Alibaba Unveils New AI Chip, Outlines Plan for Larger Model — Update

By Tracy Qu

Alibaba has rolled out its most powerful chip yet, doubling down on its bid to control the software and hardware that power its artificial-intelligence ambitions.

The new AI chip--the Zhenwu V900--outperforms its predecessor by a significant margin, Chief Executive Eddie Wu said Tuesday. With larger GPU memory and bandwidth, its performance is three times better than the chip Alibaba released just four months earlier, signaling companies' push to accelerate development as competition intensifies.

"We anticipate significant growth in annual AI chip shipment volumes," Wu said at Alibaba Cloud's annual conference in Hangzhou.

Alibaba has committed to spending more than $53 billion over three years to add AI capabilities to its traditional core business of e-commerce.

Failure to do so could see it fall behind as China's AI startups like DeepSeek and Moonshot AI keep making technological breakthroughs, while established giants like TikTok parent ByteDance also spend heavily to adapt to the AI boom.

Establishing a foothold in China's homegrown chip ecosystem would give Alibaba a lucrative advantage, especially as Beijing pushes to cut reliance on foreign semiconductors.

The new chip supports model training and inference. A single cluster built on the V900 can support up to 500,000 cards for model training and running frontier AI models, which means that many processors can work together as a single system to train models.

The Zhenwu V900 is scheduled for mass production and commercial release in the first quarter of 2027.

Adding more chipmaking capacity is not without its challenges. Alibaba is up against Huawei, whose Ascend chips lead the domestic market, as well as rivals such as Cambricon and Baidu's Kunlunxin.

Alibaba is continuing its efforts on the software side too, announce plans to train a new AI model with between 5 trillion and 10 trillion parameters, Wu said.

Parameters are variables learned during training that help determine a model's output, with the count used as a rough gauge of scale and capabilities. The more parameters a model has, the more it can do.

To put Alibaba's new AI model plan into context, its flagship Qwen 3.8 Max, released in July, has 2.4 trillion parameters.

Moonshot AI, a lab in which Alibaba owns a stake, released a model with 2.8 trillion parameters in July, catching many Silicon Valley competitors by surprise.

Alibaba didn't specify expenditure details about its new AI plans, though last month it raised about $10.2 billion through a Hong Kong share offering that it aims to use exclusively for AI investments.

The company's Hong Kong-listed shares rose as much as 5.1% during Tuesday trading, outperforming a 0.8% gain by the Hang Seng Tech Index.

Write to Tracy Qu at tracy.qu@wsj.com

(END) Dow Jones Newswires

September 22, 2026 02:34 ET (06:34 GMT)