Ali was forced to "stop"
EmphasizeNext
2026.11.20 11:39
· From Beijing
Full text 4496 words
Text | EmphasisNext
At the earnings conference call on the evening of November 19, Alibaba Cloud delivered the fastest single-quarter growth rate in the past five years: 53%. But then, management lowered its cloud revenue growth target for the next fiscal year (April 2027 to March 2028) to 30% to 45%.
Also lowered are expectations for infrastructure construction. Alibaba's previously proposed long-term energy consumption target of 20GW data centers may be reduced. The growth rate of cloud business will not return to more than 50% until 2028, and the time point for free cash flow to become positive has also been pushed to the second half of 2028. Only the long-term goal of external cloud revenue of US$100 billion and a profit margin of over 20% in 2030 remains unchanged.
When the growth rate has just hit a new high, it takes the initiative to lower expectations for next year. The bottleneck is supply. Wu Yongming bluntly stated at the meeting: Computing power is in short supply, especially in China, where it will be extremely scarce in the next two years.
Just three days before the conference call, new regulations on global computing power cut off Chinese companies from leasing large-scale computing power overseas. In the same week, Nvidia confirmed that it would stop shipping previously licensed Hopper chips to China. When external computing power sources are completely locked, how fast Alibaba Cloud can run has little to do with how many orders it takes, but only depends on how many cards it can create. This layer of hard constraints is rewriting the rankings in the domestic cloud market and forcing Alibaba to recalculate the computing power quotas of Qianwen and Taobao.
Alibaba Cloud’s upper limit is bet on Pingtou Ge’s production line
Looking at demand alone, Alibaba Cloud is at its best right now. Revenue from external customers increased 64% year-over-year in the quarter, accelerating for the 10th consecutive quarter. AI-related product revenue reached 17.9 billion yuan, accounting for nearly half (49%) of external recurring revenue, and annualized revenue crossed the US$10 billion threshold.
External orders are still pouring in, but next year's growth guidance has dropped to 30% to 45%. The gap in the middle can only be filled by domestic chips. Wu Yongming made it clear during the conference call that the new export control regulations will affect product supply next year. Alibaba’s 2030 AI infrastructure plan is completely based on the supply chain of Pingtou Ge’s domestic chips.
For the first time, he released the third-generation mass production schedule of Pingtou Ge: this year, the "Zhenwu M890" will be fully launched, and the "V900" and "J900" will be mass-produced in 2027 and 2028 respectively. At present, Qianwen has trained the next generation model on Pingtouge chip, and the inference service of Bailian platform has also been switched to domestic cards.
Stacking the chip schedule and revenue guidance together, Alibaba Cloud's operating rhythm in the next two years will be on track: 2027 will be a lean period before V900 production capacity ramps up, and revenue growth will be limited by the progress of card delivery. By 2028, V900 and J900 will gradually increase in volume, and the growth rate will return to more than 50%.
The guidance range of 30% to 45% also hides Ali's cash flow calculations. In a conference call in August this year, Wu Yongming once calculated the return on investment of the cloud business: If the payback cycle of a single card is three years and the growth rate of the cloud business is within 33%, positive cash flow can be maintained. If the payback period is shortened to less than 2.5 years and the growth rate reaches above 40%, it will be able to generate blood on its own.
The growth rate range for next year given this time is stuck on the return to capital line of 33% to 40%. Pingtou Ge has undertaken a dual task in Alibaba’s ledger: not only to fill the computing power gap after the supply of imported chips was cut off, but also to rely on self-research to reduce the cost of a single card and shorten the payback cycle to less than 2.5 years. This can also explain why Alibaba has set "growth rate back to 50%" and "free cash flow positive" at the same time in 2028, because only when the shipment scale and cost of self-developed chips cross the critical point, high growth and positive cash flow will not fight each other.
Capital expenditures this quarter already reflect the difficulty of getting cards. Alibaba’s capital expenditures this quarter were 50.5 billion yuan, a decrease of 17 billion yuan from the previous quarter. CFO Xu Hong explained that this is a normal procurement cycle fluctuation and not a strategic contraction. However, management also admitted that the main reason why the 20GW target may be lowered is chip constraints. As of the end of the quarter, Alibaba still had US$32.9 billion in net cash on its books. Alibaba does not lack money to buy cards, but it lacks cards that are available on the market.
This is also another reason why Wu Yongming rarely revealed the third-generation chip trump card of Flathead Brother in a group conference call at once. In the past, most of the self-developed chips of major manufacturers were hidden behind the scenes, and they rarely directly reported the specific chip code names and schedules for the next two to three years to the secondary market. Now the management has publicly tied Alibaba Cloud's long-term plan to return its growth rate to more than 50% in 2028 and revenue of hundreds of billions of dollars in 2030 to Brother Pingtou, which is equivalent to directly issuing an internal guarantee order with a high degree of certainty to Brother Pingtou.
On the path of Pingtou Ge to promote independent financing and even go public, this mass production plan, which is deeply tied to Alibaba Cloud and Qianwen, is the hardest commercialization road show. Once independent capital channels are opened, the high tape-out investment and production capacity advance payment from generation to generation can be shared by external equity funds, thus reducing the pressure on Alibaba Group's profit statement and free cash flow. At the same time, the strategic allotment and equity space of independent listed entities is also the in-depth interest binding of Pingtou Ge with domestic wafer factories and advanced packaging factories, and the hard bargaining chip for priority scheduling in advanced process production lines that are in short supply locally.
Computing power has shifted to a seller's market, and cloud vendors have rearranged their rankings based on whether they have cores or not.
In the past two years, the theme of the domestic AI cloud market has been price wars and competition for usage. According to this style of play, ByteDance's Volcano Engine has the strongest momentum: IDC data shows that the domestic MaaS market size in the first half of 2026 was 10.63 billion yuan, and Volcano Engine took 45.1% of the revenue share, nearly 20 percentage points higher than the second-place Alibaba Cloud. In the overall AI cloud market including underlying IaaS computing power, Alibaba Cloud still ranks first. According to Omdia, its revenue share in 2025 is 38.1%. Huoshan received the largest share of Token distribution, and Alibaba maintained the underlying computing power base.
The new regulations that came into effect on November 16 directly removed the foundation of the price war in the previous two years.
According to the new requirements of the US BIS, any overseas data center that provides more than 12 million TPP computing power to Chinese headquarter companies must apply for a chip export license. This threshold is only equivalent to about 750 H100s. At the same time, the review of overseas wafer foundries and advanced packaging has been further tightened. Chip design companies in restricted areas must enter the BIS "Approved IC Design Enterprises" white list in order to be exempted from tape-out approval on a piece-by-piece basis.
In addition, Nvidia CFO Colette Kress confirmed in the same week that it would stop shipping approved Hopper chips to China. The three routes of overseas card rental, direct purchase of cards, and overseas tape-out of major Chinese manufacturers have been narrowed at the same time, leaving almost only domestic chips for new computing power.
Domestic cloud vendors have thus divided into two camps.
There are only a few people who have self-developed chips and clear mass production schedules: Alibaba has Pingtou Ge, Huawei has Shengteng, and Baidu has Kunlun Core. Large manufacturers that do not have self-developed large chips have to lock up goods in the market: Tencent mentioned in the second quarter conference call that its computing power comes from leasing, purchasing imported high-end GPUs and more domestic GPUs. In the third quarter, it used large advances to grab production capacity. ByteDance had previously planned to deploy about 36,000 B200s through local cloud providers in Malaysia. After the new regulations were implemented, this overseas borrowing plan was the first to bear the brunt.
The anxiety of external manufacturers about card shortage has in turn increased the urgency for Pingtou Ge to go public independently. When ByteDance, Tencent, and second-tier cloud vendors are all locking up domestic GPU production capacity in the market, if Pingtou Ge is always just an internal business unit of Alibaba Cloud, external cloud vendors and Internet giants will find it difficult to safely hand over the underlying computing power to it due to concerns about horizontal competition. Only through independent financing and listing, and the introduction of external strategic shareholders and industrial capital to achieve a neutral governance structure, can Pingtou Ge truly step out of the Alibaba system and exploit the external market left after Nvidia cut off its supply as an independent commercial chip company.
The MaaS price war can only start if there is a steady supply of new cards on the market. Whoever does a good job in reasoning and optimization will dare to lower prices to grab share. Once the newly added computing power becomes a seller's market, the players with the largest call volume and the lowest token price will face the largest computing power gap.
Alibaba took the initiative to lower its growth guidance for next year this time. It was just the first to write the hard constraints on chip production capacity into its open-label ledger. Other major manufacturers face the same hurdle. In the next two years, the focus of domestic AI cloud competition will shift from grabbing demand to ensuring supply. But even manufacturers with self-developed chips have to face another internal problem: Who can use the card that is finally offline?
For a card, should it be given to Qianwen first or to the customer first?
Within Alibaba, the same Pingtouge chip is placed on the table, but there are two completely different accounts: it is installed in Alibaba Cloud and rented to external customers, which can realize a segment profit rate of 14.5%. The Qianwen large model allocated to our own family is now a pure investment.
When the computing power is sufficient, each of the two lines calculates its own account. But when the chip supply reaches Alibaba Cloud's revenue limit, every time Qianwen takes up one more card, Alibaba Cloud will have less money to sell one card. The computing power consumed by Qianwen has become the resource with the highest opportunity cost for the entire Alibaba Group.
Alibaba’s actions to shrink C-side general traffic can already be seen in the financial data. This quarter, Alibaba's AI laboratory and application division lost 10.5 billion yuan, which was three times that of the same period last year, but narrowed by a quarter compared with about 13.9 billion yuan in the previous quarter, and segment revenue increased from about 3.3 billion yuan to 4.4 billion yuan.
The other side of the narrowing of losses is that Qianwen’s user growth has almost stagnated. QuestMobile data shows that from March to June this year, Qianwen’s monthly activity only increased slightly from 166 million to 167 million, while the monthly activity of ByteDoubao reached 382 million during the same period. In this quarter’s financial report, Alibaba simply did not disclose Qianwen’s user data.
Under the computing power rationing system, Qianwen has reduced the use of free computing power in exchange for PanQ&A monthly activity, and instead focuses its computing power on scenarios that can directly generate transactions and income. The cloud-based personal intelligence Qwen Go!, which was launched in October this year, and QwenWork, which includes Taobao merchant tools, are both taking a path closer to transactions. When every card has a clear opportunity cost, the input-output ratio must be calculated clearly for every large model inference.
The exemption clauses in the new U.S. regulations also add a layer of variables to Alibaba’s internal distribution of computing power. The new regulations exempt traditional purely commercial recommendation systems, and conventional recommendation algorithms for e-commerce and social platforms do not need to apply for a license. But as long as the system can generate new text or images based on user input, it no longer enjoys the recommendation exemption.
This means that Taobao’s traditional “guess you like” can still continue to run in the compliance-exempt computing power pool. The Qianwen-style conversational shopping guide and Agent’s automatic order placement belong to generative AI, which fall within the control red line and must all be returned to Pingtou Ge’s domestic computing power pool. The deeper Taobao goes into AI e-commerce, the more prominent the conflict with Alibaba Cloud's external customers in grabbing domestic cards.
Taobao needs to find another way to make money, but it has to wait for chips to increase in volume
Alibaba Cloud's growth rate is slowing down and free cash flow will not turn positive until the second half of 2028, which means that in the next two years, Taotian E-commerce will continue to serve as the group's only cash cow. This quarter, the e-commerce group contributed 38.85 billion yuan in adjusted EBITA, almost supporting all of Alibaba’s operating profits.
But this cash cow’s traditional means of monetization are almost exhausted. This quarter, Taobao and Tmall's customer management revenue (CMR) apparently dropped 5% year-on-year, mainly because the subsidies originally included in sales expenses were changed to directly offset revenue. Even if it is restored to the same standard, CMR only increased slightly by 1% year-on-year. Thanks to the low base brought about by the cooling of the food delivery subsidy war, e-commerce profits have increased significantly year-on-year, but they have still fallen slightly from the 39.75 billion yuan in the previous quarter. The dividends of tools that increase monetization rates, such as site-wide promotion and software service fees, have been basically exhausted, and the proportion of traditional shelf e-commerce companies that can extract from each transaction has peaked.
The real-time retail approach is also converging. Instant retail revenue increased by 22% in the quarter, mainly driven by Hema. Management's evaluation of the unit economic benefits (UE) of the takeaway business was "in line with expectations." Just two days ago, JD.com disclosed that its new business losses narrowed significantly by 38.7% year-on-year. The two most aggressive attackers in the past year have stopped at the same time, and the food delivery business has shifted from burning money to grab share to repairing their respective profit statements. For Alibaba, the current core strategic value of flash sales is to serve as a high-frequency base for Qianwen’s “service” capabilities on the contract fulfillment side in the same city.
In order to break the monetization ceiling of shelf e-commerce, Taobao has bet the next generation of commercialization on Agent transactions. After Qianwen fully integrated with Taobao in May this year, users can already select products, compare prices and place orders in one sentence. ByteDance is also simultaneously testing Doubao directly connected to Douyin e-commerce for placing orders. Whoever gets through the AI shopping portal first will get the pricing power for the next generation of e-commerce traffic.
What’s blocking Taobao’s shift is still the computing power schedule. Every order placed by Agent e-commerce requires multiple rounds of large model inference calls across the product library, discount calculation, and fulfillment systems, consuming exactly the limited computing power that will be in short supply next year. Before the large-scale release of Pingtouge V900 in 2027, Taobao's AI can only be advanced in small steps according to the rationing system. The main position must rely on traditional chips to defend the basic market: 66 million 88VIP high-spending members, Xianyu, whose daily activity has increased by 23%, and the flash sale intra-city fulfillment network that has been rolled out.
In the end, several core accounts of Alibaba in the next two years: Alibaba Cloud's revenue growth, Qianwen's commercial scale, Taobao's speed of cutting traffic to Agents, and the point at which the group's free cash flow turns positive are all tied to the same production line of Pingtou Ge. Before domestic chips can truly fill the computing power gap, Taobao's profit base must last long enough.
Respect for the original work, if you are qualified, you should be appreciated.
According to the real-name requirement of the Cybersecurity Law, please bind your mobile phone number before posting a comment.
Express
11:48
Breaking the record, China achieves wireless communication verification with the highest code rate in the 6G frequency band
Malaysian Prime Minister Anwar arrived at the meeting
11:47
Xi Jinping meets with Thai Prime Minister Anutin
Qiyuan Robot signed a strategic cooperation with the Postal and Telecommunications Equipment Consumer Electronics Division
11:46
A-share lunch review: Shenzhen Component Index and ChiNext Index both fell by more than 2%, and the precious metals and liquor sectors performed sluggishly.
11:35
The relevant person in charge of Byte refutes rumors that Xie Xin and Zhang Nan have resigned: the information is inaccurate
11:32
The main domestic futures contracts rose or fell, with palladium and Shanghai silver falling nearly 9%.
11:28
Meta plans to launch Muse Charm, priced at $300
11:27
Spot gold fell below $3,520 per ounce, down nearly 3% on the day
11:23
Over 60% of China's sustainable development indicators are close to or achieving the 2030 goal
11:14
The concept of AI comic drama rebounded during the session, and Guomai Culture rose by more than 10%
11:13
Procter & Gamble plans to cut manufacturing and logistics employees by up to 30% by 2030
11:04
National Energy Group established a hydrogen energy technology company in Ordos with a registered capital of approximately 350 million
11:00
Domestic gold jewelry price falls below 1,100 yuan per gram
10:55
Korean exchange launches SIDECAR mechanism and suspends KOSPI programmed selling
10:54
Pork concept rebounded in intraday shock, Jin Xinnong went out of 3 boards in 6 days
10:51
22 private equity firms appeared in the first batch of 15 active ETFs, holding a total of 524 million shares
10:48
Alibaba Q2 of fiscal year 2027: Xianyu daily activity increased by 23% year-on-year, and order volume increased by 29%
10:47
Thai Prime Minister Anutin arrived at the meeting
10:46
Five departments issued the 2026 version of the "Guidance Catalog" for marine industry investment and financing
Scan to download App




