Data Centers Are 4% of America's Power. Why Is Everyone's Bill Going Up?

If you've opened your electric bill recently and felt a small jolt that had nothing to do with the wiring, you're not imagining it. Across large swaths of the country — Virginia, Ohio, Illinois, New Jersey, the Mid-Atlantic — household electricity rates have been climbing faster than inflation, and a familiar villain keeps getting named in the headlines: artificial intelligence. Data centers, we are told, are quietly draining the grid and sending your bill skyward.

Here's the headline-deflating part: U.S. data centers consumed roughly 183 terawatt-hours of electricity in 2024 — which the International Energy Agency itself describes as "more than 4%" of the country's total electricity consumption. Four percent. That's not nothing, but it's also not the kind of number that obviously explains a rate hike showing up in millions of mailboxes at once. So what's actually going on — and is the AI boom really to blame for your bill?

The Real Number Isn't the Share. It's the Growth Rate.

The 4% figure describes where things stand today. It says nothing about where they're heading — and that's where it gets interesting. The IEA estimates that data centers accounted for roughly half of all the growth in U.S. electricity demand in 2025. Not half of total demand — half of the increase in demand, in a single year.

That distinction matters because the U.S. grid spent roughly two decades in a period of relatively flat electricity demand. Power companies built their long-term plans, their rate structures, and their infrastructure investments around that assumption. Then, in the span of a few years, generative AI went from a research curiosity to the kind of build-out arms race that pushed Nvidia past Apple as the world's most valuable company — and demand forecasts utilities hadn't seriously revised in a generation suddenly needed rewriting, sharply, on a timeline of months rather than decades.

Pew Research Center, tracking the trajectory, projects that data center electricity consumption in the U.S. will grow roughly 133% by 2030 — climbing to an estimated 426 terawatt-hours a year. That's not a statement about today's share of the grid; it's a statement about how fast that share is about to move. When demand grows that quickly, prices respond — and they don't wait for the long-term planning cycle to catch up.

How "Their" Demand Becomes "Your" Bill

This is the part that trips people up: a data center in Virginia doesn't send Ohio residents a separate invoice. So how does its electricity appetite end up on a bill hundreds of miles away? The answer runs through something most consumers have never heard of — the capacity auction.

In the PJM Interconnection — the regional grid operator covering 13 states and Washington, D.C., serving roughly 65 million people — power plants don't just sell electricity when it's generated. They also get paid, years in advance, simply for promising to be available when the grid needs them. That promise is bought and sold at auction, and the clearing price of that auction gets baked into the rates utilities charge everyone connected to the grid — whether or not a given household lives anywhere near a server farm.

And those auction prices have moved dramatically. PJM's 2025/2026 capacity auction cleared at $269.92 per megawatt-day — up roughly ninefold from $28.92 the year before. Monitoring Analytics, the grid's independent watchdog, attributed 63% of that jump — about $9.3 billion — directly to data center growth, current and forecasted. The next auction, for the 2026/2027 delivery year, cleared even higher still: $329.17 per megawatt-day, the maximum allowed under PJM's price cap. Two consecutive record-breaking auctions, back to back — and both bills landed on the same ratepayers.

That's the mechanism in a sentence: the grid plans years ahead, AI demand arrived faster than anyone planned for, and the cost of closing that gap gets spread across everyone the grid serves — not billed directly to whoever's renting the server racks. It's a dynamic with an odd parallel to how AI's labor disruption is spreading: the technology's costs and benefits don't always land on the people closest to it.

The Bill Doesn't Care Where the Server Is

It would be convenient if this were a story with a single villain — point at the data center, blame the AI boom, move on. The reality is messier and, honestly, more useful to understand. Grids are shared infrastructure by design; that's the whole point of building them regionally instead of one plant per building. The same interconnection that lets a hospital in Pittsburgh draw on a power plant in West Virginia is the one that lets a data center's projected demand in Loudoun County ripple into a rate filing in Columbus.

What's changed isn't the structure — it's the speed. Utilities and grid operators built their forecasting models for a world where demand crept upward by a percent or two a year, predictable enough to plan a decade out. Generative AI didn't creep; it arrived in a rush, and the auctions, rate cases, and infrastructure timelines that were built for gradual change are now absorbing a shock they weren't designed for. Some of that gap will likely be closed by faster build-outs of new generation — solar capacity additions have already had their best run in over a decade — but new power plants take years to permit and build, while AI demand is arriving in quarters. Regulators in several states are already weighing proposals that would have large data center operators shoulder more of the infrastructure cost directly, rather than letting it fold into the general rate base — an acknowledgment that the current setup is straining under a kind of growth nobody priced in.

None of that changes what shows up in your mailbox this month. But it does change how to read the next headline about AI's energy appetite. The question worth asking isn't just "how much power do data centers use" — it's "how fast is that number moving, and who absorbs the gap while the grid catches up." Right now, the answer to that second question is: largely, you do. Whether that stays true is less a matter of how much AI grows, and more a matter of whether the systems built for a slower world get rebuilt fast enough to keep pace with the one we're actually in.