Alibaba's T-Head to Tape Out Second-Generation AI Chip in H2
Alibaba CEO Wu Yongming announced that T-Head's second-generation domestic AI chip, designed for large-scale model training, is scheduled to tape out in the second half of the year. Additionally, Alibaba Cloud has launched supernode instances powered by its new Zhenwu M890 chip to meet growing customer demand. The development highlights Alibaba's push to build a self-reliant AI hardware ecosystem amid tightening US export controls on advanced semiconductors. By offering domestic alternatives like the Zhenwu series, Alibaba aims to reduce reliance on foreign GPU suppliers like Nvidia for large-scale AI workloads. The upcoming second-generation chip is expected to feature high computing power and interconnect bandwidth suitable for LLM training. Meanwhile, the currently deployed Zhenwu M890 chip features 144GB of memory and 800 GB/s interchip bandwidth, serving over 650 customers as of August.
## BACKGROUND
Founded in 2018, T-Head Semiconductor is Alibaba's chip-design subsidiary focusing on processors for data centers, AI, and IoT. As China seeks domestic alternatives to Nvidia's dominant AI hardware, tech giants like Alibaba are integrating their cloud and chip divisions to optimize full-stack AI capabilities.