Federal Land for AI Data Centers Is Plentiful. The Power Grid Is the Limit.

Federal land for AI data centers is no longer scarce. The power to run a gigawatt-scale campus is. RAND, a federally funded research center, has now put an energy-first screening rubric over 31 candidate sites, 12 offered by the U.S. Department of War and 17 held by the U.S. Department of Energy, so the next round of federal lease solicitations can be screened on energy before bids go out, not after ([RAND, AI Data Center Siting on Federal Lands](https://www.rand.org/pubs/research_reports/RRA5050-1.html)).

A 2025 executive order pushed federal agencies to open federal land for AI data center development and to identify suitable sites. The RAND report is the first systematic attempt to rank that pipeline on the dimension that actually decides whether a campus gets built, which is energy. "Gigawatt-scale" here means a single campus drawing roughly the output of a large nuclear reactor, on the order of a gigawatt of continuous load, the working unit for frontier AI training and inference clusters.

The rubric sorts sites on three axes. The first is energy cost, including delivered electricity price, fuel-mix exposure, and the long-run cost of building new generation on or near the site. The second is grid interconnection feasibility, meaning how much new transmission, substation capacity, and queue time the site would need before a megawatt could actually flow. The third is infrastructure readiness, including water, roads, and the civil works needed to host a multi-hundred-acre compute campus. RAND then ran a cost model across all 31 locations to compare total energy infrastructure expense, not just the lease price.

The rankings point in one direction almost everywhere. Sites in the Southwest perform best for most scenarios because they sit close to existing gas and solar resources and have shorter interconnection queues. Several sites in the Northeast consistently rank at the bottom, not because the land is wrong, but because the surrounding transmission is constrained, fuel costs are higher, and the build-out path is longer. The report treats these as conditions, not geography. As new transmission is permitted, as nuclear restarts, and as storage gets cheaper, the map will reshuffle.

Across both departments, nearly every candidate site lacks sufficient existing generation to power a gigawatt-scale data center without significant new infrastructure. At that scale, energy infrastructure expense is comparable in magnitude to the data center construction cost itself. A campus that pencils out at $10 billion to build can carry a second $8-to-$10 billion of grid work before the first rack powers on, and that grid work is what most bidders currently leave to the government.

That gap drives the report's central recommendation: treat energy infrastructure as a parallel, co-equal workstream alongside civil construction in every request for lease proposals, not as a subordinate item to be resolved after a lease has been awarded. In practice, a request for lease proposals has to specify the power timeline, the interconnection plan, and the generation build-out the bidder will commit to, on the same schedule as concrete and steel.

The second recommendation is a screening tool bidders can be required to describe up front: the Ratepayer Protection Pledge, the federal commitment that data center energy costs will not be passed through to consumer electricity rates. The pledge is a bidder-screening concept in this report, not yet an enforceable federal requirement, and the rubric uses it to force prospective operators to explain how they will keep residential bills flat while pulling gigawatts off the same grid.

A third leg came from a stakeholder workshop that stress-tested the rankings. In one tabletop exercise, community opposition, litigation, and infrastructure disputes at a federally approved data center site cascaded into overlapping legal, political, and security crises that threatened the development itself. The report frames this as a planning risk, not a documented incident, but it is the reason the rubric puts interconnection and community conflict on the same axis as dollars-per-megawatt: a cheap site with a hostile permitting environment is not actually cheap.

The next move is on the agencies. The Department of War and the Department of Energy now have a public, peer-reviewed way to score their remaining candidate sites before the next solicitation closes, and a written argument to attach the Ratepayer Protection Pledge and a parallel energy workstream to every lease package. The Senate's separate data-center power bill, which would reshape how large loads pay for grid upgrades, is the payment mechanism, not the siting map. The rubric is in hand. The next Department of War and Department of Energy solicitations will show whether the screening runs on energy or on acreage.