Alibaba Cloud wants its global data center footprint to reach 20 gigawatts within the next six years. Eddie Wu, Alibaba's chief executive, laid out the target at the company's Apsara conference.
He also introduced the Zhenwu V900, a chip developed by Alibaba's T-Head unit. Wu says it delivers three times the performance of its predecessor, the Zhenwu M890, and calls it the most powerful AI chip in China today.
Inside the V900
The chip carries 216 gigabytes of memory and a chip-to-chip link running at 1,200 gigabytes per second. It supports number formats from FP32 down to FP4, with its compute units tuned for FP8 and FP4.
According to the company, a single cluster can hold up to half a million of the chips. Alibaba did not say when production will start.
Models and regions
The Register reports that Alibaba is now training Qwen 4 and plans two successors with five trillion and 10 trillion parameters. After the keynote, the company said it will open new cloud regions in Türkiye, Finland and the Netherlands within 12 months.
Wu framed the push in broad terms. "Machine power already drives 99.9% of the world's physical work," he said, adding that he expects AI to eventually handle a similar share of cognition.
For scale
The United States currently has 37.7 gigawatts of data center capacity under construction. Meta is building a five-gigawatt campus, and the Stargate project is aiming for 10 gigawatts by 2029.
