platform updates · September 22, 2026

Alibaba plans a 10-trillion-parameter AI model, unveils Zhenwu V900 chip

Alibaba CEO Eddie Wu used the company's Apsara Conference in Hangzhou on Tuesday to announce an ambitious compute-and-scale roadmap: plans to train a new AI model with 5 trillion to 10 trillion parameters from the Qwen team, the unveiling of the Zhenwu V900 AI chip — which Wu called the most powerful in China, with three times the performance of its predecessor — and a target of more than 20 gigawatts of global data-center capacity for Alibaba Cloud by 2032.

The 5–10T figure is a plan, not a launch: per Reuters, Alibaba is still working on the model architecture and data optimization — training has not begun, and there are no benchmarks, release dates, or pricing attached. The Zhenwu V900 comes from Alibaba's T-Head semiconductor unit; its predecessor, the Zhenwu M890, shipped only in May. The 3× performance claim is Alibaba's own — no independent benchmarks were cited — and the "most powerful in China" framing is Wu's, not a measured global comparison.

Context matters here: a plan for a model ten times the size of today's largest, announced alongside a custom chip and a six-year, 20-gigawatt infrastructure target, is capacity signaling — Alibaba telling the market (and competitors) that it intends to fund and power its way into the top tier of the model race. For context, Startup Fortune notes Qwen's current Qwen3.8 Max runs on 2.4 trillion parameters, so the target would more than double it.

Why it matters

When one of the world's largest cloud providers commits — publicly, at its flagship conference — to 10-trillion-parameter ambitions and a China-built chip stack, it sets the price-war and sovereignty conversations in motion. Track it as a watch item: nothing shipped today, but the follow-ups on silicon availability and the training cluster will matter to anyone pricing AI workloads.

Key facts

Announced September 22, 2026 at Alibaba's Apsara Conference in Hangzhou by CEO Eddie Wu
Plan to train a 5–10 trillion-parameter AI model from the Qwen team — still working on architecture; training has not begun
Zhenwu V900 chip unveiled by T-Head: claimed "most powerful in China," 3× the performance of predecessor (Alibaba's claims, no independent benchmarks)
Target: more than 20 GW of global Alibaba Cloud data-center capacity by 2032
Context: Qwen's current Qwen3.8 Max runs on 2.4T parameters — the plan would more than double it

Sources

Spot an error? We correct quickly and note it. Contact us via the contact page.

← More AI news