PJM Capacity Auction Adds Billions to Power Bills Across 13 States
Data center load is widening the gap between accredited supply and projected demand, with ratepayers already absorbing the cost.
PJM's capacity auction results, announced the week of 2026-07-13, will add billions of dollars in electricity costs across 13 states, with data center demand the main driver of the increase.7 The bill lands on ratepayers in the world's largest competitive power market, and the supply gap underpinning it is getting wider, not narrower.
Bank of America analysts forecast the United States 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 — a shortfall exceeding 100 GW.5 As utilities struggle to bring capacity online fast enough, BofA expects more data center developers to turn to behind-the-meter generation.5
The demand projection has institutional backing. Lawrence Berkeley National Laboratory, in a report backed by the Department of Energy, put data center electricity use at 9.5% to 15.3% of all US power by 2030, up from 4.7% in 2024.4 That is an upward revision from researchers' earlier forecast of a 6.7% to 12% range by 2028.4 The IEA reported in April that electricity demand from data centers jumped 17% year on year in 2025, with AI-hosting facilities rising faster.2
Ratepayers are already absorbing earlier rounds of this cost. Investor-owned utilities requested $18 billion in rate increases last year — the most since the mid-1980s — and regulators approved 66% of the dollar value sought.4 Electricity prices rose 6% in nominal terms last year, more than twice the rate of inflation.4
The Trump administration has pushed self-generation as the answer, with the White House promising that tech companies building power plants alongside data center campuses will shield ratepayers from rising bills.3 Developers are moving in that direction. More than 7.5 GW of data center projects with on-site generation are under construction, with another 60 GW-plus in pre-construction, according to BofA.5 Manufacturers including Caterpillar, INNIO, Rolls-Royce and Wärtsilä have expanded production to meet the demand.5
For gas traders, the arithmetic is less dramatic than the headline figures suggest. Even a 4-6 GW data center consumes roughly 1 billion cubic feet per day of natural gas, depending on turbine efficiency — modest against total US production.1 NYMEX Henry Hub front-month held at $2.91/MMBtu early on 2026-09-17, unchanged on the session, and the market is not pricing a data center-driven gas squeeze. The larger gas market effect is structural timing: on-site plants can be built in two to three years, while utility-scale gas, transmission and interconnection take longer, which is why developers are choosing behind-the-meter in the first place.5
The supply side of BofA's forecast may itself be generous. Planned additions include intermittent wind and solar that contribute less accredited capacity during peak demand than their nameplate ratings imply, so the 93 GW figure could overstate what is actually deliverable.5 BloombergNEF's most recent forecast for data center power demand is nearly double what the firm projected in December, and more than its analysts expect the grid to accommodate.6 Off-grid gas is often presented as the fix. Turbine supply chains and fuel logistics cannot scale to cover the entire shortfall.6
Demand-side tools could ease near-term pressure. Columbia University researchers argued in commentary published on 2026-06-23 that grid-enhancing technologies and expanded demand response can reduce the price impact of data center load.4 Their deeper argument concerns utility incentives: under traditional regulation, utilities earn roughly 9% to 10% on capital investments, a structure that "rewards capital deployment more than system optimization."4 Replacing existing transmission lines with advanced conductors could produce $180 billion in savings by 2050, according to analysis in the Proceedings of the National Academy of Sciences, which the Columbia team referenced.4
Gas markets are pricing near-term weather, storage and LNG flows, not a mid-2030s capacity gap that may be partially resolved by siting decisions, load curtailment or regulatory reform. The Columbia work notes that some AI operators are already exploring surge pricing for inference, which would flatten peak demand rather than compound it.1
The pace at which the 60 GW-plus of pre-construction on-site projects reaches financial close is the number to track. If a meaningful share converts, utility load forecasts across the 13 PJM states face downward revision and the ratepayer cost story shifts toward a utility earnings story. If the pipeline stalls, the next PJM auction prices the gap again.5