Alibaba Unveils Zhenwu V900 Accelerator and Announces Multi-Trillion Parameter Model

Chinese group Alibaba announced the new Zhenwu V900 AI accelerator, in a step reflecting the group's ambition to build a national chip ecosystem capable of competing amidst escalating US restrictions on semiconductor exports.

In addition to the new accelerator, the group announced an upcoming large language model within the Qwen series, which is reportedly expected to contain between five and ten trillion parameters—a size many times larger than the largest models announced to date.

These announcements signal an accelerating path toward vertical integration among major Chinese tech companies, as they seek to possess full capabilities across the AI chain from chip to model, away from reliance on US suppliers.

This announcement intensifies the tech competition between US and Chinese companies in the field of generative artificial intelligence, with rising pressure on companies like Nvidia, Intel, and AMD to speed up the development of a new generation of computing accelerators.

What do these terms mean?

Parameter: The basic number that the model learns from data and uses to make decisions. The higher the number of parameters, the greater the model's ability to understand context and handle complex tasks—think of it as the number of "brain cells" in AI.