Alibaba unveils Zhenwu V900, calling it China's most powerful AI chip

Alibaba unveiled its Zhenwu V900 AI chip at the company's Apsara conference on Monday, with CEO Eddie Wu presenting it as China's most powerful AI accelerator to date. Built by Alibaba's T-Head semiconductor unit, the V900 delivers roughly three times the performance of its predecessor, the Zhenwu M890, which launched in May.

The specifications reflect a chip designed for both training large models and serving inference at scale: 216GB of memory, up from 144GB on the M890, and 1,200GB/s of inter-chip bandwidth, up from 800GB/s. The V900 supports FP8 and FP4 precision formats and can scale to clusters of up to 500,000 chips. Alibaba says mass production and commercial availability are targeted for the first quarter of 2027.

The timing and framing matter as much as the specs. US export controls have restricted Chinese companies' access to Nvidia's most advanced AI accelerators, and the Zhenwu line is Alibaba's direct answer to that gap — a domestically designed alternative aimed at keeping Chinese AI labs and cloud customers supplied with training and inference hardware without depending on hardware Washington won't sell them. Alibaba disclosed it has already shipped 560,000 units across the Zhenwu chip family to more than 400 external customers, though the company's own figures elsewhere put that customer count above 650.

Alibaba paired the chip announcement with considerably larger ambitions: models scaling up to 10 trillion parameters, and a commitment to build out 20 gigawatts of data center capacity by 2032. Both targets position Alibaba Cloud to compete directly with US hyperscalers on raw infrastructure scale, not just model quality — a bet that domestic chip supply, rather than access to foreign silicon, will determine which Chinese cloud providers can actually deliver on those commitments.

Investors read the announcement as credible: Alibaba shares rose following the Apsara keynote, reflecting confidence that the company can execute on both the chip roadmap and the data center buildout it announced alongside it.