Ken Apaki keeps a spreadsheet. Actually, Ken keeps a lot of spreadsheets—his grandkids’ heights, household expenses, and most relevant to this story: his electricity bill.
On Planet Money, NPR visited Ken and his wife Carol in central Ohio to understand what’s happening to American electricity prices:
11 cents per kilowatt hour in 2020. 19 cents now. Nearly doubled in five years. Same house. Same AC unit. Same usage patterns.
Ken has a theory about why.
130 Data Centers
The Apakis live in central Ohio, and they’ve watched their neighborhood transform:
130 data centers now dot central Ohio. Google, Meta, Amazon—they’ve all moved in. And they’re all pulling massive amounts of electricity from the same grid that powers Ken and Carol’s home.
The scale of investment is staggering:
“We are spending more to build data centers than we spent to build the entire interstate highway system.”
The Diesel Generator Scandal
The power demands of AI are so acute that some companies aren’t even waiting for the grid. On AI in Business, host Jaden Schaefer described what happened with Elon Musk’s xAI:
xAI’s Memphis data center couldn’t get enough power from the grid. Their solution? Thousands of portable diesel generators, all running simultaneously to power their AI training clusters.
It’s a stark illustration of just how desperate the demand has become.
The $15 Billion Plan
The Trump administration, working with a bipartisan group of governors, is pushing a solution: make tech companies pay for new power plants.
The plan asks PJM Interconnection—the grid operator serving 65 million Americans across 13 states—to hold an emergency auction. Tech companies would bid on 15-year power contracts to underwrite construction of new generation capacity.
The logic: if data centers are driving demand, data centers should pay for supply.
Why Solar and Wind Aren’t Enough
The obvious question: why not just build more solar and wind? The answer comes down to a fundamental limitation:
“Solar and wind are intermittent. They only generate when the sun is shining or the wind is blowing. Data centers don’t get to take breaks. They need firm, dependable 24/7 electricity.”
Battery storage helps but can only cover a few hours. For continuous AI workloads, you need baseload power—which means natural gas, nuclear, or bringing old plants back online.
That’s why Microsoft paid a billion dollars to recommission Three Mile Island’s nuclear plant. It’s not nostalgia; it’s physics.
The China Problem
There’s also a competitive dimension. From AI in Business:
China is building electrical capacity far faster than the United States. If America can’t power its AI infrastructure, the training runs—and the technological advantage—will happen elsewhere.
Who Pays?
The fundamental tension is simple: when data centers connect to the grid without ensuring new generation gets built, the risk transfers to regular ratepayers.
Ken’s electricity bill doubled. He didn’t ask for 130 data centers in his backyard. But he’s paying for the strain they put on the grid.
The Trump administration’s plan is an attempt to realign incentives—make the biggest demand drivers bankroll the expansion. Whether it works, and whether it comes fast enough to prevent further price spikes, remains to be seen.
The Bottom Line
We’re spending more on data centers than we spent on the interstate highway system. AI models require power that solar and wind can’t reliably provide. And the costs are showing up in household electricity bills across America.
This isn’t a future problem. It’s Ken’s spreadsheet, today.
Search for more podcast coverage of AI infrastructure and energy on Audioscrape.