The Energy Information Administration now expects commercial power sales to reach 1,557 billion kilowatt-hours in 2026, edging past residential demand at 1,546 billion and marking a structural inversion in how the American grid is used (Reuters, 2026). Total consumption climbs to 4,306 billion kWh this year and 4,370 billion kWh in 2027, the third consecutive annual record.

The shift is narrow but consequential. Residential sales set their own all-time high last year at 1,519 billion kWh, and they continue growing. Commercial demand simply grows faster, driven by data centres that add hundreds of megawatts in single locations on timelines that outpace utility planning cycles (Economic Times Data Centers, 2026). A 1-gigawatt campus in Abilene, Texas illustrates the problem: transmission upgrades, new generation, and regional infrastructure planning all have to be sequenced around a single customer's energisation date.

Supply is not keeping pace in the way the headline numbers suggest. Natural gas holds at 40 percent of generation in 2026 before easing to 39 percent in 2027, coal slides to 15 percent, and renewables climb to 28 percent by 2027. Nuclear stays flat at 18 percent. Gas sales to power generators reach 37.4 billion cubic feet per day, a record (Reuters, 2026). The firm capacity doing the heavy lifting remains fossil-fired, even as corporate buyers advertise renewable procurement.

The cost pressure is already visible in wholesale markets. PJM's capacity auction for the 2026/2027 delivery year cleared at record levels, with data centre load growth identified as the primary driver of an 833 percent increase over the prior year (KilowattLogic, 2026).

BuiltWorld AI Operational Take: The commercial-over-residential crossover is a planning signal, not a statistical curiosity. Utilities that built forecasting models around household growth now face single customers whose demand arrives in 500-megawatt increments, often before the transmission to serve them exists. Project teams should treat local substation capacity and queue position as the binding constraint on commissioning dates, because the aggregate national numbers hide exactly where the grid is thinnest.