Alibaba Launches Zhenwu M890 AI Chip Amid U.S. Export Challenges

In a notable development for the Chinese technology sector, Alibaba has unveiled its latest AI chip, the Zhenwu M890, which reportedly delivers three times the performance of its predecessor, the Zhenwu 810E. This launch arrives amid U.S. export restrictions that have limited access to advanced chips, prompting companies like Alibaba to enhance their domestic semiconductor capabilities.

The Zhenwu M890 features 144GB of GPU memory and an interchip bandwidth of 800GB per second. Developed by Alibaba's T-Head subsidiary, this chip is designed for both training and inferencing workloads, with a particular emphasis on agentic AI—software that can autonomously execute complex, multi-step tasks. Myron Xie from SemiAnalysis noted the increasing popularity of the Zhenwu platform among enterprise clients in China, although he pointed out the absence of disclosed compute performance metrics.

New Offerings and Future Plans

In addition to the M890, Alibaba introduced the Panjiu AL128 server system, which can house 128 of these accelerators in a single rack, as well as the Qwen 3.7-Max, a large language model aimed at coding and extended workflows. The Qwen 3.7-Max is designed to maintain consistent output over lengthy sessions, reportedly lasting up to 35 hours.

The Zhenwu product line has gained significant traction, with Alibaba reporting cumulative shipments exceeding 560,000 units across more than 400 enterprise clients in 20 sectors. Looking ahead, Alibaba plans to release the V900 chip, expected to debut in mid-2027 with performance projections to triple that of the M890, followed by the J900 anticipated for late 2028.

Addressing Market Gaps

Alibaba is also seeking to raise funds by planning to list T-Head, aiming to meet the demand for Chinese alternatives to Nvidia's products. Brady Wang from Counterpoint Research acknowledged that while the M890 may not match Nvidia’s H200 in raw processing power, it could serve as a practical substitute within the domestic market.

This strategic focus on domestic chip production aligns with broader efforts by Chinese firms to create self-sufficient AI infrastructure in response to U.S. restrictions. Earlier this year, Alibaba launched a data center in Shaoguan, Guangdong, equipped with 10,000 Zhenwu chips, with plans to scale this to 100,000 chips.

Financial Commitment to AI Infrastructure

To bolster its AI and cloud capabilities, Alibaba has pledged over 380 billion yuan ($53 billion) to infrastructure development over the next three years. Chief Executive Eddie Wu has stressed the importance of comprehensive end-to-end AI solutions, which encompass proprietary silicon, operational infrastructure, foundational models, and cloud services.

As competition in the AI market heats up, Alibaba's advancements with the Zhenwu M890 represent a critical step in strengthening its position and expanding its capabilities in an environment where access to foreign technology is increasingly limited. The implications of this launch extend beyond performance metrics; they reflect a broader geopolitical context in which technological self-sufficiency is becoming a priority for countries like China.