Cheap Electricity May Decide the Next U.S.-China AI Contest
If American data centers are built more slowly, high prices for AI answers could let power-rich China undercut the U.S.

The AI Daily Brief: Artificial Intelligence News and Analysis · Washington Is Taking Control of AITraining an AI model gets the headlines, but answering user questions day after day is where the power bill comes due.
Each answer is sold in small units called tokens, and the price of those tokens depends heavily on running data centers. Those buildings full of chips use enormous amounts of electricity to do the work, known as inference, of responding to prompts.
That is where analysts see a possible opening for China. Scott Kennedy, of the Center for Strategic and International Studies, where he researches China's economy and business, said Democratic pressure for tougher China policy would "be constrained by the executive branch's wide range of latitude" and China's growing leverage, which "can't be easily undone."
Gary Ng, an economist at the bank Natixis, said "a slowdown in the U.S. could be a win for China." American data center building would not stop, he said, but "the pace will definitely slow."
Slower building would mean tighter computing capacity in the United States. That could keep American token prices high, because fewer data centers would be available to handle demand.
China, by contrast, is described as having plentiful electricity. With plenty of power, Chinese firms could run inference more cheaply and export those lower-cost answering services to customers elsewhere.
In that world, restricting China's access to advanced chips still makes strategic sense. Even if Beijing has power, it is harder to sell cheap answers at scale without enough chips to put that electricity to work.
Questions people ask
What is the difference between training AI and inference?
Training is the one-time, very expensive process of teaching a model by showing it vast amounts of text and other data. Inference is what happens afterward, every time the finished model answers a prompt, and it uses electricity continuously.
What is an AI token?
A token is a small chunk of text, roughly part of a word, that AI systems read and produce. Companies usually charge by the million tokens, so token prices are a shorthand for how much it costs to get answers from AI.
Why does electricity matter so much for AI?
Data centers draw power not only for the chips themselves but for cooling and other equipment, and they often need reliable power around the clock. Access to abundant, inexpensive electricity therefore directly lowers the cost of each answer.

