Alibaba plans AI model with 5 trillion to 10 trillion parameters, unveils new chip

Alibaba Group aims to develop a new artificial intelligence model with parameters ranging from 5 trillion to 10 trillion, according to chief executive Eddie Wu. Wu shared this announcement during Alibaba Cloud’s annual Apsara Conference in Hangzhou, outlining the company’s strategy encompassing AI models, chips, and data centres.

The Qwen team is working on model architecture and data optimisation, aiming to tackle more complex, long-horizon tasks and move towards artificial superintelligence (ASI). The upcoming Qwen 4.5 and Qwen 5 series are projected to scale up to the specified parameter range. Currently, Alibaba's flagship model, Qwen 3.8 Max, has 2.4 trillion parameters, making the planned model roughly two to four times larger.

Alibaba’s M890 AI supernode can handle inference for models above 2 trillion parameters, a capability held by only a few companies globally. The company's next-generation AI chip, the Zhenwu V900, is three times more powerful than its predecessor, the M890, and will support up to 500,000 cards for frontier model training and inference.

Expected to enter mass production and commercial release in the first quarter of 2027, the chip is part of Alibaba's plan to surpass 20 gigawatts of global data centre capacity by 2032. The launch of these AI advancements comes amid Chinese firms racing to develop domestic alternatives to Nvidia’s processors due to US export restrictions.

Wu emphasized that customer demand for AI remains exceptionally robust, driving Alibaba Cloud’s revenue growth, but acknowledged that global supply chain shortages are limiting the company's expansion. He framed the development as the beginning of a new era of "Machine Intelligence," predicted that machines would eventually produce more than 1,000 times the "thinking" of all humanity, and likened AI coding to the light bulb of the electrical age, emphasizing that the truly groundbreaking products of this era have yet to arrive.