Data Centers Drove 63% of the Largest Electricity Price Spike in U.S. History. Here's What It's Costing You.
PJM Interconnection's capacity auction prices surged from $28.92 to $329.17 per MW-day in two years. Monitoring Analytics attributes 63% of the increase to data center load forecasts. Across 13 states, 67 million ratepayers are now subsidizing AI infrastructure through bills they can't opt out of.
In December 2022, the cost of keeping the lights on in America's largest electricity market cleared at $28.92 per megawatt-day. That number set the capacity price for PJM Interconnection, the regional transmission organization that manages the electric grid for 67 million people across 13 states and the District of Columbia, stretching from New Jersey to Illinois.
Two and a half years later, that price is $329.17 per MW-day. An 11.4x increase. It is the largest capacity price spike in the history of organized U.S. electricity markets.
Who caused it? Monitoring Analytics, PJM's own independent market monitor, ran a counterfactual analysis on the 2025/2026 auction results. Their conclusion: data centers were responsible for 63% of the price increase, translating to $9.3 billion in costs recovered from customers across the PJM region through higher electricity rates.
The Arithmetic Nobody Ran
PJM serves roughly 67 million people. At an average household size of 2.5 persons, that is approximately 26.8 million households. Total capacity procurement rose from an estimated $4.1 billion at the old $28.92/MW-day rate to $16.1 billion at the new prices, according to PJM's own 2026/2027 BRA report. That is roughly $12 billion per year in additional capacity costs spread across the region.
Divide $12 billion by 26.8 million households: $448 per household per year in capacity cost increases alone.
Apply Monitoring Analytics' 63% attribution to data centers: $282 per household per year, or about $23.50 per month, is the implicit cost of data center demand on each PJM household's electricity bill. Nobody approved it.
That figure aligns with reported rate increases across the region: in Washington, D.C., Pepco residential customers saw bills increase by an average of $21 per month starting in June 2025, with Western Maryland at $18 and Ohio at roughly $16.
| PJM Region | Monthly Increase | Annual Impact | % of Avg. Bill |
|---|---|---|---|
| Washington, D.C. (Pepco) | $21 | $252 | ~15% |
| Western Maryland | $18 | $216 | ~13% |
| Ohio | $16 | $192 | ~12% |
| Northern Virginia (Dominion) | Est. $25+ | $300+ | ~16%+ |
| PJM-wide average | ~$23.50 | ~$282 | ~16% |
For context, the average American household pays roughly $145 per month in electricity, making the data center capacity surcharge roughly 16% of a typical PJM household's annual electricity spending. The average U.S. household pays about $500 per year in local property taxes toward public schools. The implicit data center electricity subsidy is now 56% the size of what families pay for schools.
How a Capacity Market Creates a Hidden Subsidy
PJM's capacity market is a forward auction, conducted three years ahead, where generators bid to be available during peak stress periods, the clearing price gets paid to all generators who clear, and every customer in the region pays through their utility bills regardless of who caused the demand to grow. A retired teacher in rural Ohio and a hyperscale data center in Loudoun County pay into the same capacity pool, and the teacher has no say in what the data center consumes.
That mechanism explains how load forecasts became a $12 billion lever: in 2022, PJM projected roughly 5,700 MW of growth by 2037 in its Dominion Zone, which includes northern Virginia's "Data Center Alley," but by 2025 that projection had ballooned to more than 20,000 MW from data centers alone, a 3.5x revision that pushed clearing prices to their ceiling.
Virginia Is Ground Zero
Northern Virginia is where the cost dynamics become extreme: according to Bloomberg data cited by EESI, Virginia electricity prices have increased by up to 267% over the last five years, far exceeding any other state, while nationally, prices held flat at roughly 13 cents per kilowatt-hour for more than a decade before 2019 and only climbed to 19 cents per kWh by late 2025. Residential prices rose 11.5% in 2025 alone, outpacing inflation by a factor of three. ICF modeling projects prices could increase by up to 40% by 2030, and that assumes no additional facilities break ground beyond those already permitted.
New supply cannot keep pace: PJM's 2026/2027 auction saw 2,669 MW of new generation or uprates, the first net additions in four auctions, but utilities received interconnection requests for more than 700 GW of power connection development in 2025. All of the United States consumed 477 GW in 2023. Wait times for gas-fired turbines now reach up to seven years, and refurbishing a 40-year-old coal plant to delay retirement can cost up to $1.3 billion. Data centers can be built in 18 months. The power plants to serve them take 6-7 years. Existing ratepayers bridge the gap.
The Strongest Case for Data Centers
But the counterargument deserves its full weight. Data centers generate substantial economic value: Virginia data centers produced an estimated $1.2 billion in local tax revenue in 2024, they create high-paying jobs, and they anchor the infrastructure that powers AI research and cloud computing.
The capacity market is also working exactly as designed: higher prices attract investment, PJM's latest auction cleared 2,669 MW of new generation as a direct response to the price signal, and if prices stay elevated long enough, new supply will eventually arrive and moderate costs.
And the 63% attribution may overstate data centers' role. Monitoring Analytics' counterfactual estimates what prices would have been without data center load in the forecast, but coal plant retirements, post-COVID demand recovery, and FERC rule changes all contributed to tighter supply-demand conditions independently. That 63% figure is modeled, not measured.
Limitations
Several caveats apply. Capacity costs are only one component of electricity bills; per-household calculations use PJM-wide averages while actual impacts vary by state, utility, and rate class; and industrial and commercial customers bear a proportional share, meaning the residential impact per household may be lower than the simple average suggests. PJM's own data center load forecasts may be inflated: IEEFA notes strong reasons to believe the 20-year growth projections will not fully materialize, meaning the market may be pricing in a worst case that never arrives.
What You Can Do
If you live in a PJM state (New Jersey, Pennsylvania, Delaware, Maryland, Virginia, West Virginia, Ohio, Kentucky, Indiana, Illinois, Michigan, North Carolina, or D.C.), check your utility's most recent rate case filing with your state's public utility commission. Capacity costs are itemized separately from energy charges. Compare the line item to what it was two years ago.
Several states are already pushing back: Maryland's Governor Wes Moore has demanded that data centers cover grid upgrade costs rather than socializing them across all ratepayers, and Pennsylvania's PUC is developing a model tariff to separate data center load costs. The policy direction is "bring-your-own-generation," which would mandate data centers secure dedicated power rather than drawing from the shared grid. If your state hasn't adopted this framework, your public utility commission is the body to contact.
The Bottom Line
Companies building AI have committed more than $300 billion in capital expenditure on data center infrastructure for 2025-2026 alone. Electricity for those facilities is being procured through a market mechanism that distributes cost increases across every ratepayer in the region, regardless of whether they use or benefit from AI services. In the PJM region, that cost is currently running at roughly $282 per household per year, an amount that no one voted for, no legislature approved, and no utility explicitly disclosed on anyone's bill. Markets work. Whether the people paying for it know that they are is a different question entirely.