Electricity is getting more expensive for tens of millions of Americans, and AI is part of the reason. In PJM, the grid that serves 67 million people across 13 states from Illinois to New Jersey, the price utilities pay to guarantee future supply jumped from 29 dollars per megawatt-day to 270 in a single year, a 9.3-fold increase. The auctions after that hit the federal price cap of 329. Run through to a household bill, that works out to roughly 25 to 30 extra dollars a month and an average rise of about 15 percent against the era before the data-center surge.
The demand is real and it is large. PJM's own market monitor found that stripping all data centers out of the forecast would cut those capacity payments by about 9.3 billion dollars, nearly two-thirds of the total. Roughly 8 gigawatts of new data-center load showed up in the most recent delivery year, with 12 expected next. A federal watchdog warned that the impact on customers has been very large and is not reversible, with more pressure due at the next auction.
The part the headlines skip
Here is where it gets more complicated than the angry-headline version. A detailed analysis from SemiAnalysis argues that much of the price shock traces to how PJM's market is designed, not just to how many servers want to plug in. The capacity auction uses a steep pricing curve that turns a modest supply crunch into a dramatic payout, and PJM has repeatedly overestimated how fast new demand would actually arrive, revising its forecasts down year after year. The same buildout in Texas, which has no equivalent capacity auction, pushed forward power prices up only 11 to 17 percent rather than ninefold.
The reliability case for those big payments also took a knock. During Winter Storm Fern in January, PJM lost around 21 gigawatts of generation, roughly 15 percent of its fleet, despite collecting record sums meant to keep the lights on. Texas weathered the same period without the price spike. The lesson is not that data centers are blameless. It is that a poorly tuned market amplifies their effect, and ratepayers absorb the difference.
Real cost, wrong panic
This is the grounded version of an anxiety that often runs to extremes. We have written before about the claim that AI will boil the oceans, which does not survive contact with the numbers. The power-bill story is different because it is measurable and it is landing now, on real statements, in regions where the build-out from companies like SoftBank and the hyperscalers is concentrated.
What to watch is who ends up paying. Regulators and lawmakers are increasingly pushing to make data-center operators carry more of the cost of the infrastructure they require, through special rate classes and direct grid contributions, rather than spreading it across every household. The technology is not the only variable here. The rules around it are, and those are still being written.
Commentarii · 0