Alibaba unveiled a new AI accelerator, a target to operate more than 20 gigawatts of global data-centre capacity and plans for an AI model with as many as 10 trillion parameters on Tuesday, laying out all three parts of its AI infrastructure strategy in a single keynote.
CEO Eddie Wu introduced the Zhenwu V900 at Alibaba Cloud’s Apsara Conference in Hangzhou, which runs from September 22 to 24. Wu said the new processor delivers three times the performance of the previous Zhenwu M890 and can be connected in clusters containing as many as 500,000 chips for training frontier models.
Wu described the V900 as “the most powerful AI chip in China today,” making clear that the claim is Alibaba’s own assessment rather than an independent benchmark.
Alibaba’s Hong Kong shares closed about 2% higher following the announcement. Its U.S.-listed shares had risen more than 3% before the opening bell, although that gain narrowed to around 0.7% in New York trading at the latest check.
Alibaba Wants More Than 20 GW of Data-Centre Capacity
The larger number from the keynote may be the infrastructure target rather than the chip specification.
Alibaba wants Alibaba Cloud’s global data-centre capacity to exceed 20 GW by 2032 as it builds the computing base needed to train and run increasingly large AI models.
The company has not changed its previously announced RMB 380 billion capital-expenditure plan. Citi analysts said the 20 GW target could translate into around $160 billion of external cloud revenue by fiscal 2033, roughly $60 billion above the bank’s existing $100 billion estimate for FY2031.
Citi said that would imply an external cloud revenue compound annual growth rate of roughly 40% from FY2026 through FY2033. The bank also noted that not all of the capacity necessarily has to sit directly in Alibaba’s capex budget, because infrastructure can also be secured through partners and operating-expense arrangements.
That buildout puts Alibaba on the China side of the same infrastructure cycle driving U.S. semiconductor valuations. FinanceFeeds recently tracked AMD’s first close above a $1 trillion market value, as investors continued to price in rising demand for AI compute.
The V900 Is Alibaba’s Attempt to Own More of the Stack
Alibaba’s T-Head semiconductor unit designed the V900 as the next step beyond the M890, giving the company an in-house accelerator alongside its cloud platform and Qwen models.
Benzinga reported that the V900 is scheduled to enter mass production and commercial release in the first quarter of 2027. Alibaba’s existing Zhenwu chips are already being used by customers across sectors including finance, automotive, energy and manufacturing, according to the report.
The strategy differs from a pure chipmaker because Alibaba can potentially consume its own processors inside Alibaba Cloud as well as offer AI services on top of them. That vertically integrated approach comes as Intel continues trying to turn manufacturing scale into external foundry demand, while Arm’s AI exposure remains tied primarily to the architecture licensed across the wider chip ecosystem.
Qwen Is Aiming for Up to 10 Trillion Parameters
The third part of Alibaba’s announcement sits above the hardware.
Wu said the Qwen team plans to train a next-generation model containing between 5 trillion and 10 trillion parameters and is exploring recursive self-improvement, where AI systems contribute to improving subsequent versions of themselves.
Taken together, the announcements show what Alibaba wants investors to value: not simply a larger Qwen model or another domestic Chinese accelerator, but an integrated stack spanning the model, the chip that trains it and the cloud infrastructure that supplies the compute.
The 20 GW target now puts a number on how large Alibaba expects that stack to become.