Alibaba’s New AI Chip and Model Plan Put Compute at the Center of China’s Tech Push

A chip announcement with a wider ambition

Alibaba has unveiled a new artificial intelligence chip and outlined plans for a much larger AI model, placing computing capacity—not just model design—at the center of its bid to compete with leading U.S. technology companies.

At the company’s annual conference in Hangzhou on Tuesday, Sept. 22, CEO Eddie Wu said Alibaba’s Zhenwu V900 is China’s most powerful AI chip and can deliver three times the performance of the company’s previous-generation M890. Those performance figures are Alibaba’s claims; the company did not provide independent benchmark results in the announcement reported by The Associated Press.

The chip is intended for the demanding calculations involved in training AI systems and generating their answers. Alibaba uses Zhenwu processors in its data centers, which supply computing power for its own services and cloud customers. The announcement matters because access to advanced chips and the ability to deploy them at scale remain key constraints in the global AI race.^[1]

A jump in planned model scale

Alibaba also said it plans to train a future AI model with between five trillion and 10 trillion parameters. Parameters are adjustable values within a model that help determine how it processes information. They are one measure of scale, but not a standalone measure of quality: real-world capability depends on factors including training data, computing methods and performance on relevant tasks.

The proposed model would be a substantial increase over Alibaba’s current top Qwen system. The company’s Qwen3.8-Max model has 2.4 trillion parameters, according to AP. Alibaba did not announce a release date for the larger model or provide benchmark scores demonstrating how it would compare with rival systems.

That distinction is important for readers and customers weighing headline numbers. The five-to-10-trillion figure describes a plan, not a finished product or a verified performance result. For now, the concrete development is the combination of a newly announced chip and a stated intention to build models at a much larger scale.^[1]

Compute capacity is the strategic wager

Alibaba’s chip and model announcements are tied to an infrastructure push. The company says it expects to have more than 20 gigawatts of computing capacity by 2032, citing rapidly rising demand for AI computing. Wu said shortages across the data-center supply chain are already limiting how quickly Alibaba can expand its infrastructure.

Gigawatts measure power capacity, not the intelligence or quality of a model. Still, the target points to the scale of investment companies believe will be needed to train and serve increasingly demanding AI systems. It also helps explain why Alibaba is developing both hardware and models while expanding its cloud operations: control over more of the computing stack could help it serve customers even when access to outside chips is constrained.

That ambition is not a guarantee of execution. Building capacity requires more than designing a processor. Data centers need power, cooling, networking and reliable supplies of components. Analysts cited by AP said China’s ability to manufacture advanced chips will also affect whether it can close the gap with U.S. competitors.^[2]

Export limits sharpen the self-reliance drive

The announcement comes as U.S.-led export restrictions limit China’s access to some advanced AI chips and chipmaking equipment. Chinese AI developers have continued to build systems, and Alibaba’s new hardware plans underscore Beijing’s broader push for domestic technology alternatives. The company’s effort does not, by itself, show that China has overcome manufacturing or supply-chain constraints.

Some Chinese AI models have gained attention overseas partly because they can be more affordable than leading closed systems, AP reported. But a chip specification or a large parameter count alone cannot establish that a model is cheaper to operate, broadly competitive or more useful for a particular task. Independent performance comparisons and details about availability will matter as the products develop.

Alibaba’s move also arrives amid heightened U.S.-China competition over artificial intelligence. AP reported that Chinese President Xi Jinping was due to meet U.S. President Donald Trump in Washington days after the announcement, with AI among the expected topics. The timing gives the company’s plans strategic visibility, but the announcement itself remains a corporate technology development rather than evidence of a shift in government policy.^[1]

What to watch next

For businesses considering Alibaba’s cloud services or Qwen models, the useful takeaway is to watch for deployed products and independently checkable results—not just targets. Key questions include when the V900 becomes broadly available, how it performs in training and inference workloads, and whether Alibaba can deliver the promised capacity amid infrastructure bottlenecks.

For the wider market, Alibaba’s AI chip announcement shows how competition is increasingly about the full stack: processors, data centers, cloud access and models. The company has made a concrete hardware announcement and disclosed an ambitious model plan. Whether those plans translate into dependable, competitive computing will be tested by manufacturing, supply and actual performance.

That makes the next milestones more informative than the headline numbers: shipping hardware, operational capacity and published evaluations could show whether Alibaba’s push changes the choices available to developers and cloud customers. Until then, the five-to-10-trillion-parameter model remains a stated target, while the V900’s claimed performance awaits broader verification.^[1]