Alibaba’s AI roadmap is getting much bigger than models alone.

Eddie Wu’s (Alibaba Group CEO) is saying that intelligence itself is becoming a scalable utility: machine-generated thinking is still below 3% of total human cognitive volume today, but Alibaba sees a path toward 1,000x human capacity.

That ambition is showing up across 3 layers. Qwen is pushing Recursive Self-Improvement and plans a 5–10T parameter model for more complex, longer-horizon tasks on the road toward ASI. Alibaba also unveiled the Zhenwu V900, which it says is China’s most powerful AI chip, with 3x the performance of the M890 and support for clusters of up to 500,000 cards, alongside expectations for sharply higher annual chip shipments.

And because AI demand is still outpacing infrastructure supply, Alibaba Cloud plans to exceed 20GW of global data-center capacity by 2032.

On compute, Alibaba says its new Zhenwu V900 delivers 3x the M890’s performance, includes 216GB of memory and 1,200GB/s inter-chip bandwidth, and can scale to 500,000-chip clusters. Mass production and commercial release are planned for Q1 2027, while Alibaba expects significant growth in annual AI-chip shipments.

Models, chips, and cloud are now being scaled as one system.