Alibaba plans trillion-parameter AI model and unveils new chip
 |
Alibaba plans 5 trillion to 10 trillion parameter AI model |
HANGZHOU — September 22, 2026
Alibaba plans to train a new artificial intelligence model with between 5 trillion and 10 trillion parameters as the Chinese technology company expands its strategy across AI models, chips and data centers.
Alibaba CEO Eddie Wu announced the plan on Tuesday at the company's annual Apsara Conference in Hangzhou. He said the company's Qwen team is continuing research into model architecture and data optimization, with the aim of improving AI's ability to handle more complex tasks over longer periods and moving toward artificial superintelligence, or ASI.
Parameters are variables that an AI model learns during training and are commonly used as a rough measure of model size. Alibaba's current flagship Qwen 3.8 Max model has 2.4 trillion parameters, meaning the planned system would be roughly two to four times larger.
Wu said Alibaba's M890 AI supernode can already run inference for models with more than 2 trillion parameters. He said only a small number of companies worldwide currently have that capability.
The Qwen team has also made progress on what Alibaba describes as recursive self-improvement. This approach involves AI systems identifying limitations, designing experiments and generating data that can be used to improve future versions of the system.
Alibaba also unveiled the Zhenwu V900, a next-generation AI processor developed by its T-Head semiconductor unit. Wu described it as China's most powerful AI chip and said it provides three times the performance of the M890.
A single cluster using the new chip can support up to 500,000 cards for training and inference of advanced AI models, Wu said. Alibaba also expects annual shipments of its AI chips to grow significantly.
The chip launch comes as Chinese technology companies work to develop domestic alternatives to Nvidia processors amid U.S. restrictions on exports of advanced computing technology to China. The restrictions have increased pressure on Chinese firms to develop their own AI hardware and computing infrastructure.
Alibaba also plans to expand its data center network. Wu said Alibaba Cloud's global data center capacity is expected to exceed 20 gigawatts by 2032, as demand for AI computing continues to increase.
Wu said customer demand for AI remained exceptionally strong and was helping accelerate Alibaba Cloud's revenue growth. He also acknowledged that shortages across the global AI data center supply chain were limiting how quickly the company could expand.
Alibaba Cloud plans to begin deploying its AI supernodes at commercial scale during the current quarter. Wu said demand over the medium and long term was significantly greater than the company's available supply capacity.
Wu described the broader development of AI as the beginning of a new era of "Machine Intelligence" that could be comparable to the Industrial Revolution. He also argued that AI coding is an early application rather than the defining product of the technology, saying more significant products have yet to emerge.
