US Utilities Face Widening Gap Between AI Power Requests and Committed Load

American utilities are discovering that the scale of AI infrastructure announcements and the scale of actual, committed electricity demand are two very different numbers — and the distance between them is growing. The mismatch is rewriting how grid operators assess load credibility, who funds new infrastructure, and whether ordinary ratepayers end up holding the bill if projected demand stalls.

What happened

American Electric Power in Ohio received upward of 30 GW in preliminary data center load inquiries. After the company introduced stricter financial requirements through its Data Center Tariff process, roughly 13 GW advanced to formal study — and only 5.6 GW moved forward with signed service agreements backed by financial commitments. That funnel illustrates the broader planning problem: utilities are now managing three distinct tiers of demand credibility, ranging from informal requests through to fully energized facilities, and the gaps between each tier are widening.

ERCOT disclosed that average peak consumption for certain proposed large-load projects reached only about half of the megawatt levels those projects originally requested, and the Texas grid operator said it may revise demand forecasts using historical project completion rates rather than announced figures. The pattern echoes earlier renewable interconnection queues, where speculative filings routinely outpaced developer capacity to secure financing, permits, and equipment.

To manage the uncertainty, Duke Energy created dedicated procurement subsidiaries to lock in transformers and other electrical equipment through specialized financing structures, while also extending a $10 billion credit facility and raising its five-year capital plan to $103 billion. AEP similarly lifted its five-year capital plan to $78 billion and projects 63 GW of incremental contracted load by 2030. Wood Mackenzie estimates the US data center electrical equipment market could grow from around $20 billion today to $65 billion by 2030.

Why it matters

The reliability stakes are no longer theoretical. The North American Electric Reliability Corporation issued a Level 3 alert this month warning that utilities broadly lack adequate processes to handle the dynamic behavior of large computational loads, including rapid voltage swings and sudden load drops that traditional industrial planning frameworks were never designed to absorb. NERC is developing a new regulatory category specifically for large facilities connected to the bulk power system.

At the same time, the question of who pays for potentially overbuilt infrastructure is moving to the center of regulatory debates. AEP has proposed requiring large new data centers to commit to paying for at least 90 percent of requested capacity for ten years before supporting infrastructure would be constructed. In Texas, a coalition of retail electricity providers noted the stark disparity in how transmission costs are allocated between large commercial customers and residential accounts. The Maryland Office of People's Counsel has filed a complaint against PJM transmission planning decisions, arguing that consumers could face significant costs if anticipated data center load growth fails to arrive on schedule.

Federal regulators at FERC have also opened proceedings examining whether hyperscale customers should directly fund grid upgrades attributable to their projects, particularly after co-location proposals inside the PJM footprint raised priority-access concerns.

What to watch

ERCOT's proposed batch-study framework — which would group projects meeting stricter criteria into coordinated transmission studies — could become a model for other grid operators seeking to separate credible demand from placeholder filings. Transmission buildout remains a structural bottleneck: independent analysis suggests the US needs roughly 5,000 miles of new high-capacity lines annually through 2035, against fewer than 1,000 miles completed in 2024. Whether utilities can accelerate construction without ultimately socializing the cost of infrastructure built around demand that hasn't fully committed remains the central unresolved question across multiple regulatory jurisdictions.