Electricity companies across the U.S. are struggling to figure out how much demand will actually materialize from the artificial intelligence boom, as the stock market speculates that vast sums of money will be spent on infrastructure to support a big data center buildout.
“There is a question about whether or not all of the projections, if they’re real,” Willie Phillips, who served as chairman of the Federal Energy Regulatory Commission from 2023 until April 2025, told CNBC. “There are some regions who have projected huge increases, and they have readjusted those back.”
The AI companies are rolling out ambitious plans to build server farms that in some cases would consume as much electricity as entire cities. But the tech industry is shopping the same big projects around to multiple utilities as they look for the quickest access to power.
“We’re starting to see similar projects that look exactly to have the same footprint being requested in different regions across the country,” GridUnity CEO Brian Fitzsimons told CNBC. GridUnity uses software to give utilities and transmission operators a clearer picture of where power projects are requesting connections across the patchwork U.S. electric grid.
This data center shopping is making it difficult for utilities to determine how much power generation they will need to ensure the reliability of the electric grid. Electricity prices, meanwhile, are rising for consumers because power supply is already struggling to keep pace with demand.
FERC Chairman David Rosner warned in September that the difference of a few percentage points in electricity load forecasts “can impact billions of dollars in investments and customer bills.”
“Put simply, we cannot efficiently plan the electric generation and transmission needed to serve new customers if we don’t forecast how much energy they will need as accurately as possible,” Rosner said.
Constellation Energy CEO Joe Dominguez warned of the problem on the nuclear power operator’s May earnings call: “I just have to tell you, folks, I think the load is being overstated. We need to pump the brakes here.”
AI bubble fears
The stock market, however, is not really pumping the breaks. The steady pace of big data center announcements is fueling one of the biggest rallies in power company stocks in two decades.
The utility sector has gained about 21% this year after rallying more than 19% in 2024. The companies that supply the U.S. with electricity have gained nearly $500 billion in value over that two-year period. The last time utilities advanced more than 40% in consecutive years was in 2003 and 2004.
OpenAI CEO Sam Altman warned in August that the stock market is facing an AI bubble, cautioning investors that they were getting “overexcited.”
While the exact magnitude of the coming demand is uncertain, experts generally agree that the U.S. is facing a historic increase in electricity consumption after a long period of flat growth. Existing data centers point to what is coming, Rob Gramlich, president of Grid Strategies, said.
“We can see it. Data centers exist,” he said. “They’re operating day in and day out, using a lot of electricity. It used to be the case a 50 megawatt data center was pretty big. Now, it’s very common to have data centers that are 20 times that size — that are a gigawatt.”
Grid Strategies, a power sector consulting firm, estimates 120 gigawatts of additional electricity demand by 2030. This includes 60 gigawatts from data centers based on forecasts from the utilities. To put that in perspective, 60 gigawatts is roughly equivalent to the 2024 peak hourly power demand of Italy, the world’s eighth largest economy.
“This is not a bubble,” Fitzsimons said. “It’s going to transform our nation completely. It’s going to continue to grow. We need a 50-year energy policy.”
But the utilities need to get firm financial commitments from the data centers, Gramlich said.
“That is going to help us rationalize all these requests and get a better handle on the total estimate,” he said. “But the industry’s got to plan based on the best information we have at the moment.”
The uncertainty about demand forecasts has raised concerns that utilities could spend billions of dollars on infrastructure that is not needed in the end. Fitzsimons said the risk of the utilities overbuilding is lower than it was two decades ago due to constraints in the market.
“They’re in a very different environment where we have massive supply chain problems,” Fitzsimons said. “We have inflation running off the hook for quite a long time now. They can’t afford overbuild. It’s going to come down to better planning.”
Infrastructure constraints
A month after warning of a bubble, Altman struck a deal with Nvidia in which OpenAI would build 10 gigawatts of data centers with the chipmaker’s graphic processing units. The plan requires as as much electricity as New York City during the energy intensive summer days, raising questions about whether they can actually secure the power needed.
It may be possible to secure that much electricity but the AI industry is facing contraints as their plans grow larger, Gramlich said. The companies are competing for scarce infrastructure, he said, which is increasing prices for essential electrical equipment like transformers, switches and breakers.
“We really don’t have the electrical infrastructure to meet the aggressive targets,” the analyst said. “We don’t have enough generation or transmission infrastructure to meet even the modest midpoint targets.”
The question is how fast new generation can be built, he said. Natural gas turbines are largely sold out through the end of the decade. The tech industry is investing in advanced nuclear power, but these technologies are not expected to reach commercial scale until the 2030s at the earliest.
Renewable energy, meanwhile, can be deployed at the quickiest speed, particularly solar and battery storage. More than 90% of the power projects waiting for a grid connection right now are solar, battery storage or wind, according to August data from energy consulting firm Enverus.
“For the past 10 years, our interconnection cues have been filled with a massive percentage of renewables,” Fitzsimons said. “The renewables are the fastest way to build out new capacity. There’s no doubt in that because of the supply chain issues we have around natural gas turbines.”
But President Donald Trump opposes solar and wind while promoting coal, natural gas and nuclear power, raising uncertainty about whether enough new generation can be built. Utilities will turn data centers away if they do not have enough power available, Gramlich said.
“If they literally do not have the power to serve a customer, they’re not going to sacrifice reliability,” Gramlich said. “That’s their core job.”
Some AI companies believe they have found a solution — just build the power at the data center off the grid, called “behind the meter” in industry lingo.
“We should invest in just about every possible way of generating energy,” Nvidia CEO Jensen Huang told CNBC in an interview on Oct. 8. “Data center self-generated power could move a lot faster than putting it on the grid and we have to do that,” he said.
