AI & EnergySubscribers only Jun 28, 2026 at 11:183Add to bookmarks

For the first time, a leader of a major American utility publicly questions the grid's ability to absorb the demand shock from AI data centers—and predicts outages.
The accelerated deployment of AI data centers is putting unprecedented pressure on the U.S. power grid. As Congress debates a moratorium on new constructions (AI Data Center Moratorium Act, AOC/Sanders, June 25, 2026), Exelon’s CEO—one of the largest U.S. utility groups—has issued a public warning: the United States faces a real risk of blackouts due to a structural electricity supply deficit.
The IEA (Electricity 2025) projects ~945 TWh of global data center consumption by 2026, with ~380 TWh for the U.S. alone. GE Vernova, manufacturer of gas turbines for hyperscale campuses, reports a multi-year order backlog with delivery times of 3 to 5 years. In this context of dual software and physical pressure on AI, OpenAI was forced to limit GPT-5.6 at the request of the U.S. government (June 26, 2026)—the first direct regulatory precedent for a deployed model.
The bottleneck is twofold and structural. On the generation side: gas plants take 4 to 6 years to build, nuclear SMRs 8 to 12 years—far beyond AI deployment cycles. On the transmission side: UHV lines are stalled by 5- to 10-year environmental and land-use approvals. Hyperscalers are signing multi-decade PPAs and campus leases committing tens of additional GWs before capacity even exists. The grid, designed for the stable demand of the 1990s, cannot absorb exponential growth without massive modernization.
Utilities focused on grid modernization (NextEra Energy, Constellation Energy) structurally benefit from demand shock, but their valuation premiums are already high. Network equipment manufacturers—transformers, UHV cables (ABB, Prysmian, Nexans)—remain chronically under capacity with intact pricing power. Regulatory escalation would weigh on hyperscalers dependent on construction permits.
Exelon’s CEO’s warning may be tactical (anti-moratorium lobbying)—to be cross-checked with NERC reports. Battery storage is advancing faster than official projections: the blackout risk may be overestimated over a 3–5 year horizon. An economic slowdown would mechanically reduce data center demand.
Create a free account to access all our content and the weekly review.
Article produced by artificial intelligence, reviewed under human editorial control.
Sign in to join the discussion.
Funny how the same grid that’s ‘too fragile’ for AI data centers handled Bitcoin mining just fine. Priorities, or just panic?
2022 EIA data shows US grid reserve margin at 19.3%-well above NERC’s 15% threshold. Alarmism sells, math doesn’t.
Data centers are the new steel mills-demand spikes always outpace grid upgrades. Where’s the federal coordination, or are we just repeating the 2000-01 California crisis?
Regulation of AI Data Centers: Legislative Risk and Energy Constraints