Alibaba unveils Zhenwu V900 chip and plans model of up to ten trillion parameters

At its Apsara Conference in Hangzhou on 22 September, Alibaba introduced the Zhenwu V900, its next-generation AI accelerator. Alibaba calls the V900 the most powerful AI chip in China and says it delivers three times the performance of its previous Zhenwu M890. No independent benchmark was published in the material reviewed for this edition.

Alibaba says a single cluster based on the V900 can support up to 500,000 cards for training and inference. Its T-Head unit also makes Yitian CPUs, Panmai network-interface chips and interconnect silicon. Alibaba says it is co-optimising chips, servers, supernodes, networks, models and inference software, and that it expects annual AI-chip shipments to grow substantially.

On models, Alibaba says the Qwen team plans to train a model with five trillion to ten trillion parameters for more complex, longer tasks. Parameter count describes scale; it is not a direct measure of quality or intelligence. Alibaba also said it expects more than 20 gigawatts of computing capacity by 2032. That is a plan, not capacity already operating.

The announcement should be read as Alibaba's own product and strategy claims. The checked sources do not establish that the V900 has beaten another chip in independent testing, that it is commercially available outside Alibaba's cloud, or that it is manufactured without foreign technology. Alibaba's language about artificial superintelligence and recursive self-improvement is the company's ambition, not an established fact.