Turns out training large language models to write mediocre marketing copy costs more than anyone expected — specifically, it costs 67 million Americans a 76% jump in wholesale electricity prices in three months.
PJM Interconnection, which operates the largest power grid in the U.S., just released a report that reads like a utility company’s hostage note. Wholesale electricity prices in the mid-Atlantic, Midwest, and South averaged $136.53 per megawatt-hour in the first quarter of 2026 — up from $77.78 in the same period last year. The culprit? AI data centers are sucking down power faster than the grid can supply it.
The report doesn’t mince words: “Data center load growth is the primary reason for recent and expected capacity market conditions, including total forecast load growth, the tight supply and demand balance, and high prices.”
Translation: this is what happens when you build the infrastructure for an AI arms race before checking whether the electrical grid can handle it.
AI Data Center Electricity Costs Are Hitting Ratepayers Hard
PJM’s territory covers 13 states — Delaware, Illinois, Indiana, Kentucky, Maryland, Michigan, New Jersey, North Carolina, Ohio, Pennsylvania, Tennessee, Virginia, and West Virginia. That’s nearly 20% of the U.S. population living in an area where the grid is now buckling under the weight of servers training models to generate stock photos of business handshakes.
The wholesale price spike doesn’t stay wholesale. It trickles down to retail customers — homeowners, small businesses, anyone plugged into the grid. And the report makes clear this isn’t a temporary blip: “Large data center load additions have already had a significant and irreversible impact on PJM customers that will be paid through May 31, 2028.”
Irreversible. That’s the word they used.
The grid simply doesn’t have enough capacity to meet the demand from data centers built to power the AI boom. And there’s no quick fix — no magic switch to flip that suddenly generates enough electricity to keep the lights on and train the next iteration of GPT.
The Power Mix Is Shifting — But Not Fast Enough
In response to the surge in demand, the energy mix has adjusted slightly. Coal generation dropped 1.7%, natural gas rose 4.2%, oil spiked 43.2%, wind fell 4.7%, and solar climbed 15%.
Oil up 43%. That’s not a typo.
Solar’s 15% increase sounds promising until you realize it’s still a fraction of the total load. Wind actually decreased — likely due to intermittency issues and the fact that you can’t just summon more wind when data centers need power at 3 a.m.
The grid is scrambling to keep up, burning whatever it can to meet demand. The result is higher AI data center electricity costs passed directly to consumers, with no end in sight until at least 2028.
Americans Would Rather Live Next to a Nuclear Plant Than a Data Center
A new Gallup poll dropped this week with a stat that should make tech executives nervous: 71% of Americans oppose data center construction in their area. Among those, 48% are strongly opposed.
More telling? Only 53% oppose nuclear power plants in their neighborhood. Meaning more Americans would rather have a reactor in their backyard than a warehouse full of GPUs churning out AI-generated podcast transcripts.
The top concern? Strain on local resources— specifically, electricity. People are connecting the dots between data centers moving in and their power bills going up. They’re not wrong.
What Happens When the Grid Can’t Keep Up?
The PJM report lays out a grim forecast: higher transmission costs, higher energy market prices, and higher capacity market prices. All baked in through 2028 at minimum.
And that’s assuming no new data centers are built — which is laughable given how much capital is pouring into AI infrastructure. Every tech giant is racing to build more capacity, more compute, more training clusters. The grid isn’t expanding fast enough to match.
So what gives? Either electricity prices keep climbing, or utilities start imposing limits on data center connections, or brownouts become routine in areas with heavy data center concentration. None of those options is great.
There’s also the possibility that data centers start building their own dedicated power plants, which some are already exploring. But that takes years, requires regulatory approval, and doesn’t solve the immediate problem of a grid already stretched past capacity.
The AI Boom Has a Hidden Price Tag
The conversation around AI tends to focus on capabilities, ethics, job displacement, copyright issues. Fair enough. But the infrastructure costs — the literal megawatts required to keep the whole thing running — rarely get the same attention.
AI data center electricity costs aren’t an externality. They have a direct, measurable impact on millions of people’s utility bills. And unlike software bugs or model hallucinations, you can’t patch a power grid overnight.
The PJM report is a rare moment of clarity from an industry that usually speaks in euphemisms. The grid is strained. Prices are up. The impact is irreversible. And ratepayers — not tech companies — are absorbing the cost.