Data centers are devouring electricity at a pace that has caught grid planners off guard. In a series of analyses published this spring, the U.S. Department of Energy laid out a blueprint for keeping up: pair fast-to-build wind, solar, and battery storage with steady output from nuclear reactors, both existing and yet to be constructed. The message from federal officials is that the tools exist to power the AI boom without breaking the grid or locking in decades of new fossil fuel generation. But the gap between a technically plausible plan and an executed one remains wide.
The scale of the problem
A DOE overview tied to research from Lawrence Berkeley National Laboratory describes data centers as one of the fastest-growing loads on the American grid, driven by the expansion of cloud computing, artificial intelligence training, and cryptocurrency mining. The DOE summary notes that data-center electricity consumption could rise sharply by the end of the decade, though the department’s press page does not reproduce LBNL’s full scenario figures. The underlying LBNL report, updated in December 2024, provides the detailed growth projections and methodology behind those estimates.
The growth is not spread evenly. The Energy Information Administration’s Short-Term Energy Outlook has flagged two regions bearing heavy load: ERCOT, the Texas grid, and PJM, which covers much of the Mid-Atlantic and Midwest. Northern Virginia’s “Data Center Alley” in PJM territory already hosts the world’s densest cluster of facilities, and Texas has become a magnet for new builds thanks to cheap land, deregulated power markets, and relatively fast permitting. The EIA warned that if clean energy additions fall short of projections in these corridors, fossil fuel generation will rise and wholesale electricity prices will climb, a cost that would ultimately land on ratepayers and businesses alike.
DOE’s two-tier playbook
The DOE’s Office of Electricity grouped the supply-side response into two tiers. The first covers resources that can be deployed relatively quickly: land-based wind, utility-scale solar, and battery storage, combined with efficiency improvements inside the data centers themselves. These technologies have mature supply chains and falling costs, making them the most immediate lever available.
The second tier is what DOE calls “clean firm” resources, meaning generation that runs around the clock regardless of weather. This category includes squeezing more output from existing nuclear plants through uprates, restarting recently retired reactors, and eventually building new small modular reactors or advanced designs. Data centers typically operate at capacity factors above 90 percent, a load profile that wind and solar alone cannot match without massive overbuild and storage. Nuclear fills that gap.
A companion analysis from the DOE’s Office of Nuclear Energy examined the specific advantages of pairing reactors with data centers. Nuclear plants deliver high-availability, weather-independent power, exactly what hyperscale operators need. The analysis also explored co-location, where a data center sits directly at or near a reactor site and draws power behind the meter, potentially bypassing transmission interconnection queues. Those queues have ballooned in recent years; Lawrence Berkeley National Laboratory’s annual queues report has documented wait times that now stretch beyond a decade in some regions, with more than 2,600 gigawatts of generation and storage capacity seeking grid connection nationwide as of its latest update.
That co-location concept is not hypothetical. Microsoft signed a 20-year power purchase agreement to support the restart of a reactor at Three Mile Island in Pennsylvania. Amazon acquired a data-center campus adjacent to a nuclear plant in Pennsylvania through its deal with Talen Energy. Google struck an agreement with Kairos Power for advanced reactor output. These corporate moves are exactly the kind of long-term contracts DOE’s framework envisions, and they signal that major tech companies see nuclear as part of their energy future, not just a talking point.
What the national labs modeled
A technical report produced jointly by Argonne National Laboratory, Idaho National Laboratory, and Oak Ridge National Laboratory went deeper, modeling nuclear capacity needs under several data-center growth scenarios. Published through the DOE’s Office of Scientific and Technical Information, the study examined grid-connected versus co-located and behind-the-meter configurations, weighed deployment pathways including uprates, restarts, power purchase agreements, and new construction, and identified constraints around reactor siting and fuel supply.
One constraint loomed large: high-assay low-enriched uranium, or HALEU, the specialized fuel required by several advanced reactor designs. The report flagged HALEU supply as a bottleneck under higher demand scenarios. The commercial supply chain for this fuel is still being built in the United States, and if production lags behind reactor deployment schedules, some of the clean firm capacity in DOE’s vision could arrive later than the data-center load it is supposed to serve. The report stopped short of publishing levelized cost comparisons between grid-tied and behind-the-meter nuclear setups, leaving data-center operators without a public, government-sourced benchmark for the economics of each approach.
The risks if clean energy falls short
The EIA’s modeling serves as a reality check on DOE’s more optimistic framing. Where DOE emphasizes what clean resources can accomplish under favorable conditions, EIA quantifies what happens when those resources do not arrive fast enough. In its scenarios, shortfalls in renewable and nuclear deployment lead to increased natural gas burn and higher wholesale power prices, particularly in ERCOT and PJM.
That is not an abstract risk. Several utilities have already filed plans for new natural gas plants, citing data-center demand as the justification. Georgia Power, Dominion Energy, and others have proposed or received approval for gas-fired capacity in the past two years. If clean alternatives cannot be permitted and built on competitive timelines, the data-center boom could lock in fossil infrastructure that operates for decades, undercutting federal and state climate targets.
Permitting remains the central bottleneck. The DOE analyses identify reactor restarts and small modular reactors as part of the solution but do not specify when enough capacity could realistically come online. Licensing timelines for new nuclear designs at the Nuclear Regulatory Commission have historically stretched for years. Co-location projects face their own hurdles: local zoning, community acceptance, water availability for cooling, and the sheer capital cost of building or restarting a reactor. None of these factors are fully quantified in the current federal documents.
Where execution will be won or lost
For grid operators, the DOE framework offers a planning vocabulary: diversify supply across fast-deploying renewables and steady nuclear, rather than defaulting to gas. For data-center companies, it validates the strategy of signing long-term nuclear contracts and exploring co-location, while flagging the fuel and permitting risks that could delay those plans. For ratepayers, the stakes are straightforward. If data-center load grows as projected and clean supply does not keep pace, the cost shows up in higher electricity prices and increased emissions.
The federal analyses published this spring should be read as boundary markers, not guarantees. They establish that a technically credible path exists to power the AI and cloud computing surge with low-carbon electricity. But technical credibility is not the same as execution. The distance between DOE’s blueprint and a functioning grid that absorbs hundreds of terawatt-hours of new demand without buckling will be closed, or not, by permitting decisions, fuel supply investments, and corporate commitments that are still taking shape as of June 2026.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
