Alibaba's AI cloud business grew 53% in the quarter ended September 30, 2026, but the group's tripled profit is mostly investment gains, and the unit that runs the Qwen models lost RMB 10.5 billion. The numbers come from Alibaba's results release of November 19, 2026. They show strong demand for AI cloud, and a high cost to serve it.
This is a business story, but it matters to builders. Alibaba's Qwen models are among the most used open-weight families, and the company's own words about inference cost explain pressure on free tiers and pricing.
Update (November 19, 2026): earnings call guides cloud growth to 30-45% and flags a possible cut to the 20 GW target
On the earnings call, CEO Eddie Wu said new US compute export control rules will hurt next year's plans, according to the call transcript published by Investing.com. Three points from that transcript:
- Cloud growth guidance. Wu said "we might expect our growth target for cloud revenue to be in the 30%-45% range" for the next fiscal year, with a rebound to 50% or higher in calendar 2028 as domestic compute scales.
- 20 GW target may fall. Wu said "based on our current preliminary assessment, it may lead to a downward revision of that long-term 20 GW target," with "a certain impact" on the overall capex plan, while Alibaba pushes harder on domestic compute and its supply chain. This is a "may", not a decision.
- Two growth figures. The 64% figure is revenue from external customers in AI Cloud and compute services. The 53% in our headline is total AI Cloud and compute revenue, per the same transcript.
We could not open the Benzinga and Bloomberg framings, which say the chip rules "could force" a scale-back, so we rely on the transcript wording. For the US rules themselves, see our Samsung Foundry and BIS rules explainer. Practical meaning: if Alibaba cuts capacity plans, expect tighter Qwen API and cloud GPU supply in China, not a change to open-weight Qwen releases, which the call did not address.
Update (November 19, 2026): most new AI chip capacity after 2027 will be domestic, T-Head first for training
A second part of the same call answers where Alibaba will get its chips. In the Q&A, answering a Macquarie question, Wu said "most of the incremental AI chip capacity we will add will be primarily domestic or proprietary domestic chips," per the call transcript on Singju Post. He split the work by job:
- Training. "Our own flagship proprietary chips will be used as a first priority to support the training and iteration" of Alibaba's models. In the prepared remarks he said "Qwen is already training its next-generation model on T-head's domestic chips."
- Inference. Alibaba will support inference "with a wider variety of domestic chips, including domestic chips from other brands."
- Timing. Wu expects mass production of the next-generation Zhenwu V900 in the first quarter of calendar 2027.
- The squeeze. He described 2027 and 2028 as "the period in which compute supply is at its scarcest in China's domestic market."
What is not confirmed: no performance or yield figures for the V900 were given on the call, and independent benchmarks of T-Head chips against Nvidia parts have not been published. For builders, the practical read is the same as above: expect China-side Qwen API and cloud capacity to stay tight through 2028, while open-weight releases are a separate question the call did not address. Questions people ask: Will Alibaba stop using Nvidia chips? The call said most incremental capacity will be domestic, not that existing Nvidia capacity goes away. When does the shift start? Wu pointed to after 2027, with V900 mass production in early 2027.
The numbers at a glance
| Metric (quarter to Sept 30, 2026) | Result | Change year on year |
|---|---|---|
| Revenue | RMB 271.5 billion (US$40.5 billion) | +10% |
| AI Cloud and Compute Services revenue | About RMB 61.0 billion | +53% |
| External-customer cloud revenue | n/a | +64% |
| Recurring external revenue (no compute equipment) | n/a | +67% |
| Cloud adjusted EBITA margin | 14.5% | Up from 10.0% (MLQ.ai) |
| Income from operations | RMB 29.1 billion | +443% |
| Net income | RMB 64.4 billion | +213% |
| Non-GAAP net income | RMB 19.9 billion | +92% |
| Free cash flow | Outflow of RMB 34.5 billion | Outflow was RMB 21.8 billion a year earlier |
| AI Labs and Applications adjusted EBITA | Loss of RMB 10.5 billion (MLQ.ai) | Loss was RMB 3.4 billion a year earlier |
Figures in RMB and US dollars follow Alibaba's release as reproduced on StockTitan and Yahoo Finance, and segment detail follows MLQ.ai.
What did Alibaba say about AI?
CEO Eddie Wu said in the release that "full-stack AI capabilities continue to improve commercialization." He pointed to 67% growth in recurring external cloud revenue and said AI-related product revenue delivered triple-digit growth for the thirteenth consecutive quarter. MLQ.ai puts that revenue at RMB 17.9 billion.
Wu also listed recent upgrades across proprietary chips, supernode architecture, agent-native cloud infrastructure, AI models and agents, and said T-Head unveiled its next-generation Zhenwu V900 AI chip. CFO Toby Xu said the company will keep disciplined investment "with a focus on return on invested capital while targeting to achieve positive free cash flow."
Those are company statements. The tables and notes are where you see what the company paid for that growth.
Why is the 213% profit jump misleading?
Net income rose 213%, but Alibaba says the rise was primarily from higher net investment income and income from operations. MLQ.ai quotes the release: interest and investment income was RMB 73.3 billion, up 265%, mainly from mark-to-market gains on equity investments.
Non-GAAP net income, which excludes those items, rose 92% to RMB 19.9 billion. That is still strong, but it is less than half the headline growth rate. Operating income rose 443%, which reflects improved results at Taobao Instant Commerce and cloud, as Alibaba describes it.
If you compare companies, use the operating and non-GAAP lines, not the 213%.
How big is the AI spending bill?
Enterprise rollout of cloud capacity, illustrating Alibaba AI cloud spending and capex in the quarter
Free cash flow was an outflow of RMB 34.5 billion, against RMB 21.8 billion a year earlier. Alibaba attributes the larger outflow mainly to higher cloud infrastructure spending, partly offset by smaller losses in Taobao Instant Commerce and more customer prepayments for cloud services. Cash and other liquid investments stood at RMB 487.9 billion at quarter end.
MLQ.ai reports capital expenditure rose 60% to RMB 50.5 billion. Over six months the free cash flow outflow was RMB 79.1 billion, compared with RMB 40.7 billion a year before.
Put simply: AI cloud demand is growing, and Alibaba is buying capacity ahead of it. The company says it is targeting positive free cash flow, but gives no date in the passages we read.
How does this quarter compare with the one before?
MLQ.ai built a three-column comparison from the release. The June-quarter column is MLQ.ai's own arithmetic (six-month totals minus the September quarter), because Alibaba reports to the SEC twice a year as a foreign private issuer. With that caveat, in RMB billions:
| Line | Sep 2026 quarter | Jun 2026 quarter | Sep 2025 quarter |
|---|---|---|---|
| Revenue | 271.5 | 269.0 | 247.8 |
| Income from operations | 29.1 | 15.2 | 5.4 |
| Adjusted EBITA | 32.6 | 27.3 | 9.1 |
| AI Cloud and Compute revenue | 61.0 | 48.4 | 39.9 |
| Non-GAAP net income | 19.9 | 20.7 | 10.4 |
Two things stand out. AI Cloud revenue jumped from about 48 billion to 61 billion in one quarter, a much larger step than group revenue, which moved only from 269 to 271.5. And non-GAAP net income slipped slightly from the June quarter, even as operating income roughly doubled, which fits the rising cost of AI investment sitting below the operating line. Treat the June column as approximate.
What happened outside AI?
The release credits much of the operating improvement to Taobao Instant Commerce, Alibaba's quick-commerce business, whose losses narrowed substantially while it kept market share, and to solid growth in marketplace e-commerce. Customer management revenue, the core advertising line, fell 5% year over year. Excluding the contra revenue effect of a new business development program, Alibaba says it would have grown 1% like for like.
That matters for context. The AI story is growing inside a group whose largest advertising line is flat. Cloud is carrying the growth narrative, and the margin gains in quick commerce are what make the group's operating income look strong. If you are judging whether AI is paying for itself, the answer from these numbers is partly: cloud margins are rising, but the AI labs segment is a large loss and cash flow is negative.
What does the Qwen segment loss tell us?
Alibaba reorganized its reporting. According to MLQ.ai, AI Cloud and Compute Services now combines the former Cloud Intelligence Group with chip designer T-Head, and AI Labs and Applications groups the model labs, the Qwen consumer business and the QwenWork agent platform.
That second segment had revenue up 51% to RMB 4.44 billion, but an adjusted EBITA loss of RMB 10.5 billion, up from RMB 3.4 billion a year earlier. MLQ.ai quotes Alibaba blaming "increased investment in AI capabilities, and higher inference cost related to Qwen app." MLQ.ai adds that the loss narrowed from the June quarter, though its figures for that quarter are its own calculation from the six-month totals.
The takeaway for builders: serving a popular AI app is expensive. Free usage of a consumer chatbot is a cost center, and the cost shows up as a segment loss. That helps explain why labs keep adjusting limits and plans. See our coverage of Qwen3.5 9B beating larger models on fraud detection for the other side of the trade: small models that cost less to run.
What about the T-Head chip listing?
A custom AI chip with fine traces, representing the T-Head Zhenwu V900 mentioned in Alibaba's results
The results release mentions that T-Head unveiled the Zhenwu V900. A Wall Street Journal report, as summarised by Techmeme, says Alibaba is preparing a Hong Kong listing for T-Head. We did not find that listing in the results text, and we could not read the Journal's report, so treat it as reported, not confirmed by the company.
Chips matter here because Chinese AI labs depend on domestic supply under export limits. Our posts on Chinese labs rushing for Nvidia capacity overseas and the Alibaba, ByteDance and Tencent RTX Pro orders show the wider supply picture. A competitive in-house chip would lower Alibaba's reliance on that supply.
What this means for builders and founders
Builders using Qwen or Alibaba Cloud. Demand for AI cloud is strong and Alibaba says the margin is improving, with cloud adjusted EBITA margin up to 14.5%. That supports continued investment in the platform. The inference cost comment suggests that free consumer access is a place where terms can change.
Founders comparing providers. Watch the capex line. A provider that outspends its cash flow needs customer prepayments and sustained growth. Alibaba says customer prepayments for cloud services rose, which is a sign that enterprise buyers are committing.
Marketers and analysts. Do not quote "profit up 213%" without the non-GAAP figure. The two tell different stories.
What is confirmed and what is not
From Alibaba's release (as reproduced by StockTitan and Yahoo Finance): revenue, AI cloud growth, operating income, net income, non-GAAP net income, free cash flow, and the executive quotes.
From MLQ.ai's reading of the release: the segment split, the AI Labs and Applications loss, the capex figure, and the June-quarter comparison, which MLQ.ai calculated itself.
Not confirmed: the T-Head Hong Kong listing, which comes from a Wall Street Journal report we could not read. We could not open the SEC filing or the original Business Wire page directly, so we rely on reproductions of the release.
Our read
AI cloud growth of 53% is the real signal. The profit headline is mostly investment marks. The RMB 10.5 billion Qwen loss and the RMB 34.5 billion cash outflow are the price of building out and serving AI at scale. Alibaba says it will target positive free cash flow. Whether it does depends on whether cloud demand keeps outrunning spending.
Related reading
- Coinbase's fine-tuned Qwen3.5 9B on fraud
- Chinese AI labs rush for Nvidia capacity overseas
- ByteDance, Alibaba and Tencent RTX Pro 5500 orders
- Alibaba OpenGCUA computer-use agents
- Alibaba CommerceGen Studio
- IEA: $4 trillion data center forecast
Figures are from Alibaba's November 19, 2026 results release as reproduced by third parties, and MLQ.ai's analysis. Check the filed report for exact wording.
