Alibaba Unveils Zhenwu V900 AI Chip, Targets 10‑Trillion‑Parameter Model and 20GW Data Centers by 2032

BABA unveiled the Zhenwu V900 AI accelerator at its Apsara Conference on September 22, calling it China's most powerful AI chip and saying it delivers three times the performance of its predecessor, the M890. Reuters reported that the chip supports clusters of up to 500,000 units and is slated for mass production in the first quarter of 2027. CEO Eddie Wu paired the chip with a plan to train a Qwen model of 5 to 10 trillion parameters, up from 2.4 trillion for the current flagship Qwen3.8-Max.

The infrastructure ambition is the other half of the announcement: Alibaba Cloud is targeting more than 20 GW of global data-center capacity by 2032. Reuters framed the push as a response to tightening US export curbs, which are driving Chinese firms to build domestic alternatives to NVDA processors. CryptoBriefing, citing the conference, put cumulative Zhenwu shipments at 560,000 units across more than 400 customers and Alibaba's AI investment commitment at $53 billion over three years.

Hong Kong-listed Alibaba shares rose 5.1% to their highest level in a month, according to Reuters, while other outlets reported gains nearer 3% depending on listing and timing. The AP noted the announcement lands ahead of Xi Jinping's upcoming Washington visit, where AI competition is expected on the agenda.

What to watch: whether the V900 hits its Q1 2027 mass-production date, which depends on access to advanced manufacturing under US restrictions, and how quickly the 20 GW buildout translates into cloud revenue relative to its capital cost. No independent benchmarks of the V900 have been published yet, so the three-fold performance figure is Alibaba's own claim.