How Can AI Data Centers Grow When Power Is Short and Residents Say No?
Dozens of AI data centers are planned across Pennsylvania. But the power plants and transmission lines to supply them are not being built at the same speed. PPL Electric, the utility that serves central and eastern Pennsylvania, says data center load in advanced planning alone totals 31.8 gigawatts. PJM, the grid operator for 13 states and Washington, D.C., projects peak demand growing 3.6 percent a year, and 94 percent of that growth comes from data centers.
A data center draws power at a steady rate around the clock, so one gigawatt of data center demand uses far more electricity over a year than one gigawatt of household peak demand. The Congressional Research Service defines a hyperscale data center as a facility using more than 100 megawatts, roughly the power used by about 80,000 American homes. PJM manages the grid for 67 million people.
How many data centers are planned in Pennsylvania?
The count depends on how you count them. Cleanview, a market intelligence firm that tracks data centers from public records, counted 303 in Pennsylvania as of November 16. Of those, 17 are operating, 12 are under construction and 274 are planned. That total covers every kind, not only AI. Other counts are lower. An independent dashboard puts proposed, under construction and preliminary projects at about 70. The Pennsylvania Department of Environmental Protection said in May that it was reviewing permits for 17 AI data center projects and knew of 34 other proposals. Either way, the number is in the dozens. Cleanview says nine campuses are planned at one gigawatt or more.
Pennsylvania has reasons to want them. Amazon announced in June 2025 that it would invest at least $20 billion to build data center campuses in Luzerne County and Bucks County, creating at least 1,250 jobs. The state called it the largest private-sector investment in Pennsylvania history.
Where does the power come from?
The power comes from three directions. First, tech companies are taking electricity from nuclear plants that already exist. Amazon has a contract to buy most of the power from the Susquehanna nuclear plant in Luzerne County, ramping to 1,920 megawatts by 2032. Because that is existing generation rather than new generation, less power is left on the grid, and gas plants fill the gap.
Second, closed plants are being revived or new ones built. Unit 1 at Three Mile Island, which closed in 2019, is being restarted with a target of 2027 after Microsoft signed a 20-year contract to buy all 835 megawatts. That is Unit 1, next to the Unit 2 reactor that had the 1979 accident. In Indiana County, the Homer City site, where a two-gigawatt coal plant closed in 2023, is being rebuilt as a 4.5-gigawatt gas plant, which would be the largest in the United States, to power an Amazon data center campus. First power is targeted for 2027. Earthjustice, an environmental group, estimates the gas plant would emit 10.5 million tons of carbon dioxide a year, 2.4 times the old coal plant’s 2022 level, because its capacity is more than twice as large.
A two-gigawatt campus has broken ground at the former coal plant site in Shippingport, Beaver County, and a campus with three gigawatts of on-site gas generation is planned in Upper Burrell, outside Pittsburgh. Google and Constellation have a $4.3 billion deal to add 890 megawatts through nuclear plant upgrades. Third comes solar and storage. Solar is the largest share of what is waiting to connect to Pennsylvania’s grid, but it still supplies only about 1 percent of the state’s power, so it is not a large share yet.
More projects are bringing their own power. Earthjustice estimates that about 80 percent of self-powered data centers coming online in the next five years will run on gas. The point is that the power is not here yet. The planned demand is enormous, and most of the generation to meet it will not arrive until 2027 or later. The last capacity auction came up 6,831 megawatts short of what was needed.
What powers Pennsylvania today?
The U.S. Energy Information Administration reports that Pennsylvania generated about 247 terawatt-hours of electricity in 2025 and sent 89 terawatt-hours to other states, making it the largest net exporter of electricity in the country. Natural gas supplied 58 percent, nuclear 30 percent, coal 7 percent, wind 1.4 percent, hydropower 1.1 percent, solar 0.8 percent and biomass 0.6 percent. Gas and nuclear together supply more than 88 percent, and renewables supply about 4 percent. Compared with 2018, gas rose from 35.5 percent to 58 percent, nuclear fell from 38.8 percent to 30 percent and coal fell from 17.1 percent to 7.1 percent. Coal shrank and gas took its place. Renewables actually slipped a little, from 4.8 percent in 2018 to 3.8 percent in 2025.
One point is easy to get wrong. Being the largest exporter of electricity does not mean Pennsylvania has power to spare at peak times. Exports are measured over a full year, while the capacity auction buys power for the single hour of the year when demand is highest. A state can have power left over across the year and still run short in that one hour. Data centers use power steadily all day, so they raise both the total and the peak. That is how annual exports and a peak shortage can happen at the same time.
The high auction price and the difficulty of connecting data centers have the same root. Demand is growing faster than power plants and transmission lines can be built. The auction is where the cost of that gap is billed to households.
Who should pay?
The market monitor and PJM disagree. Jon Gordon, a senior director at Advanced Energy United, an industry group for clean energy companies, said the region is getting closer and closer to a reliability crisis and an affordability crisis while solutions come too slowly. Jeffrey Shields, a PJM spokesman, said PJM fundamentally disagrees that it can decide who bears data center costs. That decision belongs to state utility regulators, he said, though PJM has built pathways for states to assign all data center costs to those customers.
There have been attempts. The Trump administration and PJM governors agreed in February on principles to shield households and small businesses from data center costs, and major tech companies signed a Ratepayer Protection Pledge to obtain their own power. The monitor said PJM failed to carry them out.
Is Pennsylvania the best place for a data center?
Pennsylvania’s advantage is not surplus power but the conditions to build new power. Gas in the ground, nuclear sites, former factory sites with transmission lines already in place, and a state government eager to attract developers all come together. An industry report says Pennsylvania’s commercial electricity rates are the fourth lowest among PJM states and its data center market is the fastest growing in the region. That is why developers are coming.
Even so, connecting to the grid is the bottleneck. PPL Electric alone has 31.8 gigawatts of data center requests in advanced stages. The Pennsylvania Public Utility Commission set a model large-load tariff in May for customers above 50 megawatts, requiring collateral, minimum bills and upfront payment for grid upgrades. Governor Josh Shapiro issued an executive order in August requiring large AI data center developers to meet separate standards. Residents are pushing back too. A Quinnipiac University poll in July found 74 percent of Pennsylvania registered voters would oppose building an AI data center in their community, up from 68 percent in February. More than 20 proposals have been withdrawn, delayed or rejected after local opposition. Residents worry not only about electric bills but also about the water used for cooling, noise and home values.
How do other states compare?
The national average is gas 41 percent, nuclear 18 percent, coal 17 percent, wind 11 percent, solar 7 percent and hydropower 6 percent. Renewables total 24 percent. Pennsylvania relies far more on gas and nuclear than the national average and far less on coal and renewables. Virginia is the closest match, with gas at 56.5 percent, nuclear at 26.3 percent, solar at 8.5 percent and coal at 3.4 percent. Northern Virginia is home to the world’s largest cluster of data centers. But there is one big difference. Virginia generates less electricity than it uses and bought 41.2 terawatt-hours from outside in 2025. Pennsylvania exported 89 terawatt-hours.
New Jersey gets 48.5 percent of its power from gas and 45.7 percent from nuclear, with no coal at all. In July it enacted the Data Center Fair Share Act, which creates a separate rate class for large data centers and makes them cover their own grid infrastructure costs. Ohio gets 54.7 percent from gas, 24.3 percent from coal and 11.6 percent from nuclear, so its coal share is more than three times Pennsylvania’s. West Virginia depends on coal for 86.7 percent of its power, the highest in the country.
States farther away look different again. Texas leads with gas at 48 percent but also gets 17 percent from wind, 15 percent from solar, 12 percent from coal and 6 percent from nuclear. In California solar is the largest single source and there is no coal. Illinois gets about half its power from nuclear. Washington State gets more than 80 percent from low-carbon sources, led by hydropower, and Iowa gets 58.5 percent from wind. What matters is what each state uses to meet new demand. Pennsylvania is meeting it with gas, and Virginia is buying power from outside.
The price of buying power in advance is rising too
Electricity is hard to store, so it must be made when it is used. To prepare for the day of highest demand, power plants are kept on standby, and the capacity auction buys that standby power in advance. In plain terms, it is paying generators in advance to promise they will make electricity when it is needed later. The problem is that the reservation includes demand from data centers that do not exist yet.
The numbers are already showing up in bills. Capacity costs make up about one-fifth of wholesale electricity costs, and they rose more than 80 percent in the first nine months of 2026 compared with the same period in 2025. The market for electricity actually used was competitive over the same period, the monitor said, and consumers received power at the lowest stable prices, but the capacity market was not competitive. The auction held from late June to early July for 2028 and 2029 capacity totaled $16.4 billion. When the monitor reran the same auction without the forecast growth in data center load, the total was $10.7 billion. Forecast data center load of 16,436 megawatts added $5.7 billion to the cost. The monitor warned in its annual report on November 12 that the December auction is also likely to clear at the maximum price and add $5 billion to $10 billion in data center costs to bills across the region.
The forecast itself is already shifting. As of September 21, the monitor said, 42 percent of the forecast increase in data center load for the 2026 to 2029 delivery years had already been cancelled or delayed.
Governor Josh Shapiro sued PJM and reached a settlement that capped auction prices, starting at $325 per megawatt-day. Later auctions kept clearing at the cap, and in July the auction fell 6,831 megawatts short of the standby capacity target for peak demand. The Federal Energy Regulatory Commission also rejected PJM’s proposed solution, a framework that included a reliability backstop auction to connect data center developers directly with power plant owners. The commission sent it back, saying it did not include everything needed to protect consumers. PJM submitted a revised proposal. That means the backstop auction cannot be held before December, so data center demand stays in the auction forecast and keeps pushing prices up. The monitor recommended postponing the December auction until the backstop auction can be held. The December auction, which buys capacity for 2029 and 2030, is scheduled for December 9 to 15.
Whether the monitor’s warning is right, that ordinary consumers are paying money data centers should pay, will be shown by the December auction. The fundamental question is one. Who will make the power the AI era needs, with what, and with whose money.

