Alibaba unveiled the Zhenwu V900 AI accelerator on Tuesday, a chip it says delivers three times the performance of its predecessor, the Zhenwu M890, and can be linked in clusters of up to 500,000 units to train frontier models. Chief executive Eddie Wu presented it at the company’s annual Apsara Conference in Hangzhou, alongside a target to expand Alibaba’s data centre capacity to more than 20 gigawatts by 2032. Mass production is due to begin in the first quarter of 2027.
The chip comes from T-Head, Alibaba’s in-house chip design unit, and extends a family that the company says is already used by more than 650 customers across several industries. Wu also said the Qwen team is working towards a model with up to 10 trillion parameters. Its Hong Kong shares rose 2 percent on the day.
The announcement is one of China’s clearest answers yet to losing reliable access to Nvidia. US export controls have cut Chinese buyers off from Nvidia’s most capable parts, pushing the country’s cloud providers to design their own. A 500,000-chip cluster is a claim about networking and software as much as silicon, since stitching together very large numbers of accelerators is how Chinese labs aim to make up for individual chips that are widely seen as trailing the best American ones.
Alibaba has not said which foundry will make the V900, and that is the question that decides whether 2027 volumes are realistic. US rules sharply restrict TSMC from producing advanced AI chips for Chinese designers, and SMIC’s leading-edge capacity is already contested by Huawei. A threefold gain over Alibaba’s own previous chip also says nothing about how the V900 compares with Nvidia’s current parts, the comparison Alibaba chose not to publish.
