Image Not FoundImage Not Found

A technician works at a desk in a quiet data-center operations room with server cabinets behind glass.

Alibaba’s AI Cloud Turns Profitable at Scale, but Qwen Losses Test the Full-Stack Bet

Alibaba’s September-quarter results gave investors one of the clearest looks yet at how AI can become an operating business rather than just a capital-spending story. In its newly recast AI Cloud and Compute Services segment, revenue rose 53% year over year to RMB60.973 billion and adjusted EBITA jumped 121% to RMB8.838 billion. But the same report also showed why the market is still cautious: AI Labs and Applications, the unit that houses Qwen and related products, posted a far larger adjusted EBITA loss of RMB10.503 billion.

That split matters beyond one quarter. The practical question for customers, partners and investors is no longer whether Alibaba has AI demand. This quarter says it does. The harder question is whether that demand can become durable profit once the company pays for models, chips, inference and the consumer and enterprise software needed to keep the stack strategically relevant.

In Alibaba Group’s quarterly results announcement, the company also reported 10% group revenue growth and a 260% rise in consolidated adjusted EBITA. Reported net income attributable to ordinary shareholders climbed to RMB64.535 billion from RMB20.990 billion, but Alibaba said net investment income was a major contributor. That makes the segment figures more revealing than the headline profit figure for anyone trying to understand the economics of AI.

The infrastructure layer is doing more than absorbing spending

The strongest evidence in the report sits inside the cloud numbers. External-customer revenue in AI Cloud and Compute Services rose 64%, faster than the segment as a whole, and recurring revenue grew 67%. That is a more meaningful signal than a broad AI growth claim because recurring revenue suggests repeat workloads, not just short-lived testing or one-off equipment purchases. Compute-equipment-solutions revenue also rose 55%.

The margin trend is just as important. Adjusted EBITA margin in the segment reached 14.5%, up from roughly 10% a year earlier. A business that is adding demand while widening margin is not merely filling data centers; it is showing pricing power, utilization improvement, mix shift, cost discipline or some combination of the three. Alibaba does not break that mix out, so investors cannot tell how much of the improvement came from price, volume or product mix. Still, a positive and rising margin sets this quarter apart from many AI narratives that stop at usage growth.

There is another reason the external mix matters. Alibaba’s quarter table shows 14% of AI Cloud and Compute Services revenue came from internal customers. In a vertically integrated company, internal demand can help absorb capacity and sharpen products, but it does not answer the market’s core question about third-party adoption. External customers did. Alibaba said clients in internet, automotive, embodied AI, financial services and biotechnology expanded their use of its services, and it said AI-related product revenue delivered triple-digit growth for the thirteenth straight quarter.

Readers do need to keep one accounting caveat in view. Alibaba changed its reporting structure in 2026, combining cloud infrastructure with T-Head, its chip and hardware arm, into AI Cloud and Compute Services. It also grouped model labs, the Qwen consumer business and QwenWork into AI Labs and Applications. The new presentation is strategically useful because it shows the full value chain more clearly, but it makes long-run comparisons more dependent on recast definitions.

The application layer is where the bill is landing

If the cloud business is the proof of monetization, the labs business is the reminder that full-stack AI is expensive long after demand appears. AI Labs and Applications revenue grew 51% to RMB4.435 billion, but its adjusted EBITA loss widened to RMB10.503 billion from RMB3.435 billion a year earlier. Alibaba attributed that deterioration to investment in AI capabilities and higher inference costs related to the Qwen app.

Those figures explain the strategic tradeoff better than any slogan could. Alibaba is trying to sell chips, compute, models, agent software and commerce-linked services into the same ecosystem. That can improve utilization and customer lock-in over time. It also means the company may monetize infrastructure before it monetizes applications. Enterprises often buy capacity and model access early, while consumer products and productivity tools can take longer to find pricing, retention and cost structures that work at scale.

For now, the labs loss is not a rounding error. It is larger than the cloud segment’s adjusted EBITA by about RMB1.7 billion. That does not mean the cloud gains are unimportant; it means the company is still choosing to reinvest heavily down the stack. What the results do not show is which part of AI Labs and Applications is responsible for most of the burn. The segment combines model research, Qwen consumer products and QwenWork, and Alibaba did not provide a standalone cash-flow measure for that effort.

What would make the model durable

The quarter makes a stronger case that Alibaba’s AI infrastructure business is real, external and increasingly profitable. It does not yet prove that the full AI stack is producing attractive returns. The scorecard from here is straightforward.

First, external recurring revenue needs to stay strong as the comparison base gets harder. Second, cloud margins need to hold up even if Alibaba spends again on capacity, sales or customer incentives. Third, the labs loss needs to narrow relative to labs revenue, or at least stop widening this quickly. Fourth, investors should keep watching the boundary between internal and external demand. A vertically integrated stack is a feature, but paid third-party adoption is still the cleaner test of commercial durability.

The supply side matters too. Alibaba unveiled the Zhenwu V900 AI chip during the quarter, with commercial release planned for the first quarter of 2027, and it released Qwen4-Max on Nov. 16. Those are signs of ambition, not yet proof of production scale or competitive performance. At the same time, Associated Press reported that Alibaba warned new U.S. rules could affect some global data-center expansion plans, while Tencent has also described advanced-compute availability as constrained. Strong demand does not automatically mean unconstrained capacity.

That wider context helps explain why the stock market is not treating this as a simple win. Reuters reported that Alibaba beat September-quarter revenue estimates and that investment gains helped more than triple reported profit, yet the shares were still down about 40% for the year amid concern about China consumption and AI spending. Reuters also noted that e-commerce still accounts for nearly three-quarters of Alibaba’s revenue, a reminder that AI progress, however important, is not the whole company.

What Alibaba showed this quarter is the order in which a full-stack AI strategy may pay out. The cloud gets paid first. The models and apps consume cash longer. That is enough to validate the monetization story at the infrastructure layer. It is not yet enough to say the entire stack is earning its keep.