This past winter, 400,000 Exelon customers nearly lost power on some of the coldest days of the year. Not in a hypothetical scenario. Not in a planning document. In real life, with real thermostats dropping toward dangerous temperatures.
“We came very close, this past winter, to having to curtail power for about 400,000 customers on some of the coldest days of the year. And it’s only getting worse,” Calvin Butler, Exelon’s president and CEO, warned in a Financial Times interview. His timeline for actual blackouts: 2027. That’s not a decade away.
The culprit isn’t aging coal plants or freak storms. It’s AI — specifically, the massive data centers now consuming electricity at a rate the grid was never designed to handle.
The AI Tax on the Electric Bill
Data center power demand is growing faster than the infrastructure built to support it.
U.S. data centers consumed 1.9% of national electricity in 2018. By 2023, that figure had jumped to 4.4%, according to Lawrence Berkeley National Laboratory data cited by Harvard’s Belfer Center. The Department of Energy projects the share could reach 9% by 2030. Deloitte estimates AI data center power demand alone could grow from 4 GW in 2024 to 123 GW by 2035 — more than a thirtyfold increase.
The physical reality is stark:
- AI-optimized server racks draw 30 to over 100 kW each, versus 5–15 kW for traditional racks
- A five-acre facility retrofitted for GPU-heavy AI workloads can jump from 5 MW to 50 MW — same building, ten times the power draw, per Deloitte
- Grid interconnection requests now face wait times stretching to seven years
- Wholesale electricity prices near major data center hubs have spiked over 260%, according to Enki AI’s market intelligence
- In 2024, a single voltage fluctuation in northern Virginia simultaneously disconnected 60 data centers, creating a 1,500 MW surplus requiring emergency grid intervention, per the Belfer Center
Think of it like one person microwaving lunch and tripping the breaker for the entire office floor — except the “microwave” is a billion-dollar AI campus and the “floor” is an entire regional grid.
Who Actually Pays When the Lights Go Out
The gap between AI’s appetite and the grid’s capacity lands hardest on ordinary consumers.
“We have increased demand like we’ve never seen in the last 30 to 40 years and we continue to play by the same rules,” Butler told CNBC. In PJM’s deregulated market, utilities don’t own generation — supply simply isn’t showing up fast enough to meet surging load.
Butler argues utilities should regain the ability to build generation assets — community solar, battery storage, flexible capacity — to close the gap. Some energy analysts counter that aggressive deployment of renewables and demand-side flexibility can manage the risk without restructuring market rules. The Belfer Center adds an important check: overbuilding for AI demand that never fully materializes could saddle consumers with stranded infrastructure costs.
Maryland already experienced a brief 2025 blackout affecting around 4,000 residents — thirty minutes, but a warning shot. The grid isn’t failing from bad luck. It’s running on rules written for a different era of demand, now asked to power something it was never designed for.
