AI Data Centers Push U.S. Power Demand to Fresh Records Through 2027

U.S. electricity consumption is headed for new highs in 2026 and 2027. The gains come faster than many expected just a few years ago. Data centers hungry for artificial intelligence workloads lead the charge.

According to the Energy Information Administration’s latest Short-Term Energy Outlook, total power use will climb from a record 4,195 billion kilowatt-hours in 2025 to 4,271 billion kWh in 2026 and 4,397 billion kWh in 2027. Reuters first highlighted the projections in June. Similar figures appear in coverage by Investing.com.

That acceleration marks a sharp break from the prior decade. Electricity demand barely budged between the mid-2000s and 2019. Now it grows at an average 1.7% annually from 2020 through the end of the forecast period. Data centers explain much of the difference.

Commercial Demand Overtakes Residential Use for the First Time

Commercial electricity sales will surpass residential sales in 2026. The EIA sees commercial volumes at roughly 1,547 billion to 1,550 billion kWh next year against 1,508 billion to 1,512 billion kWh for homes. Industrial sales add another 1,065 billion to 1,066 billion kWh. The crossover reflects data centers moving from niche facilities to dominant loads on the commercial side of the meter.

Servers alone already accounted for an estimated 7% of all commercial sector electricity in 2025. By 2050 that share climbs to between 22% and 33% across scenarios in the agency’s Annual Energy Outlook 2026, released in April. Server consumption could reach as high as 818 billion kWh by midcentury in the high-demand case. Standalone data centers drive much of the increase. EIA’s Today in Energy analysis lays out the long-term math.

Recent data sharpen the picture. Lawrence Berkeley National Laboratory, in a report released through the Department of Energy, put 2023 data center consumption at 176 terawatt-hours. That equals 4.4% of total U.S. electricity. The lab projects 325 to 580 TWh by 2028, or 6.7% to 12% of national supply. Energy.gov summarized those findings in December 2024.

Goldman Sachs Research goes further on near-term power draw. The bank forecasts U.S. data center demand will more than double to 66 gigawatts in 2027 from 31 GW in 2025. Capacity additions accelerate to 36.3 GW that year after 13.6 GW in 2026. The analysis draws on granular facility data from Aterio, including permits, construction status and satellite imagery. Goldman Sachs published the note in May 2026.

But. Real-world bills already show strain. A brickmaker in Sugarcreek, Ohio, watched its electricity costs jump 90% last year. Capacity charges climbed from $1,600 to $12,000 a month. The surge traces to data center growth in the region. Reuters reported the story on July 7, 2026. Factories in the Rust Belt now absorb higher charges passed through shared grids.

And the numbers keep rising. U.S. data centers consumed nearly 313 TWh in 2025, according to the Energy Institute’s Statistical Review of World Energy. That figure exceeds total generation in several large European countries. America now claims almost 40% of global data center electricity use. Recent LinkedIn commentary from energy executives amplified the review’s findings.

Projections vary. The International Energy Agency sees U.S. data center consumption rising 130% from 2024 levels by 2030, adding about 240 TWh. BloombergNEF estimates power demand from the facilities could hit 106 GW by 2035. Electric Power Research Institute models put the share at 9% of national electricity by 2030 in some scenarios, with Virginia facing 41% to 59% locally. E&E News covered the EPRI outlook in February 2026.

Cooling adds another layer. Data center space cooling can require 2.9 times the electricity of ordinary commercial floor space. In the high-demand case, that extra cooling alone consumes 84 billion kWh more by 2050. Servers themselves run with a flat load shape. They pull power at roughly the same rate around the clock. The EIA updated its commercial demand model in the 2026 outlook to track server electricity separately from other computing uses.

Generation mix shifts in response. Coal’s share of U.S. power falls from 17% in 2025 to 15% by 2027. Natural gas holds steady near 40%. Renewables climb from 24% to 27%. Nuclear stays at 18%. The pattern echoes longer-term trends. Yet faster load growth raises questions about whether new gas plants, delayed renewables or even restarted coal units will fill gaps in certain regions.

Utilities scramble. Peak demand forecasts have grown sixfold in three years at some grid operators. Data centers account for the bulk of the revision. Interconnection queues swell. Only a small fraction of requests reach commercial operation. Hyperscalers respond by signing direct power purchase agreements, building behind-the-meter generation or even exploring nuclear restarts.

Prices feel the pressure. Higher demand tightens supply in key markets. Residential and industrial customers in affected areas see rates rise even if they never visit a data center. Virginia already attributes more than one-third of its electricity use to these facilities. Texas, with its own surge, could exceed 40 GW of data center capacity by 2028 and claim nearly 30% of national demand.

Efficiency gains help at the margin. New servers pack more computation per watt. Yet explosive growth in AI training and inference workloads outruns those improvements. One analysis found training certain large models now consumes 50 gigawatt-hours, enough to power San Francisco for three days. Model sizes keep scaling. So does the energy bill.

The EIA’s outlook stops at 2027. Longer views from its Annual Energy Outlook and outside forecasters point to continued expansion. Server electricity could multiply more than 16-fold from 2020 levels by 2050 in aggressive cases. Total data center demand may triple or quadruple again in the 2030s if AI adoption follows current trajectories.

Policy makers watch closely. The Department of Energy has accelerated work on advanced cooling, high-efficiency chips and grid modernization. Congress has debated measures to map data center locations and consumption more transparently. Grid operators in PJM, MISO, ERCOT and elsewhere now treat data center forecasts as a core variable in long-term planning.

One fact stands clear. The AI boom no longer lives only in software conferences or venture decks. It sits inside steel-and-concrete buildings that draw hundreds of megawatts apiece, 24 hours a day. Those buildings reshape regional power markets. They influence fuel choices, transmission builds and consumer bills from Ohio factories to Virginia suburbs. And the surge has only begun.