bytevyte
bytevyte
Language
EN DE ES FR PT IT
ai-beats —

Alibaba Unveils Zhenwu M890 AI Processor and Multi-Year Silicon Roadmap

Zhenwu M890 AI processor

Alibaba has introduced the Zhenwu M890 AI processor, a high-performance chip designed to handle both large-scale model training and inference. The announcement, made during the 2026 Alibaba Cloud Summit, is a significant expansion of the company's internal silicon capabilities as access to international high-end hardware remains restricted. The new hardware delivers a threefold performance increase over the previous Zhenwu 810E generation.

The Zhenwu M890 AI processor is equipped with 144GB of High Bandwidth Memory (HBM) and supports an inter-chip interconnect bandwidth of 800GB/s. These specifications address the massive data throughput requirements of modern generative AI workloads. Alibaba confirmed that the Zhenwu series has already seen substantial adoption, with cumulative shipments reaching 560,000 units across its data center infrastructure.

Advanced Architecture and Supernode Integration

A central component of this release is the Panjiu AL128 supernode server. This system architecture allows Alibaba to cluster 128 Zhenwu M890 AI processor units into a single, unified compute entity. By treating the cluster as a standalone high-performance system, the company can scale processing power for the most demanding foundational models. The hardware natively supports a wide range of data precision formats, including FP32, FP16, and the increasingly critical FP4 format used for efficient inference.

The Zhenwu M890 AI processor is specifically optimized for AI agent tasks. These workloads require autonomous, multi-step execution, which places unique demands on memory latency and instruction processing. By tailoring the silicon for agentic workflows, Alibaba is positioning its cloud services to support the next wave of autonomous enterprise software.

A Multi-Year Roadmap for Silicon Independence

Alongside the current hardware release, Alibaba T-Head disclosed a clear multi-year roadmap for its semiconductor division. The company intends to maintain a rapid release cycle to keep pace with global competitors. Plans are already in place for the Zhenwu V900, which is scheduled for a third-quarter release in 2027. This will be followed by the Zhenwu J900 in the third quarter of 2028.

This long-term strategy highlights a shift toward self-reliance in the AI infrastructure market. By developing a consistent pipeline of proprietary silicon, Alibaba reduces its dependence on external vendors while optimizing its cloud stack from the chip level up to the software layer. The integration of the Zhenwu M890 AI processor into the broader Alibaba Cloud ecosystem provides a scalable path for domestic AI development.

While we strive for accuracy, bytevyte can make mistakes. Users are advised to verify all information independently. We accept no liability for errors or omissions.

AI-generated image.

✔Human Verified

Share