Alibaba — Zhenwu V900, 20GW DC target, 5–10T Qwen path

Opening

At Apsara in Hangzhou, Alibaba unveiled the Zhenwu V900 AI chip, set a >20 GW global Cloud data-center capacity target by 2032, and outlined a Qwen path to 5–10 trillion parameters, per Reuters. Reuters reports CEO Eddie Wu said the T-Head Zhenwu V900 delivers three times the performance of the prior M890 and can cluster up to 500,000 chips, with mass production and commercial release in Q1 2027. Current flagship Qwen 3.8 Max is cited at 2.4 trillion parameters; Qwen 4 is in training, with future Qwen 4.5/5 expected to scale toward 5–10 trillion. Wu said industry mid-to-long-term demand "far outpaces" supply and Cloud will bring AI supernodes online at commercial scale this quarter. CNBC adds existing Zhenwu chips are used by more than 650 customers. Hong Kong-listed shares rose about 5.1% on the day per Reuters (CNBC cited around 3%).

The numbers

Zhenwu V900 — 3× M890 performance; up to 500k-chip clusters; commercial Q1 2027 — Reuters / company. Cloud DC capacity — >20 GW global by 2032 — Reuters / company. Qwen path — 5T–10T parameters (vs 2.4T on Qwen 3.8 Max) — Reuters / company. Installed Zhenwu customers — >650 — CNBC / company.

Why it matters

Principal roadmap for China's largest cloud: own silicon cadence, multi-year GW target, and model scale claims under U.S. GPU export pressure. Company framing; shipment and utilization prints still to come.

Positioning

AI capex durability — Domestic hyperscaler still guiding capacity up; supply-chain constraint language matches the compute-bound side of the bet. HBM supply binds — Not a memory print; still relevant only if V900/HBM attach shows up in trade or supplier guides later.