A new analysis getting attention Thursday puts China's operational capacity for running artificial intelligence computing at about 24 gigawatts, a measure of electrical power that reflects how much AI computing infrastructure, mainly data centers packed with specialized chips, a country currently has up and running. That is closing in on the roughly 56 gigawatts estimated for the United States, according to the report from research firm SemiAnalysis, covered by the Financial Times.
The gap may close further. The same analysis estimates another 50 gigawatts of capacity is planned or already under construction in China, spread across more than 1,000 facilities the researchers tracked. Gigawatts are used here as a stand in for the sheer physical scale of AI infrastructure, since running large numbers of advanced chips around the clock requires enormous and steady amounts of electricity.
One notable distinction the analysis draws is around what is actually limiting growth in each country. In the United States, the pace of new AI data center construction has increasingly run into bottlenecks around permitting and securing enough electrical power from local utilities. In China, the report describes the opposite dynamic, with state utilities and local governments generally treating new AI electricity demand as a welcome source of growth, which has made power availability less of a constraining factor there, even as the broader buildout still depends on continued access to advanced computer chips.
The comparison matters because computing capacity, the physical ability to train and run large AI models, has become one of the clearest ways to measure which countries and companies are positioned to lead in AI development over the coming years, alongside access to advanced chips and large pools of engineering talent.
Watch for how quickly the additional planned capacity in China actually comes online, and for any updates on access to the advanced chips that ultimately determine how much of that physical capacity can be put to use.
Why measure AI infrastructure in gigawatts?
Running large numbers of advanced AI chips requires enormous, steady amounts of electricity, so power capacity is a practical way to measure the physical scale of a country's AI computing infrastructure.
Does more computing capacity guarantee AI leadership?
It is one important factor, alongside access to advanced chips and engineering talent, but not the only one that determines a country's position in AI development.