AI isn’t to blame for rising US electricity bills — but it soon will be
Forecast capacity needs are rising, perhaps by a tenth over the next three years
Rising power prices, long a political hot potato in the UK, are now burning fingers in the US too. Households there paid 30 per cent more for electricity than they did in 2020, according to the Energy Information Administration. One difference is that in the US, it’s AI data centres that are increasingly being fingered as the culprit.
Data centres are electricity guzzlers, for sure. But even though they are growing rapidly, they accounted for perhaps 4 per cent of electricity demand in 2024, according to Pew Research. Outside major data centre hotspots the rise in bills is driven more by the price of gas — last year, it was 60 per cent higher than in 2024 at the hub — and the investments needed to upgrade creaking electricity grids.
Nonetheless, the AI boom is already making its mark on energy prices. At PJM, which runs the electricity grid in much of the eastern US including Virginia’s Data Center Alley, the total cost of wholesale power increased by 44 per cent last year according to Monitoring Analytics.
About a third of that increase comes from PJM’s so-called capacity auctions, where power generators say what it would take to induce them to provide the amount of power the grid needs — which then sets the price all generators get and in turn influences what customers pay. Forecast capacity needs are rising, perhaps by a tenth over the next three years, according to Aurora Energy. New power plants can’t be built that quickly.
That’s a recipe for spiking prices. In the auction for the year ending in May 2028, held in December last year, PJM set a maximum level for bids, but it proved too low. Moreover, even that cap was a big step up from the pre-ChatGPT era, so ordinary folk will end up paying more through their bills.
That has politicians howling. The White House, presumably sensing that energy bills are a voting issue, wants tech giants to pay the cost associated with data centres. In essence, that would involve splitting the market in two. On one side would be ordinary citizens, whose power needs would be met at a lower price. On the other, the Meta Platforms and Alphabets of this world would be free to bid whatever they liked. The catch is that anyone who has electricity to sell might try to get it into the blowout tech auction, weakening the ordinary one.
Alternatively, the tech giants could try to simply lean on the grid less heavily. Perhaps they could agree to flip the switch off at peak times, turning to their own off-grid power sources instead. Microsoft boss Satya Nadella warned this week that AI companies are dependent on “social permission” for their access to power and other resources.
In reality, the AI premium on household bills is a tomorrow problem for most of the US. But it is one that tech giants, utilities and politicians should be thinking about now. Otherwise they may find that the AI boom quite literally runs out of juice.