American households, factories, and server farms will collectively push U.S. electricity consumption past back-to-back annual records between 2024 and 2026, with total demand climbing from 4,097 TWh to roughly 4,250 TWh. The single largest force behind that jump is the rapid expansion of data centers running artificial intelligence workloads. Federal energy analyses show data-center electricity use nearly doubling within five years, a pace that threatens to outstrip the grid’s ability to add clean generation and transmission capacity fast enough.
Data-center load and the 2028 demand cliff
The scale of the problem becomes clear in a single comparison. U.S. data centers consumed 176 TWh of electricity in 2023, roughly 4.4 percent of the national total, according to a Department of Energy analysis evaluating the increase in electricity demand from data centers. By 2028, the same DOE work projects that share will rise to between 6.7 and 12 percent of all U.S. electricity, translating to 325 to 580 TWh per year. At the upper bound, data centers alone would consume more electricity than the entire residential sector of several mid-sized states combined.
That range matters because the gap between the low and high scenarios is enormous. If demand lands near 325 TWh, utilities and grid operators have a manageable buildout challenge. If it reaches 580 TWh, the country would need to bring online the equivalent of dozens of large natural-gas or nuclear plants in under four years, on top of every wind and solar project already in development. The EIA has documented that U.S. electricity demand has been rising steadily since 2020, with data-center electricity use identified as a primary driver of that growth. The question is no longer whether demand is climbing but how steeply.
A testable way to track which scenario is unfolding is to watch how quickly fossil generation ramps. If data-center load trends toward the upper end of the DOE’s 2028 range, monthly EIA generation data should show natural-gas output rising at least 4 percent above the 2023 baseline by mid-2027, even as renewable capacity additions continue. Gas plants can be permitted and built faster than new transmission lines or large renewable installations, so they tend to absorb demand surges first. Observing that metric over the next 18 months will signal whether the grid is meeting AI-driven load mainly through fossil fuel generation or through accelerated clean-energy deployment.
Federal data linking AI expansion to record electricity use
Three separate federal and international analyses converge on the same finding. The DOE’s report, drawing on work by Lawrence Berkeley National Laboratory in its 2024 United States Data Center Energy Usage Report, quantifies historical data-center consumption and projects growth through 2028, with AI-driven expansion as a central variable. The EIA’s Today in Energy discussion confirms that data-center electricity use is driving recent demand growth and notes that higher-than-expected load scenarios could shift the generation mix toward fossil fuels because of shorter construction timelines for gas plants. And the International Energy Agency’s Electricity 2026 outlook reports that U.S. power demand is climbing in 2024 and 2025, growing faster than overall energy demand, a structural shift that reflects both electrification and computing-sector expansion.
The consistency across these sources is striking. Each uses different methodologies and data windows, yet all point to the same conclusion: data centers, and the AI workloads running inside them, are the dominant new source of electricity demand growth in the United States. The DOE’s own Office of Electricity has separately flagged the need for clean energy resources to meet data-center demand, citing estimates from the Electric Power Research Institute on data centers’ growing share of national electricity through 2030. While the precise contribution of AI versus more traditional cloud and enterprise computing remains uncertain, federal analysts increasingly treat AI acceleration as the swing factor that will determine whether overall data-center consumption lands near the low or high end of projected ranges.
Grid capacity gaps and unanswered generation-mix questions
Several critical questions remain open. The headline figures of 4,097 TWh for 2024 and roughly 4,250 TWh for 2026 draw on secondary modeling rather than a single primary DOE or EIA dataset that states those exact numbers. The underlying federal reports quantify data-center load specifically but stop short of an explicit statement that AI accounts for “most” of the incremental national demand through 2026. AI is clearly the fastest-growing component of data-center load, but other factors, including manufacturing reshoring, electric vehicle charging, and building electrification, also contribute to the broader demand increase.
The generation-mix question is equally unresolved. EIA analysis discusses how faster-than-expected load growth could lift fossil generation shares, but no publicly available state-level or utility-level primary records detail exactly how many new gas-plant filings are directly tied to data-center contracts. Without that granularity, it is difficult to assess whether the grid is heading toward a temporary fossil-fuel bridge or a longer-term dependence on gas generation that would complicate federal and state climate targets. The same ambiguity applies to nuclear and long-duration storage: both are often cited as solutions for 24/7 AI power needs, but permitting timelines and cost uncertainties make it hard to know how much of the 2028 demand cliff they can realistically cover.
Transmission constraints compound the challenge. Even where wind and solar projects are technically capable of serving new data centers, congested interconnection queues and local siting opposition can delay delivery of that power by years. In practice, this often means that the fastest way to energize a large AI campus is to build or expand nearby gas-fired capacity, then layer in renewable power purchase agreements over time. That sequencing may keep the lights on, but it risks locking in emissions-intensive infrastructure that remains in operation long after cleaner alternatives are available.
Policy responses and planning blind spots
Policymakers are only beginning to grapple with the implications of this surge in digital load. At the federal level, the DOE and national laboratories are focusing on efficiency standards for servers and cooling systems, as well as incentives for locating data centers near abundant clean resources. Some state regulators are exploring requirements that new data centers procure a minimum share of their power from renewables or contribute to local grid-upgrade costs. Yet these measures are piecemeal, and they lag the speed at which AI developers and cloud providers are announcing new facilities.
Planning blind spots persist on both sides of the meter. Many utilities still base long-term resource plans on historical demand trends that underweight the possibility of rapid AI-related load additions. Conversely, some data-center developers have been slow to disclose firm power needs early in the siting process, leaving grid planners scrambling to accommodate multi-hundred-megawatt campuses on short timelines. Bridging this gap will require more transparent coordination, with utilities, regulators, and large customers sharing realistic scenarios rather than optimistic assumptions.
One emerging idea is to treat major AI data centers as quasi-industrial customers that must participate directly in demand management. That could include requirements for on-site storage, flexible computing schedules that shift non-urgent workloads to off-peak hours, or contractual obligations to curtail during system emergencies. Such measures would not eliminate the need for new generation and transmission, but they could soften peak impacts and reduce the risk that AI clusters trigger local reliability crises.
The stakes for the clean-energy transition
How the United States manages AI-driven electricity demand over the next five years will shape the trajectory of its broader energy transition. If data-center growth forces a prolonged reliance on gas-fired generation, the country could see emissions plateau or even rise just as other sectors begin to decarbonize. If, instead, policymakers and grid operators successfully channel this demand into accelerated renewable and storage deployment, AI could become a catalyst for modernizing and expanding the grid at unprecedented speed.
The difference between those outcomes hinges on choices being made now: how quickly interconnection backlogs are cleared, whether utilities update their planning assumptions to reflect AI’s upper-bound scenarios, and how firmly regulators insist that new digital infrastructure align with climate objectives. The federal analyses already on the table do not offer certainty about the exact scale of AI’s power needs, but they do provide a clear warning. Electricity demand is rising faster than many expected, and data centers are at the center of that shift. Whether the grid can keep up-without locking in a new wave of fossil capacity-will be one of the defining energy challenges of the decade.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
