Electric utilities, grid planners, and power-plant developers across the United States are scrambling to keep pace with data-center electricity consumption that has roughly doubled since 2023, reaching an estimated system impact of about 42 gigawatts. The surge, driven largely by artificial intelligence workloads and cloud computing expansion, is already reshaping how regional grid operators such as PJM Interconnection and the Electric Reliability Council of Texas plan for new generation capacity. With federal projections showing demand could triple again by 2028, the pressure on existing power infrastructure is intensifying faster than most forecasters anticipated just two years ago.
Why 42 gigawatts of data-center load changes the power equation
The scale of the problem comes into focus against a single baseline number: data centers consumed 176 terawatt-hours of electricity in 2023, equal to roughly 4.4 percent of all U.S. power use, according to a Department of Energy report drawing on research by Lawrence Berkeley National Laboratory. That share was already large enough to rank data centers among the country’s most power-hungry industrial sectors. A near-doubling to the 42-gigawatt level means these facilities now rival the total installed generating capacity of some mid-sized states.
The immediate tension sits with grid operators responsible for keeping the lights on. The Energy Information Administration has built data-center growth directly into its near-term load forecasts, relying on inputs from PJM and ERCOT, the two regional operators most exposed to new facility construction in Virginia, Texas, and the broader mid-Atlantic corridor. When those operators signal higher demand, the EIA’s Short-Term Energy Outlook adjusts its baseline for national electricity consumption upward, which in turn signals to generators and investors that new capacity is needed quickly.
Speed matters because power plants and transmission lines take years to permit, finance, and build. If the 42-gigawatt level holds through 2026, the gap between available supply and contracted demand will likely push at least one additional gigawatt of gas-fired generation into active development within PJM and ERCOT service territories. Interconnection queues in both regions already show a sharp rise in applications from gas-plant developers seeking to serve data-center clusters. The risk is that short-term demand locks in fossil-fuel infrastructure for decades, even as the country pursues broader clean-energy goals.
Data centers also concentrate demand in ways that strain local distribution networks. Unlike residential load, which ebbs and flows with daily routines and weather, hyperscale server farms can draw near-constant power around the clock. That flat profile is attractive to generators but challenging for utilities that must reinforce substations, feeders, and transformers to handle multi-hundred-megawatt customers connecting at once. In some fast-growing regions, new data centers are already being told they must wait years for sufficient grid capacity, delaying tech-sector investments or pushing projects into neighboring states.
Federal data and congressional analysis behind the doubling claim
Three independent federal sources anchor the core numbers. The Department of Energy’s assessment, built on data tracked by Lawrence Berkeley National Laboratory’s data-center research program, established the 176-terawatt-hour baseline for 2023 and projected that national data-center electricity use will climb to between 325 and 580 terawatt-hours by 2028. The wide range reflects uncertainty about how fast AI training clusters scale and whether efficiency gains in chip design and cooling can offset raw demand growth.
The EIA’s own analysis treats data centers as a primary driver of recent U.S. electricity demand growth and links that growth to potential price increases for all ratepayers. Its modeling framework incorporates STEO baselines alongside direct grid-operator submissions, meaning the agency’s published forecasts already assume significant data-center buildout. An EIA discussion on fossil generation warns explicitly that faster-than-expected data-center expansion could increase reliance on natural gas and coal plants, pushing wholesale electricity prices higher across affected regions.
A Congressional Research Service report, designated R48646, provides a legislative-grade cross-check. That CRS brief restates the 2023 national energy-use estimate and cites both Lawrence Berkeley National Laboratory and the Electric Power Research Institute as its underlying sources. The report also clarifies definitional boundaries, specifying what qualifies as data-center load and excluding certain categories such as cryptocurrency mining from some estimates. This distinction matters because inconsistent definitions have historically muddied comparisons between studies.
Together, these three sources create a consistent evidence chain: a verified 2023 baseline, a modeled current system impact near 42 gigawatts, and a projected demand range for 2028 that, at its upper bound, would require data centers to consume more than 14 percent of current U.S. electricity generation. No single private-sector forecast has produced numbers that contradict this federal consensus, though the spread between the low and high 2028 scenarios is wide enough to support very different infrastructure strategies.
Gaps in the evidence and what to watch through 2026
The 42-gigawatt figure itself carries an important caveat. No primary DOE or EIA dataset publishes a single current-year demand number for data centers in gigawatt terms. The figure is derived from the 2023 terawatt-hour consumption data and extrapolated using announced construction pipelines, utility filings, and grid-operator projections. The underlying hourly demand traces from Lawrence Berkeley National Laboratory that would confirm the doubling in real time have not been released publicly for 2024 or 2025. Until those datasets appear, the 42-gigawatt estimate should be treated as a reasonable but provisional system-impact measure rather than a measured point value.
Another gap is geographic granularity. Federal projections describe national totals, but the buildout is highly uneven. Northern Virginia’s “Data Center Alley,” the Dallas–Fort Worth metroplex, and parts of Ohio, Georgia, and Arizona are seeing much faster growth than the country as a whole. Without transparent, location-specific data on how much load is actually connected versus merely proposed, state regulators and local communities are left to infer impacts from interconnection queues and land-use filings, which can overstate ultimate buildout.
Efficiency trends add further uncertainty. Chipmakers and cloud providers argue that each new generation of hardware delivers more computations per watt, and that advanced cooling, smarter workload scheduling, and on-site renewables will blunt the growth in grid demand. The federal projections do account for efficiency improvements, but they cannot fully anticipate breakthroughs in processor design or software optimization. Conversely, if AI applications shift toward ever-larger models that run continuously, efficiency gains could be overwhelmed by sheer scale.
Policy choices between now and 2026 will heavily influence which side of the DOE demand range becomes reality. Faster permitting and tax incentives for renewable generation and storage could allow new data-center clusters to be paired with low-carbon power, limiting the need for new gas plants. Stricter interconnection rules or capacity charges could encourage operators to invest in on-site batteries and demand-response capabilities, reducing peak strain on the grid. Conversely, if policy remains static while data-center construction accelerates, grid planners may default to familiar gas-fired solutions that can be financed and built quickly.
For utilities, the next two years will test planning assumptions that long treated electricity demand as essentially flat. Many integrated resource plans drafted before the AI boom are already obsolete, forcing mid-course corrections that add risk for both investors and ratepayers. Regulators are beginning to ask for scenario analyses that explicitly model high-growth data-center cases, including stress tests for extreme weather events coinciding with peak computing demand.
Communities hosting large data centers will also face trade-offs that go beyond abstract gigawatt figures. Local officials must weigh tax revenue, jobs, and digital infrastructure benefits against land use, noise, water consumption for cooling, and higher electricity prices that can accompany rapid load growth. Clearer federal data and more transparent modeling from utilities could help those communities make informed decisions instead of reacting piecemeal to each new project proposal.
Ultimately, the 42-gigawatt estimate is less a precise measurement than a flashing warning light. It signals that data centers, once a niche consideration in grid planning, are now central to the national electricity story. Whether that story leads to a more resilient, cleaner power system or to a new wave of long-lived fossil infrastructure will depend on how quickly planners, policymakers, and technology companies translate today’s projections into concrete decisions about what gets built, where, and on what terms.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
