05-08-2025
‘Unlike Anything We've Seen': The Energy Industry is Counting on the AI Boom
'Load growth has been flat, basically flat, for 20 years or more in the U.S.,' said Tom Wilson, a grid expert at the Electric Power Research Institute, a think tank. 'And so the idea of load going up allows you, at the very highest level, to spread any sort of system-wide cost that should be allocated to the various players over a larger number of kilowatt-hours.'
The broader uncertainty across the energy policy landscape helps explain why utilities are clinging to tech so fiercely now. And while tech may not have stuck its neck out particularly far on behalf of renewable tax credits, it's still going to be the power sector's best customer.
The electric power industry's relative dispassion may also give it a valuable role in the continuing partisan battles over solar and wind power's reliability, as my colleague Nico Portuondo reported last month. When Senate Energy and Natural Resources Committee Chair Mike Lee (R-Utah) dinged renewables at a hearing on electricity demand, Jeff Tench, executive vice president at Vantage Data Centers, offered a mild corrective.
'Our observation and our requirement is for more electrons, and Vantage is relatively agnostic as to the source of those electrons,' he said.
So far, data centers have only increased total U.S. power demand by a tiny amount (they make up roughly 4.4 percent of electricity use, which rose 2 percent overall last year). But the two industries' fates are already linked. When Chinese firm DeepSeek unveiled an AI model in January that it billed as 10 to 40 times cheaper and more efficient than U.S. models like ChatGPT, the stock of tech giants like NVIDIA and Oracle plummeted — as did that of power providers like Constellation, Vistra and GE Vernova.
There are risks in a hidebound, tightly regulated industry like power, which is essentially physical in nature, hitching its wagon to mercurial tech.
'There is a scenario where utilities benefit from this,' said Michael Wara, director of Stanford University's climate and energy policy program. 'And there's also a scenario where they overplay their hand dramatically.'
There are several ways utilities could do that. One is taking demand estimates at face value and overbuilding. The tech industry is famous for its ability to improve its efficiency — and, simultaneously, for its tendency to overstate the energy use of new widgets. Computing history is littered with laughable-in-retrospect claims, like the one about a Palm Pilot using as much electricity as a refrigerator.
'Nobody has any idea how much demand is going to be from AI in five years, and anyone who says that they know that is lying,' said Jonathan Koomey, a researcher who's devoted decades to debunking demand projections and coined Koomey's Law, which holds that computing energy efficiency doubles every 18 months.
There are solutions, though. In Virginia, for example, where data centers make up a quarter of demand and are projected to quadruple again by 2040, Dominion Energy is requiring data centers commit to buying fixed amounts of power for 14 years, to protect against unexpected efficiencies.
