A model of that scale would be far larger than any publicly announced system today. Parameter counts of frontier models currently released by major labs are believed to sit in the low trillions at most, making Alibaba's target a statement of intent about the scale of its future AI ambitions.
New Zhenwu V900 AI chip
Alongside the model plan, Alibaba unveiled the Zhenwu V900, a new AI chip the company described as the most powerful in China, offering three times the performance of its predecessor. The announcement places Alibaba more firmly in the race to build domestic AI accelerators, as access to advanced foreign chips remains constrained by export controls.
The chip is aimed at powering both Alibaba's own cloud workloads and the massive training runs a trillion-parameter-scale model would demand. Industry watchers see the launch as a sign that Chinese tech giants are increasingly betting on in-house silicon to underpin their AI strategies.
20 GW of data centre capacity by 2032
Wu also said Alibaba aims for its global data centre capacity to surpass 20 gigawatts by 2032, underscoring how central AI compute has become to the company's long-term planning. The target would put Alibaba Cloud's infrastructure ambitions among the largest announced by any single company.
The combined announcements on models, chips and data centres suggest Alibaba is pursuing full control over its AI stack, from hardware through to frontier-scale model training. Analysts expect the plan to intensify competition between Alibaba, other Chinese cloud providers and the US labs racing to build ever-larger AI systems.
