Alibaba Group plans to train a new artificial intelligence model featuring between 5 trillion and 10 trillion parameters, Chief Executive Eddie Wu announced on Tuesday, September 22, 2026. Speaking at the Apsara Conference in Hangzhou, Wu outlined a strategy that integrates new AI models, proprietary semiconductors, and expanded data center capacity. Upcoming series, including the Qwen 4.5 and Qwen 5, are projected to reach this 5-to-10 trillion parameter scale, which serves as a rough measure of a model's size and complexity.
This development follows the release of Alibaba's current flagship model, Qwen 3.8 Max, which contains 2.4 trillion parameters. The planned models would be approximately two to four times larger than the current version. Wu stated that Alibaba's proprietary M890 AI supernode already facilitates inference for models exceeding 2 trillion parameters, a capability he said is held by only a few companies globally.
Alongside the new models, Alibaba introduced the Zhenwu V900 AI chip, developed by its T-Head semiconductor unit. Wu described the V900 as the most powerful AI chip in China, reporting that it delivers three times the performance of its predecessor, the M890. A single cluster using the V900 can support up to 500,000 cards for training and inference. The company scheduled mass production and commercial release of the chip for the first quarter of 2027.
The scale of Alibaba’s infrastructure expansion affects enterprise clients and cloud service users who rely on high-performance computing for AI applications. By targeting a data center capacity of 20 gigawatts by 2032, Alibaba is positioning itself to support a rise in "Machine Intelligence." For businesses using Alibaba Cloud, this could eventually manifest as improved processing for complex, long-horizon tasks, though Wu noted that global supply chain shortages currently limit the speed of this expansion.
The launch of the Zhenwu V900 chip comes as Chinese technology firms navigate U.S. export restrictions on Nvidia processors. By developing domestic hardware capable of supporting 500,000-card clusters, Alibaba aims to provide alternatives to foreign technology.
What happens next depends on the scheduled production timelines and the resolution of supply constraints. Alibaba Cloud plans to bring its AI supernodes online at a commercial scale during the current quarter of 2026. The Zhenwu V900 chip is set to enter mass production in early 2027. Investors and industry analysts will monitor whether these hardware and model milestones translate into the "significant growth" in shipments Wu projected amid robust customer demand.