The V900 adds a Chinese accelerator to Alibaba's full-stack strategy as the company targets more than 20GW of data-centre capacity by 2032. Cloud monetization is improving quickly while infrastructure is absorbing cash at a scale that puts returns at the center of the shareholder case.
Alibaba used its 2026 Apsara Conference to turn a familiar AI strategy into a much larger physical commitment. The company unveiled the Zhenwu V900 accelerator, due for mass production in early 2027, and set an objective of operating more than 20GW of data-centre capacity globally by 2032. It had already committed RMB380 billion to cloud and AI infrastructure over three years. The V900 puts proprietary compute more explicitly inside that spending programme.
The financial context is more revealing than the conference headline. In the June quarter, Alibaba's AI Cloud and Compute Services revenue rose 45% to RMB48.4 billion and adjusted EBITA increased 133% to RMB5.6 billion. Segment EBITA margin reached about 12%. At group level, however, capital expenditure rose 75% to RMB67.7 billion and free cash flow was a RMB44.7 billion outflow, with Alibaba attributing the deterioration mainly to higher cloud-infrastructure spending.
Demand is monetizing faster and cloud profitability is improving, while the physical infrastructure required to sustain that growth is pulling cash in the opposite direction. Alibaba is already showing investors both sides of the 20GW plan in the same quarter.
V900 changes the dependency map
The V900 is scheduled for mass production in the first quarter of 2027. Alibaba says the chip delivers roughly three times the performance of its predecessor and can be deployed in very large clusters. Its strategic value comes from reducing reliance on foreign accelerators as U.S. controls constrain access to the most advanced American hardware.
Alibaba also owns enough of the surrounding stack to optimize around its own silicon: cloud orchestration, models, storage, software and customer workloads. That matters because a cloud operator buys system economics, not an isolated benchmark. A chip that trails the best foreign accelerator on one measure can still be commercially useful if the broader stack compensates through availability, software integration or lower total cost.
The burden shifts to Alibaba as a result. More vertical integration means more control over supply, but it also leaves the company responsible for chip performance, yields, software compatibility and the capital efficiency of the infrastructure built around it.
The 20GW target should therefore be read as a financing and utilization problem before it is treated as a revenue forecast. Alibaba has not publicly reconciled that 2032 objective with its present energized capacity, regional power pipeline or expected utilization. Its latest quarter already shows what an aggressive build looks like in cash terms: almost RMB68 billion of capex and negative free cash flow even as cloud margins improve.
The encouraging part for shareholders is that the business consuming the capital is also the business producing Alibaba's fastest growth and improving segment profitability. The unresolved part is whether those gains can outrun depreciation and the cash cost of a build that now reaches far beyond the three-year RMB380 billion programme. A bridge from current energized capacity to 20GW, with associated capex and utilization, would turn today's strategy into something investors can model.