Utilities across the U.S. are struggling to gauge how much demand the artificial intelligence boom will actually bring to fruition, as stock markets predict billions of dollars will be spent on infrastructure to support the construction of large data centers.
Willie Phillips, who served as chairman of the Federal Energy Regulatory Commission from 2023 to April 2025, told CNBC that “there is some question as to whether all the forecasts are real.” “There are some regions that are projecting significant increases, so we’ve readjusted those.”
AI companies are rolling out ambitious plans to build server farms that, in some cases, will consume as much power as an entire city. But the tech industry, seeking the fastest access to power, orders the same large projects from multiple power companies.
GridUnity CEO Brian Fitzsimons told CNBC that “we’re starting to see requests for similar projects in different regions around the country that appear to have exactly the same footprint.” GridUnity uses software to give utilities and transmission operators a clearer picture of where power projects require connections across the patchwork U.S. power grid.
This data center purchase makes it difficult for power companies to determine how much power they need to generate to ensure grid reliability. Meanwhile, power prices are rising for consumers as power supply is already struggling to keep up with demand.
FERC Chairman David Rosner warned in September that a difference of just a few percentage points in power load forecasts “could impact billions of dollars in investments and customer bills.”
“Simply put, if we don’t predict as accurately as possible how much energy our customers will need, we can’t efficiently plan the generation and transmission needed to serve new customers,” Rosner said.
constellation energy Chief Executive Officer Joe Dominguez warned of the problem during the nuclear power producer’s May earnings call, saying, “I have to say this, folks, I think the burden is overstated. We need to pump the brakes here.”
Fear of AI bubble
But the stock market isn’t actually applying the brakes. A steady pace of large-scale data center announcements is fueling the biggest rally in power company stocks in 20 years.
The utilities sector is up about 21% this year after rising more than 19% in 2024. Companies that provide electricity to the United States have gained nearly $500 billion in value over the past two years. The last time utilities saw consecutive increases of more than 40% was in 2003 and 2004.
In August, OpenAI CEO Sam Altman warned that the stock market was facing an AI bubble and warned investors against getting “overexcited.”
Although the exact scale of future demand is unknown, experts generally agree that the United States is facing historic increases in electricity consumption after a long period of flat growth. Existing data centers are a sign of things to come, said Rob Gramlich, president of Grid Strategies.
“I understand that. Data centers exist,” he said. “They’re running day in and day out, and they’re using a lot of power. It used to be that 50 megawatt data centers were pretty big. Now it’s very common to have data centers that are 20 times that size, or gigawatts.”
Grid Strategies, a power sector consulting firm, predicts 120 gigawatts of additional electricity demand by 2030. That includes 60 gigawatts from data centers based on the utility’s projections. To put that in perspective, 60 gigawatts roughly corresponds to the 2024 peak-hour electricity demand of Italy, the world’s eighth-largest economy.
“This is not a bubble,” Fitzsimmons said. “This country is going to change completely. It’s going to continue to grow. We need a 50-year energy policy.”
But utilities need solid financial commitments from data centers, Gramlich said.
“This helps us streamline all these requests and get a better picture of the total estimate,” he said. “However, the industry needs to base its plans on the best information available at this time.”
Uncertainty in demand forecasts is raising concerns that utilities could end up spending billions of dollars on infrastructure they don’t need. Utilities spent $178 billion upgrading the power grid last year and are expected to spend $1.1 trillion in capital spending by 2029, according to the Edison Electric Institute.
Fitzsimmons said the risk of utility overbuilding is lower than it was 20 years ago due to market constraints.
“They’re in a completely different environment where we have massive supply chain issues,” Fitzsimmons said. “We’ve had inflation for quite some time. We can’t afford to overbuild. Eventually we’re going to need better planning.”
Infrastructure constraints
A month after the bubble warning, Altman signed a contract with another company. Nvidia OpenAI will build a 10 gigawatt data center with the chipmaker’s graphics processing units. The plan would require as much electricity as New York City during the energy-intensive summer months, raising questions about whether the city would actually be able to secure the electricity it needs.
While it may be possible to secure that much power, Gramlich said the AI industry faces constraints as plans grow larger. He said the two companies are competing for scarce infrastructure, driving up the price of essential electrical equipment such as transformers, switches and breakers.
“We don’t really have the power infrastructure to meet our demanding targets,” the analyst said. “We don’t have enough power generation or transmission infrastructure to meet modest interim goals.”
The question is how quickly a new generation can be built, he said. Natural gas turbines will be mostly sold out by 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.
On the other hand, renewable energy, especially solar power and battery storage, is the fastest to deploy. More than 90% of power projects currently waiting to connect to the grid include solar, battery storage and wind power, according to August data from energy consulting firm Enbels.
“For the past decade, our interconnection chain has been largely filled with renewable energy,” Fitzsimmons said. “Renewable energy is the fastest way to build new power generation capacity. There’s no question about that, given the supply chain issues we’re seeing with natural gas turbines.”
But President Donald Trump has promoted coal, natural gas and nuclear power while opposing solar and wind power, raising uncertainty about whether enough new power will be built. If there isn’t enough power available, the utility company will deny access to the data center, Gramlich said.
“We’re not going to sacrifice reliability even if we literally don’t have the power to serve our customers,” Gramlich said. “That’s their core job.”
Some AI companies believe they have found a solution. Simply build power into your off-grid data center, which is known in industry parlance as “behind the meter.”
“We should be investing in almost every conceivable way to generate energy,” Nvidia CEO Jensen Huang said in an Oct. 8 interview with CNBC. “We can move to generating electricity in our data centers much faster than putting it on the grid, and we have to do that.”
