Alibaba launches Zhenwu V900 AI chip and targets 5‑10 trillion‑parameter models, sparking stock surge

Alibaba unveiled its Zhenwu V900 AI chip alongside a data-center buildout plan of more than 20 gigawatts of global capacity by 2032 and a next-generation model targeting 5 to 10 trillion parameters. The company has committed roughly $53 billion over three years to the expansion. Mass production of the chip is not scheduled until the first quarter of 2027.

One specification claim needs care, because the headlines are looser than the disclosures. The V900's "triple processing power" is three times Alibaba's own predecessor, the Zhenwu M890 released earlier this year, not three times anything Nvidia ships. No international benchmark for the V900 exists, and Chinese analysts have so far only compared the older M890 against Nvidia's H200. The chip does carry substantive specifications: 216 GB of HBM, 1.2 TB/s of inter-chip bandwidth, and clustering up to 500,000 cards. The parameter target would put the next model roughly two to four times the size of the current Qwen 3.8 Max at 2.4 trillion.

The share reaction was front-loaded and then faded, which matters more than any single percentage. BABA rose as much as 5.1% in Hong Kong trading and opened around 4% higher in New York near $120.36, but held a gain of only about 0.5% at roughly $116.31 later in the US session. Baidu was essentially unmoved, which points to an Alibaba-specific repricing rather than a sector-wide move. Our own data shows mentions running 4.7 standard deviations above baseline, so attention is far outpacing the price follow-through.

The gap between announcement and shipment is the honest frame here. Every performance figure is vendor-reported and unbenchmarked, and nothing reaches customers before early 2027, so today's move prices a specification sheet rather than delivered silicon. What would convert it into something firmer is independent benchmark data, confirmed production timelines, and named early customer deployments.