US Power Demand Set for Records in 2026-27 as Data Centers Grow

Two more record years in the forecast

The US Energy Information Administration expects American electricity use to set new records in 2026 and again in 2027. Its October Short-Term Energy Outlook names AI-focused data centers and broader electrification as the main reasons.23 The agency puts consumption at 4,288 billion kilowatt-hours this year and 4,356 billion in 2027, up from the record 4,195 billion set in 2025.21 Data centers serving artificial intelligence and cryptocurrency mining account for much of the increase. Homes and businesses are also switching from fossil fuels to electricity for heating and transportation.22

The sector numbers matter as much as the totals. The EIA projects 2026 retail sales of 1,541 billion kWh to residential customers, 1,549 billion to commercial customers and 1,055 billion to industrial users.24 The first two would top the records both sectors set in 2025, which were 1,515 billion and 1,493 billion. The industrial record of 1,064 billion kWh, set in 2000, would stand.24 Data centers are counted in the commercial sector, and the agency expects commercial demand to keep rising through 2027, up 2.8% from the previous year.1

The forecast as a whole describes a grid where computing load now adds steadily to a base that was flat for years. The effects show up most clearly in a few regions, not in the national average.

A forecast that keeps getting revised

The most useful thing about this outlook is how it compares with earlier editions. Since spring, the EIA's monthly reports have moved its numbers in ways that show how hard this demand is to pin down.

The May outlook projected 4,248 billion kWh for 2026 and 4,379 billion for 2027.5 In July, the 2026 figure rose to 4,269 billion and the 2027 figure jumped to 4,399 billion.25 September's outlook put 2026 at 4,270 billion but cut 2027 to 4,349 billion.27 October raised 2026 again to 4,288 billion and set 2027 at 4,356 billion.23

The forecasts disagree in a consistent way. The near-term number has climbed every month, while the 2027 number peaked in July and has come down since. By this outlet's arithmetic, the October figures mean growth of about 2.2% this year and about 1.6% next year. The July outlook had predicted growth of about 3.0% in 2027.10

The forecasts also disagree on when commercial demand overtakes residential. In May, the EIA expected the commercial sector to pass residential for the first time in 2027.5 By July it had moved that date to 2026, and it predicted residential sales would actually fall that year.25 October's numbers still have commercial slightly ahead in 2026, but only by about 8 billion kWh, because a hot summer pushed household use up.23 The EIA reported that third-quarter consumption rose 4% from a year earlier, with residential up 6% and commercial up 5%, mostly because of cooling demand.1

Our reading is that much of the 2026 increase comes from weather on top of steady structural growth. The smaller 2027 forecast suggests forecasters are getting more careful about how quickly new data center load will actually connect.

Why the large-load numbers are unreliable

Other reporting supports that caution. In Texas, the grid operator ERCOT has received about 474 gigawatts of requests for new demand, more than five times the state's record peak of roughly 91 GW. Governor Greg Abbott's office says about 90% of those requests come from data centers.18 Abbott ordered state regulators and ERCOT to audit data center projects before connecting them. ERCOT paused approvals for sites of 75 MW or more.18

There is reason to be skeptical of the request totals. According to NERC, data center sites that entered service in ERCOT between 2022 and 2024 drew on average only 49.8% of the power they had requested at peak.15 PJM's independent market monitor told federal regulators that 4,017 MW of forecast data center load had been canceled or delayed. That is 42% of the growth PJM had assumed.18

These gaps between requested and actual load do not mean the boom is exaggerated. They mean the size and timing of the demand are uncertain. In September, the EIA said Texas's West South Central region still accounts for the largest share of sales growth in its forecast, despite the pause on new data center connections.27

PJM is where the strain shows

The national wholesale price forecast hides large regional differences. The EIA expects wholesale power to average $52 per megawatt-hour in 2026, 11% above 2025. It attributes the increase mainly to severe weather, including winter storms and a hotter-than-normal summer.21 Most hubs have been fairly stable. But the agency expects PJM, which covers the Mid-Atlantic and parts of the Midwest, to see a 41% price increase, while prices at the Pacific Northwest's Mid-Columbia hub fall 23%.21 The EIA also notes that wholesale prices are only one part of household bills. Residential rates also include transmission and distribution upgrades and, in some regions, capacity auction costs.1

On that point, the EIA's weather explanation differs from the analysis done inside PJM. The EIA highlights weather as the main reason wholesale prices rose. PJM's own market monitor puts more of the blame on data centers. It estimates that including data center load in the last four capacity auctions raised costs by $29.4 billion.19 Capacity prices went from $28.92 per megawatt-day for 2024/25 to the $325 cap for 2028/29.19 PJM's own analysis linked 63% of the jump to the 2026/27 price to data center growth.16

The two explanations can both be true. Weather pushes up energy prices in a given year, while data center growth raises the longer-term capacity costs that end up in bills over several years. For ratepayers, the second effect is likely to matter more over time. In Maryland, the state's consumer advocate said BGE's summer supply rate rose from about 10.2 to 12.5 cents per kWh, largely because of capacity market effects tied to data centers.19

Physical limits are tightening too. PJM recorded unrestricted peak demand of 168,158 MW on July 2. That broke a record set in 2006, during a heat wave in which the Department of Energy authorized PJM to switch large loads of 50 MW or more to their own backup generators.17 ERCOT set its own record of 91.1 GW on July 22, 6.5% above its previous peak.13

The fight over who pays is unresolved

Regulators are struggling to decide how costs should be shared. PJM proposed a framework under which new large loads of 50 MW or more without their own qualifying capacity would be the first to be cut off in grid emergencies. It also launched a backstop capacity auction with up to $20 billion to spend, aimed at a 6.8 GW shortfall.16 Federal regulators rejected most of the large-load plan in an October 9 order. That leaves open the question of whether households or data center developers will pay for the extra capacity.19

Meanwhile, developers are increasingly building their own power. Enverus forecasts that data centers will account for 22.5 GW of behind-the-meter demand, meaning power generated on site rather than drawn from the grid. PJM and ERCOT together make up nearly half of that. Meeting it would require about 29.6 GW of new gas-fired generation.14 Moving load off the grid could ease bills for other customers, but it also increases reliance on gas.

Gas holds steady while renewables gain

The EIA expects renewables to rise from about 24% of US generation in 2025 to 25% in 2026 and 27% in 2027. Coal falls from 17% to 15% over the same period. Natural gas holds at 40% this year and slips to 39% in 2027, and nuclear stays at 18%.24 Gas burned for power is forecast at 37.0 billion cubic feet per day in 2026, above the record of 36.8 bcfd set in 2024.23 So even with renewables growing, gas generation is still rising in absolute terms, because the overall amount of electricity keeps growing.

The bottom line

The main headline is solid: the US is on track for record electricity use, and data centers are a key reason. What the forecast revisions show is that 2026's increase owes a lot to weather, while the outlook for 2027 has become more cautious since summer. The pressure is concentrated in particular regions rather than spread across the country. Where it is concentrated, especially in PJM, the bigger problem is not running short of electricity. It is a pricing and cost-allocation fight that federal regulators have just sent back unresolved. Bills in those regions are likely to depend more on how that fight ends than on the national demand totals.