US utilities face 100+ GW capacity gap as AI data centers surge

The United States faces a widening electricity supply gap as AI-driven data center growth accelerates faster than utilities can add generation capacity, according to a new Bank of America analysis.

The firm projects the country will need more than 230 GW of new generating capacity over the next five years, but regulated utilities are expected to add only about 93 GW of accredited supply—leaving a shortfall exceeding 100 GW through 2030.

Data centers alone could add roughly 125 GW of electric load during this period, pushing overall electricity demand growth to a 4.1% compound annual growth rate from 2026 through 2030. This marks a sharp reversal from a decade of flat consumption driven by efficiency improvements and distributed solar adoption.

Why it matters

This supply-demand mismatch will reshape how data center operators and utilities approach power procurement and infrastructure investment. The gap threatens to constrain AI deployment in the United States and could drive up electricity costs for all customers as utilities scramble to preserve reliability. The analysis suggests the market is shifting from demand-constrained to delivery-constrained, fundamentally altering the economics of large load interconnection.

Behind-the-meter generation gains traction

With large gas turbines largely sold out through 2030, data center developers are increasingly turning to on-site generation. More than 7.5 GW of data center projects with behind-the-meter generation are already under construction, with another 60+ GW in pre-construction stages, according to the Bank of America report.

These facilities are expected to combine self-generation with traditional grid connections rather than operating entirely off-grid, improving reliability while shortening project timelines. Natural gas reciprocating engines are gaining favor because they can be deployed more quickly than large turbines and respond rapidly to changing loads. Manufacturers including Caterpillar, INNIO, Rolls-Royce, and Wärtsilä have expanded production to meet rising demand.

Utilities extend coal, deploy batteries

Utilities and regulators are keeping existing generating assets online longer to preserve reliability. The analysis identifies coal plants across Maryland, Wisconsin, Indiana, Utah, Kansas, Nebraska, and Mississippi that have had retirement dates delayed or canceled to maintain dispatchable capacity.

Battery storage and transmission expansion could help address reliability challenges, though the analysts caution that transmission projects often require years to permit and build. The Champlain Hudson Power Express, which took 16 years from planning to energization, illustrates the development timelines facing new infrastructure.

Equipment constraints limit options

Large gas turbines remain the preferred technology for flexible power generation, but manufacturing capacity is largely committed through 2030. New units can take years to enter service after shipment, creating a bottleneck for utilities attempting to add firm capacity quickly.

The analysts note that planned generation additions may overstate available supply because intermittent resources like wind and solar contribute less accredited capacity during peak demand than their nameplate ratings suggest. Firm resources will remain essential even as renewable generation expands.

Higher power prices could drive some demand reduction, but academic research suggests electricity demand is relatively inelastic in the short to medium term. A 10% increase in real electricity prices typically results in only a 1% to 2% decline in consumption.

"The market is no longer constrained by demand—it is constrained by where power can actually be delivered," the Bank of America analysts wrote in their Global Research Report.

The forecast comes largely from BofA's semiconductor analysts, whose projections for rapid AI computing infrastructure deployment underpin the firm's outlook. The analysts note that utilities have revised demand forecasts upward in each of the past three years as AI-related electricity demand has materialized faster than expected.

These details were first reported by Utility Dive.