Alibaba’s newest AI accelerator underscores growing momentum among Chinese tech firms to reduce reliance on foreign processors.
Alibaba Group has unveiled a new artificial intelligence processor, the Zhenwu M890, marking a major step in its effort to reduce reliance on foreign semiconductor technologies amid tightening export restrictions from the United States. The launch positions Alibaba more directly against global chip leaders such as Nvidia while supporting China’s broader push for self-reliant AI infrastructure.
Developed by Alibaba’s chip design subsidiary T-Head, the Zhenwu M890 delivers nearly three times the performance of the earlier Zhenwu 810E processor. The chip is purpose-built for AI “agent” workloads — advanced software systems capable of executing complex, multi-stage tasks autonomously with limited human supervision. Alibaba said the processor is optimized for heavy memory usage and high-speed communication, enabling AI models to maintain long contextual understanding and collaborate efficiently in real time.
The company also outlined a forward-looking silicon roadmap. A successor chip, the V900, is scheduled for release in the third quarter of 2027, followed by the J900 in 2028. Each generation is expected to deliver significant performance improvements, demonstrating Alibaba’s commitment to sustained in-house chip innovation.
The announcement reflects growing momentum across China’s technology sector to develop domestic alternatives to advanced processors restricted under U.S. trade controls, echoing similar initiatives led by Huawei.
Alibaba introduced the chip at its annual Alibaba Cloud Summit alongside the Panjiu AL128, a server system integrating 128 AI accelerators in a single rack. The platform is now available to enterprise customers in China through Alibaba Cloud’s Bailian AI service platform.
T-Head reported shipments of more than 560,000 Zhenwu chips so far, with adoption across over 400 customers in industries including automotive and financial services.
Alibaba also launched Qwen 3.7-Max, the latest version of its large language model designed for advanced coding and long-running AI agent operations, capable of operating continuously for up to 35 hours without performance degradation.
