Data centers that haven’t been built yet are driving up electricity prices and could leave consumers on the hook for expensive power infrastructure if demand projections are wrong.

The race to build facilities that provide artificial intelligence has fueled a boom in data centers that train and run large language models, like OpenAI’s ChatGPT and Anthropic’s Claude, upending a utility industry that grew used to 20 years of no increase in electricity demand.

But now, some investors and energy market analysts are questioning whether the AI race has turned into a bubble, one that would prove expensive to unravel as new transmission lines and power plants are built to support those data centers.

Consumers served by the largest electric grid in the U.S. will pay $16.6 billion to secure future power supplies just to meet demand from data centers from 2025 through 2027, according to a watchdog report published this month.

The grid is PJM Interconnection, serving more than 65 million people across 13 states, including the world’s largest data center hub in Virginia and fast-growing markets like northern Illinois and Ohio.

About 90% of that bill, or $15 billion, is to pay for future data center demand, according to Monitoring Analytics, PJM’s independent market monitor. This amounts to a “massive wealth transfer” from consumers to the data center industry, the watchdog told PJM in a Nov. 10 letter.

“A lot of us are very concerned that we are paying money today for a data center tomorrow,” said Abe Silverman, general counsel for the public utility board in New Jersey, one of the states served by PJM, from 2019 until 2023. “That’s a little bit scary if you don’t really have faith in the load forecast.”

Residential electricity prices in September rose 20% in Illinois, 12% in Ohio, and 9% in Virginia compared to the same period last year, according to data from the federal Energy Information Administration. Each of those states are among the top five markets for data centers in the U.S.

The costs associated with securing power for data centers is directly reflected in consumer’s utility bills, said Joe Bowring, president of Monitoring Analytics. “When the wholesale power costs go up, people pay more, when it goes down people pay less,” he said.

Forecast uncertainty

PJM is forecasting 30 gigawatts of extra demand from data centers through 2030, but it’s unclear how much will actually materialize in the end. That’s the equivalent of the average annual power consumption of more than 24 million homes in the U.S.

Data center developers are shopping projects around in different locations before committing to a site, so there is likely duplication in the forecasts, said Cathy Kunkel, a consultant at the Institute for Energy Economics and Financial Analysis (IEEFA).

“We’re in a bit of a bubble,” Silverman, the New Jersey official, said. “There is no question that data center developers are coming out of the woodwork, putting in massive numbers of new requests. It’s impossible to say exactly how many of them are speculative versus real.”

Independent power producers such as Constellation Energy, the biggest owner of nuclear plants in the U.S., and Vistra Corp. warned earlier this year that data center demand forecasts are likely inflated.

“I just have to tell you, folks, I think the load is being overstated. We need to pump the brakes here,” Constellation CEO Joe Dominguez said on the company’s earnings call in May.

Meanwhile, Vistra CEO James Burke also said in May that data center demand could be overstated by three to five times in some jurisdictions as developers scout their projects around the country.

‘Stranded cost’

The risk is that utilities invest in expensive infrastructure to meet data center demand, but not all those facilities are eventually built or they end up using less electricity than expected, said Kunkel, the consultant.

“It does tend to be consumers — residential, commercial, and other industrial ratepayers — that end up paying for overbuilt electrical infrastructure,” Kunkel said. The potential problem will come if capacity is built that isn’t needed, that “would tend to leave ratepayers holding the stranded cost bag.”

Data center demand forecasts have declined when utilities implement stricter rules.

In Ohio, for example, American Electric Power recently had requests for 30 gigawatts of electric connections from data centers.

AEP proposed stricter rules “to mitigate the risk that transmission infrastructure will be built for speculative data center projects,” according to a filing with the state utility commission in May 2024.

The AEP rules require data centers to pay for 85% of the energy they claim to need, even if they actually use less, to cover infrastructure costs. It also implemented an exit fee if data centers cancel their project or can’t meet the terms of their contract.

AEP’s data center requests in Ohio dropped by more than half, to 13 gigawatts after the utility commission approved the rules last July.

“When faced with potential financial commitments, the most speculative or uncertain data center projects did not submit load study requests — as was intended,” the Columbus, Ohio-based utility said in a statement.

The number of requests might decline further as the new rules force data centers to make binding contracts, it said.

The Data Center Coalition, a lobbying group for big tech companies, and other industry advocates have opposed AEP’s stricter rules as “discriminatory.”

Meeting demand

There is also a risk that the electrical grid grows less reliable as many large data center projects move forward. The 13 gigawatts of data center requests that AEP views as a more accurate figure, for example, is equivalent to about a dozen large nuclear plants. The infrastructure, in power plants and transmission lines, required to meet that demand is immense, the utility said.

The solution is for PJM to reject data centers’ requests for grid connection if there is not enough power to supply them, Bowring of Monitoring Analytics said. Data centers can either wait until there is enough power to supply them, or they can bring their own generation with them and jump the line, he said.

Monitoring Analytics filed a complaint with the Federal Energy Regulatory Commission last week calling on PJM to adopt this approach.

“That will give data centers a clear incentive to bring their [own] generation,” Bowring said. That formula would also help clear up uncertainty over demand forecasts because data centers are unlikely to pay for infrastructure if they are not serious, he said.

Otherwise, the costs that consumers are bearing from data center demand will continue to grow, the watchdog warned FERC in its complaint.