Alibaba opened its 2.4-trillion-parameter model on Wednesday. Builds for eight Chinese chips were already waiting.
When Alibaba finished publishing the weights of Qwen3.8-2.4T-A95B on Hugging Face, the FlagRelease account run by Beijing's state-backed AI academy posted ready-to-run versions for nine different accelerators — Huawei Ascend, Moore Threads, Baidu's Kunlunxin, Hygon, MetaX, Enflame, Tsingmicro and one more domestic vendor, alongside Nvidia — in INT8 and FP8. The upload history shows the work started before the launch: the weight files for the Nvidia build went up on 11 August, a day before Alibaba's own repository was finished. The academy says it has done same-day chip ports for twelve open models since February.
Why it mattersThe hard part of using Chinese accelerators has never been the silicon; it is that each new model needs weeks of porting before it runs on anything but Nvidia. A shared open-source stack that ships day-zero builds for eight domestic chips at once removes that lag, and it quietly turns every new Chinese open model into an argument for buying domestic hardware. It also shows how closely the model labs and the chip effort are now coordinated: someone had the weights before the public did.
Read in the app — free, in 9 languages
Related stories
Anthropic expects to raise as much as SpaceX did — the largest listing in history — and could file publicly this month2026-08-21Nvidia is paying $6 billion for the machine that builds a rival's models — and hiring 109 of the people who ran it
2026-08-21The doctors and lawyers who grade AI answers are now a half-billion-dollar business — and this one grew fivefold in eight months
2026-08-21Anthropic still leads on business AI spending. For the first time since May, the gap is closing rather than widening.
2026-08-21Alibaba's AI business grew at its fastest pace in five and a half years. Its profit fell about 75 percent.
2026-08-20