The Maze: Alibaba's AI bet is starting to look less like a research project and more like an infrastructure business. In the June quarter, AI Cloud and Compute Services grew 45% to RMB48.4 billion, its fastest cloud expansion in 22 quarters. AI-related product revenue reached RMB12.4 billion and extended triple-digit growth to a twelfth quarter. The catch sits below the growth line: Alibaba spent RMB67.7 billion on infrastructure, while its model-and-applications arm produced a much larger loss. The stack is monetizing. It is not yet cheap.
Cloud is converting AI demand into recurring revenue. The official results show both total AI Cloud and Compute Services revenue and external-customer revenue rising 45%. Segment adjusted EBITA climbed 133% to RMB5.6 billion, even as Alibaba kept spending on customer growth and technology. That is the commercial proof management needed: more enterprises are paying for compute, models and tools, while the infrastructure layer is gaining operating leverage.
The moat is the stack, not one clever model. Alibaba combined its cloud unit with T-Head Semiconductor, which designs GPU, CPU, storage and networking chips. Its Zhenwu processors now serve more than 650 external customers across over 20 industries. Above that hardware sit orchestration software, cloud infrastructure, Qwen models and agents such as QwenWork, distributed through Alibaba Cloud and DingTalk. The logic is simple: owning more layers can improve performance, reduce supply dependence and keep more of every AI workload's economics inside Alibaba.
Commerce gives that stack a giant proving ground. Qwen Shopping Assistant now covers the path from inspiration to after-sales service inside Taobao, while merchant agents automate listings, store management, advertising and customer service. Alibaba says 250 million users have tried AI-driven shopping features across its commerce integrations. That is not the same as 250 million buyers. It is still valuable: shopping prompts and merchant workflows create demand for inference, generate feedback for models and make Alibaba's cloud products easier to sell with concrete use cases.
The bill is arriving before the full payoff. Capital expenditure jumped 75% to RMB67.7 billion as Alibaba added capacity ahead of agent adoption and paid more for chip components. Free cash flow was negative RMB44.7 billion. Net income fell 75%, although lower investment gains, impairment and provisions also mattered. The sharper split is inside AI: cloud infrastructure earned RMB5.6 billion of adjusted EBITA, while AI Labs and Applications lost RMB13.9 billion as model investment and Qwen inference costs rose.
Why it matters: Alibaba has moved from promising AI demand to showing where the money lands. Cloud growth accelerated from the prior quarter's 40% pace, and infrastructure profit more than doubled. For retailers and brands, the strategic tradeoff is bigger than a better shopping assistant. Alibaba wants to own the prompt, merchant workflow, product data, model feedback and compute bill. The next test is whether 45% cloud growth can keep outrunning the cost of capacity and consumer-scale inference.
