Alibaba unveils Zhenwu M890 chip

Alibaba Group on May 20 unveiled the Zhenwu M890, a new AI chip developed by its semiconductor arm T-Head, as the Chinese company widened its in-house hardware push amid tighter U.S. export curbs on advanced Nvidia products. Reuters reported the chip is positioned as part of a domestic alternative stack for AI computing in China. Alibaba Cloud said the M890 was launched alongside a rebuilt cloud stack and a new large language model at a company event on Wednesday. Bloomberg reported the chip has 144GB of memory and is designed for both AI training and inference workloads. ### What exactly did Alibaba launch? (money.usnews.com) T-Head introduced the Zhenwu M890 as the successor to the Zhenwu 810E, with Alibaba and Reuters saying the new chip delivers three times the performance of the earlier model. CNBC also reported the announcement from Chongqing, where Alibaba presented broader updates to its AI offerings. (alibabacloud.com) Alibaba Cloud paired the chip with a server platform aimed at larger-scale AI workloads. Secondary reports citing the company said Alibaba also outlined a system architecture built around the M890 for model training and concurrent inference. ### Why is Alibaba building this now? (money.usnews.com) Reuters said Alibaba is intensifying efforts to build domestic alternatives to Nvidia processors as U.S. export restrictions tighten. That places the M890 inside a broader Chinese effort to localize AI compute infrastructure rather than rely on imported accelerators that face licensing and shipment limits. (business-standard.com) Business Standard, citing Reuters, said Chinese firms are ramping up development of domestic AI computing infrastructure. Alibaba’s launch adds its cloud business, model business and chip unit to that effort in a single product cycle. ### What do the published specs tell us? (money.usnews.com) Bloomberg reported the M890 is equipped with 144GB of GPU memory. Reuters and Alibaba said the chip is meant to handle both training and inferencing, while Alibaba described it as suited to the “agentic” wave of AI systems that can execute multi-step tasks. (business-standard.com) CNBC reported Alibaba said the chip is three times more powerful than its predecessor. That claim has not been independently benchmarked in the reporting reviewed here, and Alibaba did not, in the cited coverage, publish a direct comparison against a current Nvidia product. ### How does this fit Alibaba’s larger AI strategy? (bloomberg.com) Alibaba Cloud said the chip launch came with a new flagship model, Qwen3.7-Max, and a rebuilt cloud platform designed for autonomous AI agents. The company said its model and application services platform is expected to surpass RMB 10 billion in annualized recurring revenue in the June quarter and reach RMB 30 billion by year-end. (cnbc.com) China Daily reported Alibaba said the Zhenwu series has shipped more than 560,000 units cumulatively and serves more than 400 customers across more than 20 industries, including telecom, automotive and finance. Those figures, if sustained, would give Alibaba an installed base for deploying the new generation. ### What should readers watch next? (alibabacloud.com) Alibaba Cloud’s next test will be commercial deployment. The company has already tied the M890 to its cloud and model offerings, and future disclosures on customer adoption, server availability and follow-on chips will show how quickly the platform moves beyond launch claims. Reuters-linked coverage and Alibaba’s own cloud updates are the main places where those next details are likely to appear. (chinadaily.com.cn) (money.usnews.com)