Alibaba unveils Zhenwu V900 chip and sets Qwen target at 5 to 10 trillion parameters, shares jump 5%

Start with the chip itself. Per Reuters, the Zhenwu V900 was developed by T-Head Semiconductor, delivers three times the performance of its predecessor M890, which launched in May, and can be linked in clusters of up to 500,000 chips to train and run the largest AI models, with mass production and commercial release planned for the first quarter of 2027. Alibaba chief executive Eddie Wu said the company expects significant growth in annual AI chip shipments. TechNode reports that T-Head also showcased its full-stack chip solutions at the conference and held hands-on workshops around the T-Head SAIL software stack, covering model training, inference and software optimization. In TechNode's framing, the V900 is part of a broader effort at Alibaba to build an AI computing stack spanning chips, networking, storage and software rather than relying on standalone accelerators.

The model half of the announcement is blunter than the chip half. Per Reuters, Eddie Wu said Alibaba's Qwen team plans to train a new model with 5 trillion to 10 trillion parameters to tackle more complex, longer-horizon tasks as it pursues artificial superintelligence, a system that surpasses human capabilities. For comparison, Alibaba's current flagship, Qwen 3.8 Max, has 2.4 trillion parameters, a rough measure of a model's size and capability. In a separate statement, Alibaba said it is currently training its next-generation model, Qwen 4, with future Qwen 4.5 and Qwen 5 models expected to scale up to 5 trillion to 10 trillion parameters. Wu also said the Qwen team had made meaningful progress in getting models to improve themselves by identifying weaknesses, running experiments and generating training data with limited human involvement.

The third point is why these two threads were announced together. Reuters frames the backdrop as Chinese technology companies racing to develop domestic alternatives to Nvidia's AI processors amid tightening US export curbs. Put chip, model and data center side by side and they are three stages of one problem: bigger models need more compute; the more that compute depends on external supply, the more exposed you are to controls; so from foundation models to semiconductors to the data centers needed to train and deploy increasingly powerful systems, Alibaba is pushing on every layer. The report notes these efforts range from foundation models and semiconductors to the data centers themselves.

The fourth point is what Wu said on stage, which reveals how the company reads the current phase. Per Reuters, he said the truly groundbreaking products of the Machine Intelligence era have not yet arrived and likened the coming period to the Industrial Revolution; he predicted machines would eventually produce more than 1,000 times the thinking of all humanity, up from less than 3% today. He also likened AI coding to the light bulb of the electrical age, an early application rather than the breakthrough product itself. Read alongside the product list announced that day, the message is fairly clear: the chip and the bigger model are groundwork, and the company does not treat today's application shapes as the endpoint.

The fifth point is the infrastructure target and the constraint it runs into. Per Reuters, Wu set a target for Alibaba Cloud's global data center capacity to surpass 20 gigawatts by 2032. He said customer demand for AI was exceptionally robust and was accelerating Alibaba Cloud's revenue growth, but that supply chain constraints limit the pace of expansion; his words were that the industry's mid-to-long-term demand far outpaces our supply capabilities. He added that Alibaba Cloud would begin bringing its AI supernodes online at commercial scale this quarter. China Daily reports he told the conference that Alibaba Cloud would go all-in on building the infrastructure needed to support increasingly powerful AI models and applications.

Finally, the market reaction, and one note of caution. Per Reuters, Alibaba's Hong Kong-listed shares rose 5.1% the day of the announcement to their highest in a month. Worth flagging: the two numbers drawing the most attention in this round of announcements, a 5 to 10 trillion parameter model and a first-quarter 2027 production date, are both plans rather than delivered facts. The model is still in training, and the chip's production timeline is the company's own schedule. Parameter count also has no linear relationship to performance, a point the industry has debated for some time. Every fact and figure here comes from Reuters reporting as syndicated through the McClatchy media network, plus TechNode and China Daily, with no speculation added.