The AI Data Center Fight Reaches Your Power Bill

The AI build-out has spent two years as a story about chips, capital and model scores. This autumn it became a story about the electricity bill on the kitchen table. Communities across the political spectrum are pushing back on the data centers rising near them, and the reason is concrete rather than ideological: the power to run those buildings is expensive, and much of that cost is landing on ordinary customers.

The numbers behind the anger are stark. In the PJM grid region, which covers 13 states and Washington DC, capacity prices climbed from about $29 per megawatt-day in the 2024 auction to $329 for 2026 and 2027, according to the Institute for Energy Economics and Financial Analysis. That is more than a tenfold jump. Analysts attribute a large share of the increase to data center demand, and the resulting cost gets recovered from everyone connected to the grid. In Washington DC, Pepco residential customers saw bills rise by roughly $21 a month starting in the summer.

A bipartisan pushback

What is notable is who is complaining. This is not a left or right issue at the local level. Data centers have gone up quietly in parts of Louisiana, while similar projects in Montana and Virginia have drawn loud opposition at planning meetings. The common thread is not politics but the bill, and the sense that residents are being asked to subsidise infrastructure built for a handful of very large companies. The Natural Resources Defense Council has warned that if the full cost of serving new data centers is spread across household bills, families in the PJM footprint could pay tens of billions more over time.

A bipartisan group of state legislators is now pressing PJM to adopt ratepayer protections, so that the companies driving the demand carry more of the cost rather than leaving regular customers as what one framing called the unpaid sponsors of the AI boom. The details of who pays are being fought out in regulatory filings that rarely make headlines, but the outcome will shape electricity prices for years.

The industry's answer

The companies building this capacity are not blind to the problem. Some are trying to take pressure off the grid rather than add to it. Emerald AI raised a large round on the promise of making data centers flexible enough to ease off at peak times, and operators such as Crusoe pitch purpose-built sites with their own power arrangements. Anthropic's $35 billion Texas deal is partly a bet that building where power is cheaper avoids exactly this kind of fight.

It is worth keeping the energy question in proportion. The claim that a single AI query uses vast multiples of a web search has been examined here before and does not hold up as neatly as the viral versions suggest, and the water figures are similarly more modest per query than headlines imply. The aggregate is still real, though. A lot of buildings drawing a lot of power at once puts pressure on a grid that takes years to expand, and the bill arrives before the new transmission lines do. That timing gap, more than any single statistic, is what has turned a technical debate into a political one.