Alibaba Designs AI Chips Around Agents, Shifting the Race

Alibaba has introduced a dedicated AI processor for intelligent agents, accompanied by a multi-year silicon roadmap and a new large language model, demonstrating its strategy to build an integrated AI stack rather than merely responding to US export restrictions.

According to Reuters, Alibaba’s semiconductor subsidiary, T-Head, states that the Zhenwu M890 delivers three times the performance of its predecessor, the Zhenwu 810E. However, the architectural intent behind the chip is more significant than the raw performance increase: the M890 is engineered specifically for AI agents, which require systems to maintain long context windows, coordinate with other models in real time, and execute complex multi-step tasks with minimal human oversight.

These requirements, which place heavy demands on memory bandwidth and inter-model communication, differ substantially from the optimization goals of standard inference chips. This distinction reveals Alibaba’s vision for the future of AI compute. Rather than designing around current dominant use cases, the company is preparing for the workload profiles expected to define enterprise AI in the coming years.

Built for AI agents, not just inference

More important than the chip itself is the comprehensive roadmap Alibaba released alongside it. Following the M890, the V900 is scheduled for the third quarter of 2027, promising another approximate threefold performance increase, followed by the J900 in the third quarter of 2028. This deliberate, sustained cadence of in-house silicon upgrades mirrors the tick-tock product cycles Nvidia has used to maintain its lead in AI accelerators.

The parallel with Huawei is noteworthy. Last year, Huawei outlined a similar roadmap for its Ascend line, reflecting a shared reality: Chinese technology companies have determined that relying on foreign silicon, even if export restrictions were to ease, poses an unacceptable structural risk. Consequently, semiconductor development is being treated as a long-term capability-building exercise rather than a simple procurement issue.

Alibaba’s commitment to this initiative is substantial. Last year, the company pledged over 380 billion yuan (approximately US$53 billion) for cloud and AI infrastructure over three years, marking its largest-ever investment in the sector. The M890 and its successors are direct outcomes of this spending.

Traction that predates the announcement

T-Head reports shipping more than 560,000 Zhenwu units to date, with over 400 external customers across 20 industries, including automakers and financial services firms, deploying the chips. This represents a significant production footprint rather than lab hardware, providing Alibaba with extensive real-world deployment data ahead of the M890’s launch.

The new chip will be available to Chinese enterprise customers via Alibaba Cloud’s domestic model platform, Bailian, packaged within the Panjiu AL128 server system, which stacks 128 M890 accelerators into a single rack.

The software side of the stack

Alongside the hardware, Alibaba announced Qwen 3.7-Max, the latest version of its flagship large language model, designed for advanced coding and long-running agent tasks. The company states that the model can operate continuously for up to 35 hours without performance degradation, a specification that only makes sense when designing for extended autonomous operation.

This timing is intentional. Releasing a chip and a model optimized for the same workload class on the same day constitutes a platform strategy. Alibaba is creating a closed loop: its own silicon from T-Head, its own model in Qwen, and its own cloud delivery through Bailian. Each component reinforces the others, and the combined stack aims to reduce enterprise customers’ dependence on external vendors.

Half a million chips shipped. Successors arriving in 2027 and 2028. T-Head is not hedging its bets. At some point, building around US export controls transitions from a workaround to a core strategy. Alibaba appears to have crossed that threshold.

(Image source: The White House)

See Also: Alibaba Qwen is challenging proprietary AI model economics

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and co-located with other leading technology events. Click here for more information.

AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.