Alibaba Builds Agent-Focused AI Chip, Outlines Roadmap

Alibaba introduced the Zhenwu M890 accelerator, a new large language model called Qwen 3.7-Max, and a multi-year chip timetable that lists the V900 in the third quarter of 2027 and the J900 in the third quarter of 2028. The company presented the hardware and software as parts of an integrated AI stack spanning silicon to cloud.

The M890 was developed by Alibaba’s semiconductor arm, T-Head. Alibaba said the processor delivers roughly three times the performance of the prior Zhenwu 810E. The company described the chip as optimized for AI agents-software that must retain long stretches of context, coordinate with other models in real time and complete multi-step tasks with limited human input. Those agent workloads place heavier demands on memory bandwidth and inter-model communication than standard inference tasks.

The roadmap schedules the V900 for Q3 2027 and the J900 for Q3 2028, with each generation expected to increase performance over its predecessor. T-Head reported it has already shipped more than 560,000 Zhenwu units to over 400 external customers across about 20 industries, including automakers and financial services firms.

Alibaba plans to make the M890 available to domestic enterprise customers through its cloud model platform, Bailian. The company will offer a Panjiu AL128 server that houses 128 M890 accelerators in a single rack for large-scale deployments.

On the software side, Alibaba described Qwen 3.7-Max as targeted at advanced coding tasks and extended autonomous operation. The company reported the model can run continuously for up to 35 hours without performance degradation, a specification tied to prolonged agent workloads rather than short, single-turn inference.

Alibaba announced the chips and model alongside a capital commitment of more than 380 billion yuan (about $53 billion) for cloud and AI infrastructure over three years. The company presented the hardware, models and delivery platform as linked elements of its internal technology stack.

Chinese technology companies have increasingly planned multi-year chip and model programs. Alibaba’s announcements set precise dates for successor accelerators and add to a pattern of domestic firms defining multi-year development schedules for accelerators, models and cloud services.