What happened
On Alibaba’s earnings call for the second quarter of its fiscal year 2027, held on 19 November 2026, CEO Eddie Wu (吴泳铭) said that Qwen (千问) is already training its next model on Alibaba’s own T-Head (平头哥) chips. That is according to the Chinese-language outlet CLS (Cailian Press), which we read on 21 November 2026. The translation and summary are ours. We did not read an official transcript or recording of the call, so this is the press account of what Wu said. Our notes say TechNode repeated the information on 20 November; we do not link that report.
Key details
- Training: per CLS, Qwen is training its next model on T-Head chips. The article does not name the model or say how large it is.
- Inference: per CLS, Alibaba’s Bailian (百炼) platform already offers inference on Alibaba’s in-house chips.
- Claimed advantage: Wu said a full-stack AI infrastructure gives Alibaba an edge on a domestic supply chain, and that closer coordination between chips and models lowers the cost per unit of compute. These are the CEO’s claims, not independent findings.
- Not given in the report we read: the model name or size, the type and number of chips, the release date, and any cost figure.
Why it matters
Alibaba training a frontier-class model family on its own chips would reduce its reliance on outside suppliers, which is relevant to anyone following where Chinese labs run their training. This statement does not show that yet: there are no specifications, no results and no source other than the press account of an earnings call.
We will update this article if Alibaba publishes the call transcript, names the model, or releases it with details of the hardware used.
Disclosure: Claude, made by Anthropic, was one of the AI tools used to research and draft this article. Anthropic competes with Alibaba’s Qwen. The statements are Alibaba’s, as reported by CLS.