view it fundamentally as an in-house supplier for its parent company, raising doubts over whether it provides its absolute best products to outsiders and whether it genuinely puts clients first. Operating with relative autonomy post-IPO — backed by its own management team and institutional investors — will help shake off the stigma of being a single-client servant. This newfound independence will reinforce T-Head's relationships with its clients and suppliers while deepening market trust in its brand. Earlier this year, JPMorgan Chase estimated T-Head's potential valuation at $25 billion to $62 billion. However, with the share prices of chipmakers trading at steep discounts compared with the first half of the year, T-Head has missed the prime IPO window. If calculated based on the recent price-to-sales ratios of Moore Threads (688795.SH) and MetaX (688802.SH) — which range from 40 to 70 times — and factoring in Alibaba's estimate of over 10 billion yuan in revenue for T-Head this year, applying a relatively conservative price-to-sales multiple would place T-Head's market capitalization at roughly $60 billion. To subscribe to Bamboo Works weekly free newsletter, click here

芯片估值回歸理性 平頭哥錯失上市良機

阿里巴巴近日公布,將旗下的芯片設計公司平頭哥,獨立分拆在香港上市 重點: 阿里透過保密形式向港交所遞交申請 市場估計,平頭哥將集資20億至30億美元   劉智恒 「平頭哥」是蜜獾的別稱,被健力士世界紀錄大全列為世界最大膽的動物,其身軀雖然細小,但身體結構堅韌,脾性獨特,勇猛善戰,與龐大野獸戰鬥毫無懼色,加上具有超強抵禦毒性能力,縱使毒蛇對它也退避三舍。 2018年阿里巴巴集團控股有限公司(9988.HK; BABA.US)成功收購集成電路設計公司中天微系統後,將其併入達摩院的芯片團隊,整合成立一家半導體公司。當年團隊有意將其命名為「蜂鳥」,碰巧馬雲剛從非洲旅遊回來,蜜獾的性格深深烙入他的腦袋,於是獨排眾議,一鎚定音,以平頭哥作為新公司的名稱。 根據阿里發布的公告,已於今年九月以保密形式向港交所遞交,申請分拆平頭哥上市。公司表示,分拆後更能反映平頭哥自身價值,可讓投資者易於評估阿里的業務。同時,平頭哥可吸引對該業務關注的投資者,並能提升價值,若未來有需要融資也更容易及方便。 平頭哥的產品主要分三大類,一是真武系列的AI芯片,用於人工智能模型訓練及推理,二是倚天系列伺服器CPU,主要是供雲端伺服器及資料中心使用,三是玄鐵,主要提供RISC-V處理器核心設計及相關IP。 年度化收入逾百億 由於阿里是透過秘密方式入表申請,沒有具體營運及財務數據披露,但今年初公司的首席執行官吳泳銘在分析師會議上,亦透露了一點經營狀況。他說截至今年二月底,平頭哥生產的芯片累計交付已突破47萬片,年度化收入可超過百億元人民幣(下同)。 到五月,平頭哥半導體副總裁高慧又透露,截至四月底,真武系列AI芯片累計出貨達56萬片,服務超過400家客戶,覆蓋20多個行業。八月,真武系列芯片通過阿里雲,在自動駕駛、互聯網、金融等20多個行業的650家客戶中,已實現商業化應用。 據媒體報道,平頭哥計劃明年赴港上市,可能募集20億至30億美元。不過,具體發行規模和上市時間,阿里尚未披露。 釋放估值 有利融資 雖然阿里公布分拆平頭哥後,股價不升反微跌,但對集團的長遠來說,分拆應該利多於弊。首先是可以釋放芯片業務的價值,過去平頭哥只被視為阿里雲業務中的一環,市場難於判斷,投資者通常以母企傳統框架去估值,令到其價值被低估。分拆後,按市場對AI芯片估值計,縱使近期已從高位大幅回調,但較傳統互聯網平台的估值仍來得要高,而阿里又持有平頭哥大部分股權,平頭哥上市估值大釋放,對於阿里實屬有利。 另外,平頭哥上市後,有利於自身未來融資發展,畢竟芯片的研發及迭代更替,都要耗費海量資金,若有一個上市平台,可對接股本及債務市場,獲得銀行授信等,均有利於公司的集資及借貸。若一直在阿里旗下,集團要為平頭哥不斷投入資金,反變成阿里的負擔。 強化市場形象 過去平頭哥的芯片只是為阿里服務,隨著發展日益壯大,開始提供予外部客戶。然而,在市場人士看來,始終會感覺你是為母企提供芯片的公司,會懷疑你是否提供最好的產品,以及是否服務客戶為先。上市後運作相對獨立,有自己的管理階層及機構投資者,有助於打破其為單一機構服務的印象,強化平頭哥與客戶及供應商的關係,及對其品牌的信任。 年初時,摩根大通估算平頭哥潛在價值約為250億至620億美元。唯今天芯片企業的股價相對上半年來說已大打折扣,平頭哥錯失最佳上市時機。若以摩爾線程(688795.SH)及沐曦(688802.SH)近期市銷率介乎40至70倍計算,相對阿里估計平頭哥今年逾100億元人民幣的收入,以較保守的市銷率計,平頭哥的市值約600億美元。 欲訂閱咏竹坊每周免費通訊,請點擊這裏
T-Head is a chip maker

平頭哥估值吸引 阿里擬拆出個未來

消息指,中國電商龍頭阿里巴巴正考慮分拆旗下晶片業務平頭哥獨立上市,與另一巨頭百度採取類似策略 重點: 傳阿里巴巴正評估分拆平頭哥獨立上市的可行性,期望躋身一眾高估值中國AI晶片企業行列 競爭對手壁仞科技自1月2日在香港上市以來,股價已接近翻倍,摩爾線程則自去年12月在上海掛牌後,股價累計上漲近四倍   陽歌 如果一把鎖打不開,那就換另一把試試看。這似乎正是電商巨頭阿里巴巴集團控股有限公司(BABA.US; 9988.HK)目前的最新寫照。據報道,該公司正考慮將旗下晶片部門平頭哥半導體有限公司分拆,並尋求獨立上市。 若計劃成真,這項通常被稱為「釋放股東價值」的分拆行動,距離阿里巴巴當年宣布將集團拆分為六大業務板塊、並安排部分或全部獨立上市的大規模重組,已相隔近三年。熟悉公司動向的人都知道,該計劃最終未能落實,阿里巴巴此後基本維持現有架構至今。 頗具諷刺意味的是,如今被納入潛在分拆名單的平頭哥,其實與阿里巴巴的雲智能業務關係密切,儘管平頭哥官網僅將自己描述為「阿里巴巴集團全資擁有的半導體晶片業務主體」。在當初的集團分拆藍圖中,雲智能正是最早放棄分拆的業務之一,當時給出的理由之一,是美國對中國企業實施出口管制,禁止其向英偉達(NVDA.US)等美國供應商採購先進AI晶片。 平頭哥以及越來越多中國晶片新創公司,正試圖透過自主研發AI晶片,也就是正式名稱為圖形處理器(GPU)來填補這一缺口。其中部分公司隸屬於大型科技集團,例如華為旗下的海思,以及百度(BIDU.US; 9888.HK)的崑崙芯。 也有一些是獨立運作的新創企業,例如摩爾線程(688795.SH)、寒武紀(688256.SH)以及壁仞科技(6082.HK),這些公司的股價近期均大幅飆升。壁仞自1月2日上市以來,股價已接近翻倍;摩爾線程則自去年12月IPO後,累計漲幅超過四倍。三家公司目前的市銷率(P/S)均達三位數,其中摩爾線程與壁仞的市銷率更雙雙超過200倍。 相比之下,英偉達目前的市銷率僅為24倍,而被視為二線GPU廠商的邁威爾科技(MRVL.US),市銷率甚至只有9.2倍。中國晶片公司的天價估值,顯然也是阿里巴巴選在此時考慮分拆的重要背景之一,市場普遍預期,中國將為這些企業提供可觀的政策與資金支持,作為擺脫對西方技術依賴的一環。 不過,仍難免讓人懷疑,這些估值是否已明顯偏高,一旦目前港股、滬深股市的升勢告一段落,是否就會迎來修正。 在這樣的背景下,阿里巴巴加快推進平頭哥分拆計劃,也就不足為奇。公司顯然希望趁著估值高企之際順勢出手,籌集可觀資金以維持該業務持續運轉。事實上,幾乎所有GPU公司目前都處於高度燒錢階段,資本市場正是支撐它們繼續運作的主要動力之一。 率先披露消息的彭博報道,在潛在分拆方案下,阿里巴巴將先對平頭哥進行重組,引入員工持股安排,隨後再評估上市可能性。雖然報道指出IPO時間表尚未敲定,但在市場熱度仍高的情況下,預料進程不會拖太久,最快或在未來三至四個月內啟動。 跟上百度腳步 阿里巴巴並非唯一加快推進晶片業務分拆上市的科技巨頭。就在上個月,搜尋引擎龍頭、同時積極布局自動駕駛的百度,也宣布計劃分拆旗下崑崙芯,據媒體報道,該交易集資規模可能高達20億美元。崑崙芯最初主要為百度內部供應晶片,但近年也開始對外銷售產品,例如向中國移動等大型電信營運商供貨。 平頭哥的發展路徑與崑崙芯相近,目前亦已將晶片對外銷售給中國第二大電信商中國聯通。中國移動與中國聯通,正是國內最積極建設新一代數據中心的兩大營運商,這些設施未來將成為AI算力的重要樞紐,因此對晶片需求極為龐大。阿里巴巴的雲業務本身亦在中國及海外營運大量數據中心,這正是當初公司決定建立內部晶片部門的重要原因之一。 潛在的分拆計劃,已為阿里巴巴與百度股價帶來明顯利好。彭博報道曝光後,阿里巴巴在紐約上市的股票於周四上漲5%。過去52周,阿里股價已翻倍有餘,表現跑贏中國最大互聯網企業騰訊同期56%的升幅。近期股價上揚亦將阿里巴巴市值推升至4,230億美元,與騰訊6,930億美元的市值差距進一步收窄。 在市盈率(P/E)層面,阿里巴巴近期的升勢已使其追上騰訊,兩家公司目前同樣約為24倍,高於外賣平台龍頭美團(3690.HK)的19倍,也遠高於Temu母公司拼多多(PDD.US)僅約11倍的水平。 平頭哥與崑崙芯最終將選擇在哪裡上市,仍有待觀察。我們的推測是,兩者都可能以國際化程度較高的香港為首選,一方面反映其全球化野心,另一方面也因香港市場估值相對理性。同時,也不禁令人猜想,華為是否也會考慮為旗下海思晶片進行類似分拆,畢竟其崑崙系列晶片普遍被視為中國目前最先進的產品。 若平頭哥成功分拆,將延續阿里巴巴近來一連串相對正面的發展。在原先的集團分拆計劃告吹後,阿里股價曾低迷一段時間。公司最新業績顯示,截至9月底的季度營收按年增長5%至2,480億元;若扣除兩項正在出售的大型線下零售資產,實際增幅可達15%。 雲業務仍是阿里巴巴最亮眼的板塊之一,季度收入按年增長34%,達約400億元,佔集團總收入約16%。另一個成長動能來自即時零售業務,阿里已將餓了麼外賣整合至淘寶即時零售服務,該業務歸屬於「即時電商」板塊,在9月季度收入按年大增60%至230億元,佔核心電商收入比重升至17%。 總體來看,阿里巴巴這次分拆計劃帶有明顯的順勢而為色彩,背後動力來自投資人當前對中國AI晶片企業給予的高估值。若計劃成行,再加上雲服務與即時零售業務的強勁成長,或可為阿里股價在重拾昔日光環的過程中,提供進一步動能。 欲訂閱咏竹坊每周免費通訊,請點擊這裏