Physix Frontier · News Briefing Card (QbitAI · Sep 25, 2026)

T-Head launches Zhenwu V900 chip, expands SAIL open source

KEY FACTS

At the Yunqi Conference on September 22, Alibaba unveiled the Zhenwu V900 AI chip, claiming 3x the performance of the M890.
On September 23, T-Head announced further open-sourcing of its AI software stack T-Head SAIL, covering framework adaptation and acceleration libraries.
The Zhenwu chip already serves more than 650 customers across over 20 industries, and Ant Group, Xiaohongshu, and XPeng have adopted SAIL.
Xiaohongshu built a model migration and operator optimization agent based on SAIL's open-source code.
As of September 2026, ModelScope hosts 39 quantized models, with cumulative downloads exceeding 348,000.

KEY DATA

3x the M890Zhenwu V900 performance multiple

over 650Number of customers served

over 20Number of industries covered

over 348,000Quantized model downloads

PHYSIX OBSERVATION

Leading chip specs are only a ticket to entry; what customers really care about is migration cost. By open-sourcing its software stack, T-Head effectively lets developers optimize it themselves, which both eases the vendor's maintenance burden and hands ecosystem initiative to the community. This move is headed in the right direction, but whether it can build CUDA-like stickiness still depends on subsequent upstream merges and community activity.

Source: QbitAI report