Alibaba unveiled a new AI chip, the Zhenwu V900, and set a target of more than 20 gigawatts of global data center capacity by 2032 at its cloud unit's annual Apsara Conference in Hangzhou on Tuesday, Sept. 22, CNBC reported. CEO Eddie Wu introduced the chip in his keynote and called it "the most powerful AI chip in China today."
The chip
The V900 was designed by T-Head, Alibaba's chip subsidiary, and handles both AI training and inference. Alibaba says it delivers three times the performance of the Zhenwu M890, which it released in May. According to TechNode, the V900 has:
- 216GB of memory
- 1,200GB/s of chip-to-chip bandwidth
- Native support for the low-precision FP8 and FP4 formats used to run large AI models efficiently
- Support for clusters of up to 500,000 chips, with more than 1,000 able to work together as a single system
The chip is scheduled for mass production and commercial release in the first quarter of 2027, according to CNBC, though The Register reported that Wu did not give a production date on stage. Alibaba has not disclosed the chip's manufacturing process or which foundry will make it. More than 650 customers already use Zhenwu chips, CNBC reported.
The data center target
"Our target is that by 2032, the global data center capacity operated by Alibaba Cloud will surpass 20GW," Wu said, according to a company release carried by ANI. Alibaba did not say how much capacity it runs today. Wu also said global supply shortages are "currently limiting the speed at which we can scale our compute infrastructure," the Associated Press reported.
Alibaba did not give a spending figure for the 20GW target. Its existing plan, announced in February 2025, is to invest 380 billion yuan (~$53 billion) in AI and cloud infrastructure over three years, per Alibaba Cloud. By the end of June, it had spent 190 billion yuan (~$27 billion) of that, the Alibaba-owned South China Morning Post reported. Alibaba also recently raised about $10.2 billion in a Hong Kong share offering earmarked for AI, The Wall Street Journal reported.
Qwen roadmap and the stock
Alibaba also said its next large language model, Qwen 4, is in training, with Qwen 4.5 and Qwen 5 to follow. The South China Morning Post reported that the roadmap scales to a model of 5 trillion to 10 trillion parameters. Alibaba's current flagship, Qwen 3.8-Max, has 2.4 trillion. The company's previous generation, Qwen3.7-Max, launched in May.
Alibaba shares rose as much as 5.1% in Hong Kong on Tuesday, against a 0.8% gain in the Hang Seng Tech index, the Journal reported. CNBC put the gain at about 3%.
Beijing has blocked purchases of Nvidia's H200, according to Brookings, while Nvidia CEO Jensen Huang recently said the company expects to sell twice as many chips next year.
Comments