Alibaba unveils Zhenwu V900 chip and plans trillion-parameter model

By Splitfeed EditorialLast updated Sep 26, 2026, 20:18 UTC

On September 22, 2026, Alibaba held an AI event at which it unveiled the Zhenwu V900, a new AI chip the company claims delivers triple the processing power of competing hardware, positioning it as a direct rival to Nvidia. At the same event, Alibaba disclosed plans for a next-generation Qwen model with a parameter count ranging from 5 trillion to 10 trillion. Separately, Tencent chose the same day to release Hy Image 3.5 Preview, a generative image model aimed at closing the gap with ByteDance and Alibaba in that space, and Tencent shares rose following the announcement.

The scale Alibaba is targeting with its planned Qwen model is significant. Chinese large language models announced before this point had not publicly reached parameter counts in the trillion range at this magnitude, making a 5 to 10 trillion parameter target a meaningful step up in ambition. The Zhenwu V900 chip matters equally because access to advanced AI chips has been a persistent constraint for Chinese technology companies following US export controls on Nvidia hardware. A domestically developed chip claiming to match or exceed Nvidia's performance, if the claim holds under independent scrutiny, would change the calculus around those constraints. The core dispute at this stage is whether Alibaba's performance claims for the V900 will be validated by third-party testing, and whether the planned Qwen model will reach deployment at the stated scale.

No reactions from the Pro-AI, Anti-AI or Middle Ground camps had been published at the time of this summary. Typically, the Pro-AI camp would treat a domestically produced chip challenging Nvidia and a model at this parameter scale as evidence that AI capability is advancing rapidly and spreading geographically. The Anti-AI camp would likely raise questions about the resource costs and safety considerations of building models at trillion-parameter scale, as well as skepticism about unverified performance claims. The Middle Ground camp would probably call for independent benchmarking of the V900 before drawing conclusions, and note that parameter count alone is a limited proxy for model quality or usefulness.

The most consequential near-term developments to watch are independent benchmark results for the Zhenwu V900 chip comparing it against current Nvidia hardware, and any further details Alibaba releases about a timeline and training infrastructure for the Qwen model at the announced scale.