Alibaba Cloud deploys Qwen3.8 on Zhenwu chip: silicon substitution becomes operational

A state-of-the-art Qwen model served on domestic silicon - the message is more strategic than technical.

The fact

Alibaba Cloud announced the deployment of its Qwen3.8 model on the Zhenwu chip (Tech in Asia, July 24, 2026). The announcement follows a series of Chinese moves - CXMT IPO, Huawei/CloudRobo, Ascend expansion - that are building, brick by brick, a compute stack distinct from the Nvidia line.

Our analysis

The important thing is not the raw performance - third-party benchmarks are lacking at this stage. The important thing is the gesture: a Chinese hyperscaler serving its flagship model on domestic silicon publicly validates that substitution has moved from "test" to "prod". Each announcement of this type narrows the gap of dependence and makes any further tightening of export controls more costly for Washington: the levers blunt as substitution progresses.

To watch

• The publication of third-party benchmarks (latency, cost per token, throughput) on Zhenwu and, more broadly, performance comparisons with Nvidia alternatives available in China.
• Any mention of opening Zhenwu to external customers (Baidu, ByteDance, government).
• The next announcements from Ascend/Huawei on Qwen or DeepSeek inference.