Analysis
Data centers had actually been making US electricity cheaper for years by absorbing fixed grid costs across more usage, according to a Fortune analysis published Sunday -- but the report warns that a $7 trillion AI buildout with no guaranteed matching demand now threatens to reverse that trend entirely. The dynamic matters because it reframes the AI infrastructure story away from pure capex numbers and toward a genuine household and small-business cost question.
The strain is already showing up in real market data: PJM Interconnection, the largest US electric grid operator serving 13 states and Washington, DC, ran a capacity auction for the year starting June 2028 that tied a $16.4 billion record, with data centers alone accounting for roughly $6.3 billion of that total. Separately, power-hungry data centers have pushed supply costs for that same grid up more than 60%, according to the system's own watchdog.
“Separately, power-hungry data centers have pushed supply costs for that same grid up more than 60%, according to the system's own watchdog.”
The scale of projected future demand is what makes the risk structural rather than temporary: data centers alone could add roughly 125 gigawatts of US electric load between now and 2030, pushing overall electricity demand growth to a 4.1% compound annual rate, while total data center power consumption is projected to climb from 176 terawatt-hours in 2023 to somewhere between 325 and 580 terawatt-hours by 2028 -- a wide range that itself reflects how uncertain the actual AI demand curve remains.
The Trump administration has responded with a voluntary pledge effort, rallying utilities and data center developers to commit that AI-driven demand growth won't be passed on to household and business electricity bills, following a summer heatwave that already strained grid capacity nationwide.
What to watch: whether the White House's voluntary utility pledge holds up as actual enforceable commitments rather than a public-relations gesture, and whether the wide 325-580 terawatt-hour demand range narrows as more AI infrastructure projects either proceed as planned or get scaled back amid financing and circular-demand concerns.
