Alibaba Group is planning to train a new artificial intelligence model with 5 trillion to 10 trillion parameters, according to CEO Eddie Wu. The company also unveiled its Zhenwu V900 AI chip, which it claims is China's most powerful, and aims to expand its global data center capacity to over 20 GW by 2032.
Alibaba's significant investment in AI models, custom chips, and data center infrastructure signals its ambition to compete at the forefront of the generative AI race, potentially impacting cloud computing services, enterprise AI adoption, and the semiconductor industry, particularly within China.
Alibaba Group is making an aggressive push into artificial intelligence, announcing plans to develop a new AI model with a massive 5 trillion to 10 trillion parameters. The company also unveiled its latest homegrown AI chip, the Zhenwu V900, which it claims is the most powerful in China and offers three times the performance of its predecessor. These announcements were made at the Alibaba Cloud Summit in Hangzhou, signaling a broad strategy encompassing AI models, custom silicon, and data center infrastructure.
The new AI model, Qwen 3.7-Max, is designed for agentic workloads, capable of sustained, multi-step operations and complex engineering tasks without human intervention. In an internal benchmark, it ran for 35 consecutive hours on a Zhenwu M890 chip, executing over 1,000 tool calls and outperforming the chip manufacturer's official version by tenfold. The model is optimized for various agent harnesses and can handle multi-file software projects and orchestrate multi-agent office workflows.
Alibaba Cloud also introduced the Panjiu AL128 Supernode Server, a rack-scale system with 128 AI accelerators designed for the unpredictable inference request patterns generated by AI agents. This server is available through Alibaba's model service platform, Bailian, which now features Agentic RL for continuous model refinement and built-in safety guardrails. T-Head, Alibaba's chip design subsidiary, formally debuted the Zhenwu M890, its most powerful AI processor to date, featuring 144 GB of on-chip memory and supporting precision formats down to FP4 for both high-accuracy training and low-cost inference. The company also launched the ICN Switch 1.0 networking chip to enable congestion-free communication across multiple accelerators.
Alibaba aims to significantly expand its global data center capacity, targeting over 20 GW by 2032.
Pick the topics you care about. Get only what matters, on your cadence.