Alibaba used its Apsara Conference in Hangzhou on September 22 to unveil the Zhenwu V900, billed as China's most powerful AI accelerator — two days before Xi Jinping and Donald Trump are due in Washington, where export controls are expected on the table.
The chip, from Alibaba's T-Head semiconductor unit, is said to deliver three times the performance of the Zhenwu M890 launched in May 2026 and to handle both high-precision training and inference. The company said devices can be clustered in groups of up to 500,000 for frontier-scale models, with mass production and commercial release aimed at the first quarter of 2027. Independent comparisons with Nvidia's H200 or Blackwell parts were not published at launch.
Alibaba also sketched a new Qwen model with 5 trillion to 10 trillion parameters, up from 2.4 trillion on Qwen 3.8 Max, and a target of more than 20 gigawatts of global data-center capacity by 2032. Shares rose 3 to 5 percent on the announcements.
The timing is the product. U.S. controls still throttle access to Nvidia silicon, so a domestic accelerator unveiled on the eve of a leaders' meeting is a talking point as much as a shipping SKU. Diplomats and buyers will not know whether Zhenwu closes the gap until wafers leave the line next year. Until then, the chip's job is to show that China's AI buildout has a Plan B — and that Plan B can be clustered by the hundred thousand.
Image source: i.ibb.co