Thought Leadership
Alibaba Group has unveiled its latest AI processor, the Zhenwu M890, designed to power autonomous AI agents and reduce reliance on imported high-end GPUs from Nvidia.
The announcement was made at the company’s annual Alibaba Cloud Summit, where Alibaba showcased a broader push into in-house semiconductor development amid tightening US export restrictions on advanced chip technologies. The chip was developed by Alibaba’s semiconductor arm T-Head and marks another step in China’s wider strategy to build a self-sufficient AI hardware ecosystem.
Zhenwu M890 Targets the Next Generation of AI Agents
According to Alibaba, the Zhenwu M890 delivers up to three times the computing performance of its predecessor, the Zhenwu 810E. The chip is specifically optimized for AI agent workloads—systems designed to independently execute complex, multi-step tasks with minimal human intervention.
Unlike traditional large language model inference, AI agents require sustained memory access, long-context processing, and real-time coordination between multiple models. Alibaba says the M890 is engineered to handle these demands efficiently, reducing latency during model synchronization in live environments.
Long-Term Silicon Roadmap and Strategic Multi-Billion-Dollar Investments
Alongside the presentation of the current model, Alibaba Group outlined a binding multi-year roadmap for its in-house silicon development. The plan stipulates that a successor to the M890, the V900, will follow in the third quarter of 2027. The company forecasts another tripling of performance compared to the M890 for this chip. A further processor, the J900, is scheduled for introduction in the third quarter of 2028. With this tightly structured cadence, the group signals a continuous technological upgrade cycle for its hardware division.
These measures are part of a broader national strategy in the People’s Republic of China to increase independence from foreign supply chains in strategic key technologies. Similar steps were already taken last year by technology group Huawei.
To support this objective at the infrastructure level, the Hangzhou-based company announced an investment program last year: over a period of three years, Alibaba will invest more than 380 billion yuan (approximately US$53 billion) in the expansion of its cloud and AI infrastructure. This represents the largest financial commitment the company has ever made in this sector.
The investments are based on the assumption that demand for dedicated AI compute capacity will continue to increase significantly, driven by the widespread adoption of agent-based enterprise applications.
Panjiu AL128: High-Density AI Server System
For deployment in data centers, Alibaba Group simultaneously introduced the Panjiu AL128 server system. The system integrates a total of 128 of the new Zhenwu M890 accelerators within a single server rack, enabling high compute density. Chinese enterprise customers can access these hardware resources immediately via Alibaba’s domestic model platform “Bailian” as part of Alibaba Cloud.
Alibaba’s semiconductor subsidiary T-Head also presented updated shipment figures for the Zhenwu processor family during the event. According to the company, more than 560,000 units of Zhenwu processors have been delivered to date. The current customer base comprises over 400 external companies across 20 different industry sectors. These include primarily automotive companies and financial services providers, which use the chips for internal data analytics and automation processes.
Qwen 3.7-Max Extends the Software Stack
On the software side, Alibaba introduced Qwen 3.7-Max, the latest version of its flagship large language model. The model is optimized for advanced coding tasks and long-running AI agent workflows. Alibaba claims it can operate continuously for up to 35 hours without performance degradation, a capability aimed at complex industrial automation scenarios.
