One data center uses as much power as 100,000 homes. Guess who's paying for the grid.

The demand shock: AI needs a second grid

Every ChatGPT query, every generated video, every model training run happens in a building full of GPUs. Those buildings are getting enormous. A typical hyperscale data center draws about 100 megawatts of continuous power — as much as a mid-sized city's worth of homes. The newest AI campuses are planned in gigawatts.

The US grid wasn't built for this. Electricity demand was roughly flat for two decades — utilities planned for efficiency gains, not for an industry that wants city-sized loads connected on a two-year timeline. Now the grid needs new generation, new transmission, and new substations, all at once. Someone has to pay for that buildout.

Spoiler: it's you.

How a data center 40 miles away raises your bill

You don't share a meter with a data center, so the connection isn't obvious. It happens through three mechanisms, all perfectly legal and mostly invisible:

The capacity auction is the sharpest example. PJM — the grid operator covering 13 states from Illinois to New Jersey — held a capacity auction that hit record highs as data center demand collided with retiring power plants. The result landed directly on residential bills: Baltimore-area households saw bills jump more than $17 a month, with another increase of up to $4 coming in mid-2026.

And here's the part that stings: while households pay full retail rates for the buildout, large data centers frequently negotiate discounted industrial rates as a condition of choosing a location. Home electricity bills are skyrocketing; for data centers, not so much.

The receipts: what bills are actually doing

Want your own number? Our Data Center Next Door tracker shows what a typical home in your state pays now versus a year ago, using EIA's latest data.

For a decade — 2013 to 2023 — electricity prices basically moved with inflation. That correlation has broken. Goldman Sachs expects electricity prices to outpace inflation at least through 2026, driven specifically by data center demand. This isn't a one-time adjustment; it's a structural shift in who the grid is being built for.

The concern has gone mainstream: a November 2025 nationally representative survey found 78% of Americans worried that new data centers will push their bills up. In states like California, regulators are now studying whether data center growth will force residential rates higher — in a state where electricity is already among the most expensive in the country.

The bubble connection: heads they win, tails you pay

If you've read our AI bubble breakdown, you know the money flows in circles — Nvidia invests in companies that buy its GPUs, clouds fund startups that spend the money on cloud compute. The electricity story is where that circle finally touches the real economy, because power plants and transmission lines are not speculative assets. They take decades to pay off.

That creates an asymmetry worth understanding. If AI demand keeps growing, the infrastructure gets used and ratepayers subsidized Big Tech's buildout. If the AI bubble deflates and projected demand doesn't materialize, the grid is left with overbuilt capacity — and utilities recover those costs from ratepayers anyway. Utility cost recovery doesn't have a "the bubble popped" clause.

Either way, the bill arrives with your name on it, not theirs.

Utilities are the line item nobody budgets seriously — until it's up 20% in a year. If you're modeling your monthly costs, don't treat that line as fixed.