Alibaba’s AI Cloud Revenue Jumps 53% as Next Qwen Trains on T-Head Chips

Adjusted net profit rose 92% to RMB19.9 billion and capex climbed 60%, while Eddie Wu said new US compute rules will affect next year’s cloud services.

WHY IT MATTERS

With cloud growth guided to 30% to 45% next year and its 20 GW compute target under review, Alibaba's AI expansion now depends on how fast T-Head can supply chips.

Alibaba Group’s AI Cloud and Compute Services segment, which combines Alibaba Cloud with the chip design unit T-Head, grew revenue 53% to RMB60.97 billion in the quarter to September 30, according to the company’s results announcement of November 19. Group revenue rose 10% to RMB271.52 billion, a little above the RMB269.81 billion analysts had forecast, Reuters reported via MarketScreener. Non-GAAP net income, the adjusted measure Alibaba reports, climbed 92% to RMB19.91 billion. Net income attributable to shareholders more than tripled to RMB64.54 billion, driven mainly by mark-to-market gains on equity investments and higher operating income.

Chief executive Eddie Wu used the earnings call to describe a change in where Alibaba’s computing power comes from. Alibaba’s Qwen team "is already training its next-generation model on T-Head’s domestic chips," he said, according to the English interpretation in the Motley Fool transcript. After 2027, he added, "most of the incremental AI chip capacity we will add will be primarily proprietary domestic chips." "By 2027, domestic chips will certainly underpin all our training workloads, including both pre-training and post-training," Wu said, according to Nikkei Asia.

Wu also set a slower near-term pace. He said new US compute export rules "will affect next year’s products and services" and put next year’s cloud revenue growth in a range of 30% to 45%, with a rebound to 50% or more in calendar 2028 as Alibaba’s domestic compute capacity grows. He kept the 2030 goal of US$100 billion in external cloud revenue with a margin above 20%; Reuters described the target as a 20% EBITDA margin.

Cloud profit more than doubled while the Qwen unit’s loss tripled

Inside the cloud segment, revenue from external customers rose 64%, and recurring revenue from external customers grew 67%. AI-related product revenue reached RMB17.92 billion, its thirteenth consecutive quarter of triple-digit growth. Segment adjusted EBITA, Alibaba’s measure of operating profit, rose 121% to RMB8.84 billion. The Next Web noted that the segment’s figures include T-Head.

The consumer side of Alibaba’s AI push costs far more than it earns. AI Labs and Applications, a new segment holding the model labs, the Qwen consumer business and the QwenWork enterprise agent, grew revenue 51% to RMB4.44 billion. Its adjusted EBITA loss reached RMB10.50 billion, against RMB3.44 billion a year earlier, which Alibaba attributed to heavier investment in AI capabilities and the inference costs of running the Qwen app. Chief financial officer Hong Xu said on the call that the loss narrowed from the previous quarter.

Spending kept rising. Alibaba’s capital expenditure rose 60% to RMB50.48 billion in the quarter, and free cash flow was an outflow of RMB34.47 billion. A year earlier the outflow was RMB21.84 billion. 36Kr pointed out that capex was still below the RMB67.7 billion spent in the June quarter. Reuters reported that adjusted earnings per American depositary share of RMB8.10 fell short of the RMB11.16 analysts expected, and that China e-commerce revenue fell 5%. The shares rose 3.6% in morning trading after the results, though they remain down about 40% this year.

T-Head becomes the main source of Alibaba’s new compute

Wu’s remarks make T-Head the planned default for Alibaba’s own capacity. He said T-Head is speeding up its three-generation roadmap for graphics processors, and that Alibaba will reserve its own chips for training its foundation models and those of close partners while buying from other domestic suppliers for inference, Nikkei Asia reported. 36Kr, carrying a Zhidx account of the call, quoted Wu as saying that Alibaba’s model service platform Bailian already runs inference on domestic chips and that most future models, for Alibaba, its partners and its customers, will be trained entirely on domestic chips.

The next chip generation sets the pace. Alibaba’s announcement says the Zhenwu V900, unveiled in September, will go on commercial release in the first quarter of 2027, and Wu said it delivers three times the computing performance of the current M890. T-Head is preparing its own Hong Kong listing, a plan EastFrontier covered on November 11.

Wu expects scarcity to work in Alibaba’s favor for a time. "2027, 2028 will be the period in which compute supply is at its scarcest in China’s domestic market," he said, linking that to firm pricing for compute. Exporting T-Head chips is "a viable option," he said, but only after domestic demand is met.

Alibaba may cut its 20 GW compute target and revise its capex plan

The lower range follows the US Commerce Department’s interim final rule of November 16, which requires licenses for advanced American chips sent to data centers anywhere that give China-headquartered companies large amounts of computing power, as EastFrontier reported. Asked about Alibaba’s long-term goal of 20 GW of compute, Wu said changing policies "may lead to a downward revision of that long-term 20 GW target" and that there would be "a certain impact on our overall CapEx plan." Alibaba is still bringing new data centers in the Netherlands, France and Brazil online.

Some targets stand. Wu said Alibaba remains confident of reaching RMB30 billion in annualized recurring revenue for the full fiscal year, which 36Kr tied to its model-as-a-service business. The revised 20 GW target and capex plan, and the first commercial shipments of the Zhenwu V900 in early 2027, will show how much of Alibaba’s growth T-Head can carry.