Alibaba (BABA) Stock: New Zhenwu Chip Targets Nvidia in AI Push

TLDR

• Alibaba unveiled the Zhenwu M890 AI chip, three times more powerful than its predecessor, built for agentic AI workloads.
• The chip was developed by Alibaba’s T-Head unit and is immediately available to Chinese enterprise customers via Alibaba Cloud.
• A multi-year chip roadmap includes the V900 (Q3 2027) and J900 (Q3 2028), each promising another ~3x performance leap.
• Alibaba also released Qwen 3.7-Max, an upgraded LLM built for advanced coding and long-running agent tasks up to 35 hours.
• Over 560,000 Zhenwu units have been shipped to date across 400+ customers in 20 industries.

---

Alibaba launched a new in-house AI chip and updated its flagship large language model on Wednesday, doubling down on its push to build a domestic AI ecosystem independent of Nvidia.

Alibaba Cloud announced two new T-Head chips at its summit: Zhenwu M890, a unified training and inference AI processor, and ICNSwitch, a dedicated interconnect chip. A standalone switch chip signals cluster-scale ambition. pic.twitter.com/t8rnRpkxb5

— Poe Zhao (@poezhao0605) May 20, 2026

The new chip, the Zhenwu M890, delivers three times the performance of its predecessor, the Zhenwu 810E. It was unveiled at the Alibaba Cloud Summit in Hangzhou and is available now to Chinese enterprise customers through Alibaba Cloud’s Bailian platform.

Alibaba stock (BABA) was down 0.60% at the time of the announcements.

Alibaba Group Holding Limited, BABA

The M890 was built by Alibaba’s semiconductor design unit, T-Head. It carries 144GB of GPU memory, up from 96GB in the previous model, allowing it to handle much larger data loads.

Unlike the last generation, which focused on model inference alone, the M890 supports both training and inference. That makes it a more complete chip for companies building and running AI applications.

The chip is purpose-built for agentic AI — systems that carry out complex, multi-step tasks with limited human input. These workloads require models to retain long stretches of context and coordinate with each other in real time.

T-Head said it has shipped more than 560,000 Zhenwu units to date, with over 400 external customers across 20 industries including automotive and financial services.

A Roadmap That Keeps Going

Alibaba didn’t stop at one announcement. The company outlined a multi-year chip roadmap that includes the V900, expected in Q3 2027, followed by the J900 in Q3 2028.

The V900 is projected to deliver another roughly 3x performance gain over the M890. That kind of sustained cadence is a signal that Alibaba is treating chip development as a long-term infrastructure play, not a one-off project.

Alongside the M890, Alibaba unveiled the Panjiu AL128, a server system that packages 128 of the new accelerators into a single rack.

Qwen Gets an Upgrade Too

Alibaba also released Qwen 3.7-Max, the latest version of its flagship LLM. The model is engineered for advanced coding and long-running agent tasks, and can operate continuously for up to 35 hours without performance degradation.

The Qwen update and chip launch were announced together at the Hangzhou summit, positioning both as part of the same broader push into agentic AI infrastructure.

The move comes as Washington continues to restrict the sale of Nvidia’s most advanced chips to Chinese buyers. Alibaba was historically a major Nvidia customer, and the export controls have accelerated the shift to domestic alternatives.

Other Chinese chip companies including Biren Technology and MetaX have recently listed on mainland China and Hong Kong exchanges, pointing to a broader buildup in the sector.

Alibaba previously pledged over 380 billion yuan ($53 billion) in cloud and AI infrastructure spending over three years — its largest-ever commitment to the space.

On a recent earnings call, CEO Eddie Wu said AI-related product revenue is expected to become the primary driver of growth in the cloud segment. The company has framed its in-house chips as offering “the highest value for money compute power,” with the potential to improve margins.