Alibaba’s semiconductor arm has unveiled the Zhenwu V900, an AI processor the company calls China’s most powerful, alongside plans to train models several times larger than its current flagship.
Chief executive Eddie Wu said at the Apsara conference in Hangzhou that the chip delivers three times the performance of its predecessor, the M890, and can be clustered up to 500,000 units to train and run the largest models. Mass production and commercial release are set for the first quarter of 2027.
Models, semiconductors and the data centres that host them form the three legs of the group’s AI plan, Wu told attendees. By 2032, Alibaba Cloud intends to run more than 20 gigawatts of capacity worldwide.
Scaling the model side too
The Qwen team is training Qwen 4, with the company saying future versions will reach five to ten trillion parameters. Its current flagship, Qwen 3.8 Max, carries 2.4 trillion.
Wu said the team had made meaningful progress on models improving themselves, identifying weaknesses and generating training data with limited human involvement. Shares rose about 5% on the announcements.
The push reflects how hard US export curbs have pressed Chinese groups to build domestic alternatives to Nvidia’s processors, from silicon through to the data centres that house it.
