Alibaba Unveils Zhenwu M890 AI Chip to Counter Nvidia Restrictions

Alibaba Group unveiled the Zhenwu M890 artificial intelligence chip and the Qwen3.7-Max large language model at the Alibaba Cloud Summit in Hangzhou. Developed by its semiconductor subsidiary T-Head, the Zhenwu M890 offers three times the performance of the previous 810E generation and features 144 GB of GPU memory. The chip is positioned as a domestic alternative to Nvidia's H100 accelerators as U.S. export restrictions limit China's access to high-end hardware.

To support these processors, Alibaba Cloud introduced the Panjiu AL128 Supernode Server, which integrates 128 accelerators into a single rack, and updated its Bailian model service platform with a new Agentic RL reinforcement learning mechanism. The accompanying Qwen3.7-Max model is designed for complex reasoning and agent coding, with the capability to operate for up to 35 hours. Alibaba has already delivered 560,000 Zhenwu series chips to over 400 customers across 20 industries.

Further expanding its infrastructure, Alibaba is partnering with China Telecom to open a new data center in southern China powered by these in-house chips. T-Head has reportedly achieved scaled mass production of its GPUs and is planning a future IPO to fund continued development, with a roadmap including the V900 chip in 2027 and the J900 in 2028. CEO Eddie Wu projected that AI-related products will account for 50% of the cloud unit's external revenue within approximately one year.