⚡ Energy
Your Power Bill Is $344 Higher Because of AI Data Centers. PJM's Own Watchdog Says the Damage Is 'Not Reversible.'
Monitoring Analytics, the independent watchdog for the grid serving 67 million Americans, finds data centers drove $3.8 billion of a $16.25 billion wholesale cost increase in five months. An original per-household cost model shows the data center premium now exceeds 20% of the average residential electricity bill in 13 states.
$344. That is approximately what each household served by America's largest power grid paid, annualized, because data centers exist on that same grid, not because those households used more electricity or because natural gas prices spiked, but because AI servers in Northern Virginia, in suburban Ohio, in exurban New Jersey needed power, and there was not enough to go around, so the price of the power that was available went up for everyone.
That number comes from the grid's own books, and the grid's own watchdog says the damage is permanent.
PJM Interconnection runs the wholesale electricity market for 67 million people across all or parts of 13 states and the District of Columbia: Delaware, Illinois, Indiana, Kentucky, Maryland, Michigan, New Jersey, North Carolina, Ohio, Pennsylvania, Tennessee, Virginia, and West Virginia. It is, by every measure that matters, the backbone of the eastern U.S. power system. And its 2026 Quarterly State of the Market Report, produced by the independent market monitor Monitoring Analytics, contains a sentence that should worry anyone who pays an electricity bill in those states: "The price impacts on customers have been very large and are not reversible."
The $3.8 Billion in Five Months
Start with the topline, which tells the story in two numbers: total wholesale power costs across PJM hit $40 billion in the first five months of 2026, up 68% from $23.8 billion in the same period of 2025, a year-over-year increase of $16.25 billion that came from several sources including rising natural gas prices, transmission congestion, retiring coal plants, and increased electrification. But Monitoring Analytics broke out the data center share: of that $16.25 billion increase, data centers accounted for $3.8 billion, fully 23% of the total.
Not 23% of total costs, but 23% of the increase, a distinction that matters because it means that even if every other cost driver had held steady, data centers alone would have pushed wholesale costs up by $3.8 billion in five months, and annualizing that figure gives you $9.12 billion per year.
Now divide by households: PJM serves 67 million people, the U.S. Census Bureau's average household size is 2.53 people, giving you approximately 26.5 million households, and spreading $9.12 billion across them produces $344 per household per year — money that flows not to the power plants that generate the electrons but to the market mechanisms that ration them.
How much is that relative to what people actually pay? EIA puts the average annual residential electricity bill in PJM states at roughly $1,660, meaning data centers are adding a premium of approximately 20.7% on top of that baseline, and the grid operator's own watchdog says it is locked in through at least May 2028.
The Numbers Behind the Numbers
The Q1 2026 State of the Market report offers a granular decomposition of how wholesale prices moved. PJM's real-time load-weighted average locational marginal price hit $87.57 per megawatt-hour in the first quarter, up 67.8% from $52.20 in Q1 2025, an increase of $35.37 per MWh.
Where did that $35.37 come from? The report decomposes it into five components, and the second-largest one points directly at data centers.
| Component | $/MWh | Share of Increase |
|---|---|---|
| Fuel & consumables | $14.92 | 42.2% |
| Transmission constraint penalties | $9.73 | 27.5% |
| Market power | $3.56 | 10.1% |
| Emissions costs | $1.26 | 3.6% |
| Scarcity | $0.85 | 2.4% |
Fuel costs are the largest single factor, because gas prices rose and PJM dispatches gas-fired plants on the margin, but the second-largest component — transmission constraint penalties at $9.73 per MWh — is directly tied to congestion, and congestion is driven by where load is growing fastest, which in PJM means Northern Virginia, home to the world's largest concentration of data centers, where a single building can draw 100 megawatts from lines designed to serve a county.
Total congestion costs across PJM quadrupled, rising from $503.3 million in Q1 2025 to $2.015 billion in Q1 2026, a 300.4% increase, while uplift payments (the costs of keeping reserve power plants spinning as a backup) more than doubled to $979.7 million. These are costs that every customer on the grid shares, because PJM socializes them across its entire footprint, meaning a retired teacher in rural Ohio is paying for the transmission constraints created by a Meta data center campus in Loudoun County, Virginia.
The Capacity Auction Tells the Real Story
Wholesale energy prices capture what happened today, but capacity auctions capture what will happen over the next few years, because they are the mechanism through which PJM secures commitments from power plants to be available during peak demand periods, and the capacity auctions are where the data center impact is most alarming.
According to Monitoring Analytics' annual report, inclusion of existing and forecast data center load growth in PJM's last three capacity auctions resulted in a combined total increase in system costs of $23.1 billion, which, spread over three delivery years, comes to $7.7 billion per year, or $291 per PJM household per year, from capacity charges alone, a separate pathway from the energy market costs described above, meaning the two stack on top of each other.
Joseph Bowring, the Independent Market Monitor, was blunt: the capacity market auctions for the 2025/2026, 2026/2027, and 2027/2028 delivery years "were not competitive, primarily as a result of forecast demand for data centers," a statement that carries considerable weight when you consider that this is a federally appointed market watchdog whose explicit job is to ensure the market for keeping the lights on functions fairly. Capacity costs in 2025 rose 262% year-over-year, jumping from 6.5% to 16% of total PJM wholesale costs, and in absolute terms, total PJM wholesale power costs hit $67 billion for the full year of 2025, up 54% from $43.5 billion in 2024.
The Shortfall Is Growing 31-Fold
The reserve margin is the buffer between how much power a grid can generate and how much it actually needs. Think of it as the grid's emergency cushion. In PJM's 2026/2027 capacity auction, the shortfall between cleared capacity and the target reserve margin was approximately 210 megawatts, tight but manageable.
In the 2027/2028 auction, the shortfall ballooned to 6,520 megawatts, a 31-fold increase in a single auction cycle, equivalent to roughly six large nuclear power plants or enough generating capacity to power about 5 million homes, and PJM does not have it because the power plants do not exist yet, cannot be built fast enough, and cannot be sited where the grid needs them most because of the same permitting and interconnection backlogs that have plagued U.S. energy infrastructure for decades.
Meanwhile, data center facilities can be built or expanded at double or triple the pace of new electricity generation systems, and that asymmetry is the core of the problem: load is arriving faster than supply, every new data center opens a faucet, and nobody has figured out a bigger drain.
What It Looks Like in Your Mailbox
Wholesale costs do not translate directly to retail bills, because utilities buy power through a mix of long-term contracts, spot purchases, and regulated rate cases, and regulatory lag can delay pass-through by months or years, but the direction is unambiguous and the CPI confirms it.
Bureau of Labor Statistics CPI data shows residential electricity prices are up 4% year-over-year as of mid-2026. That is roughly double the rate of overall inflation. Barclays economist Pooja Sriram told CNN that residential electricity prices rose "about twice as fast in 2025 as compared to the average seen in years prior," and that through the first five months of 2026, they climbed even faster than in 2025. "I think that is one of the clearest imprints of AI data center demand driving up residential electricity costs," she said.
The EIA's Short-Term Energy Outlook forecasts average annual residential electricity prices will rise 5.1% in 2026 and 2.4% in 2027, with Texas facing demand growth of 10% to 15% over 2025–2027 as the largest data center expansion outside of Virginia reshapes its grid, and the burden is not distributed evenly: Virginia ratepayers are subsidizing the cooling systems of servers that serve users in California, and Ohio households are paying for the grid upgrades that keep Amazon Web Services running in Columbus.
On July 2, during a heat dome that pushed PJM demand to 166.2 gigawatts, approaching the all-time record of 165.6 GW set in 2006, spot prices in PJM's Virginia zone, the heart of data center country, surged past $600 per megawatt-hour, a fifteen-fold spike from the $40 per MWh they had been that morning, which in practical terms means a factory in Richmond whose energy contract was pegged to spot pricing saw its hourly power cost jump from $4,000 to $60,000 while the servers next door hummed along on fixed-rate contracts that insulated them from the chaos they helped create.
The Strongest Case for Data Centers
The strongest counterargument is fiscal. Northern Virginia's "Data Center Alley" generates more than $3 billion in annual property tax revenue for Loudoun and Prince William counties, according to county budget documents. Those tax receipts have enabled Loudoun County to maintain one of the lowest residential property tax rates in the Washington metro area while funding some of the best public schools in the state. Construction crews, electricians, network engineers, security staff all have real jobs that depend on these facilities. Meta, Amazon, Google, and Microsoft have collectively invested tens of billions in Virginia alone.
A full accounting must include both columns. If data centers add $344 per household to the annual power bill but reduce property taxes by $500, the net impact is positive. But the electricity cost increase is regressive. It hits every household on the grid, including millions in West Virginia and Kentucky who see none of the tax revenue benefits, while the property tax reduction is hyperlocal, concentrated in a handful of wealthy Northern Virginia suburbs.
There is also an efficiency argument: newer data centers are dramatically more energy-efficient than older ones, and the industry is moving aggressively toward on-site renewable generation, nuclear power purchase agreements, and advanced cooling technologies. Google's 2025 environmental report showed its data centers achieved a power usage effectiveness of 1.10, meaning only 10% of electricity goes to cooling and overhead. That is impressive. It does not change the fact that a 1-gigawatt data center campus still consumes 1 gigawatt, and 1 gigawatt is enough to power 750,000 homes.
Limitations
This analysis relies on Monitoring Analytics' attribution of $3.8 billion in wholesale cost increases to data centers. Their internal methodology accounts for load growth forecasts, capacity auction results, and congestion patterns, but a precise isolation of data center impacts from other concurrent load growth (electric vehicle charging, building electrification, industrial reshoring) is inherently imperfect. Understand the $344 figure as an order-of-magnitude estimate, not a precise line item on anyone's bill.
PJM is one of seven major regional transmission organizations in the United States. Its 67 million people represent about 20% of the U.S. population. Other grids, including ERCOT in Texas, CAISO in California, and MISO in the Midwest, face their own data center dynamics, but the cost structures and regulatory frameworks differ significantly. A national figure would require separate analysis of each market.
Wholesale costs do not pass through to retail bills in a one-to-one ratio. Regulated utilities file rate cases; deregulated markets adjust faster. Our $344 estimate represents the wholesale cost increase attributable to data centers, not the amount that has appeared on retail bills to date. Some of it has. Some of it will. Some may be absorbed by utilities and recovered later. That trajectory is clear even if the exact timing varies by state and utility.
The Bottom Line
Every ChatGPT query, every AI-generated image, every automated customer service response runs on electricity that comes from somewhere. For 67 million Americans in PJM's footprint, that somewhere is the same grid that powers their refrigerators, their air conditioners, and their children's night lights. And the grid's own independent watchdog, whose job is to ensure the market works fairly, says the price impact is large and irreversible.
What to do about it depends on where you sit.
If you are a PJM ratepayer: Check whether your utility has filed or plans to file a rate case that includes data center load growth. Many utilities build these costs into general rate increases without itemizing them. Virginia's State Corporation Commission and Pennsylvania's Public Utility Commission have both opened proceedings on data center cost allocation, and public comment periods are open. Your electricity bill has a line item that says "generation" or "supply." That is where this shows up.
If you are a state policymaker in PJM territory: The $23.1 billion in capacity auction cost increases is already locked in through May 2028. At stake is whether data center operators should pay a share of grid upgrade costs proportional to their load contribution, or whether all customers should continue to split the bill evenly. Oregon and Georgia have both passed legislation requiring data centers to pay for their own transmission interconnection. PJM states have not.
If you are a data center operator: On-site generation is no longer a sustainability talking point; it is a cost-of-doing-business question. Every megawatt you pull from the grid costs your neighbors $344 a year more, and regulators are noticing. The companies that build their own power, whether nuclear SMRs, on-site solar-plus-storage, or behind-the-meter gas turbines, will face fewer political headwinds than those who keep drawing from the communal bathtub.
Nobody designed the grid for this. It was designed for a world where demand grew 1% a year and the biggest load on the system was an aluminum smelter in West Virginia. That world ended when someone decided to train a language model in Ashburn, Virginia, and the power bill for training it landed, in pieces too small to notice individually but too large to ignore collectively, on 26.5 million doorsteps.