DOE picks 16 federal sites

AI data centers are running into a very physical problem — land, power, cooling, and permits. That is the real story here. On April 3, 2025, the Department of (energy.gov)from data-center developers, energy companies, grid operators, tribes, states, and local communities. The point is simple: use federal land that already has energy infrastructure, security, and research assets to speed up big AI buildouts. (energy.gov) ### Why is DOE doing this now? Because AI demand is smashing into the grid. Training a(energy.gov) huge amounts of electricity plus water or other cooling systems. DOE framed the move as part of a broader push to keep U.S. AI leadership while lowering costs by co-locating data centers with new power generation on DOE land. The agency tied the effort to the January 23, 2025 executive order on removing barriers to American leadership in AI. (energy.gov) ### What exactly did DOE announce? Not final winners — a shortlist. DOE identified 16 “potential s(energy.gov), a path to faster permitting for new generation, including nuclear. The RFI is the market test. DOE wants to know who would build, operate, and finance these projects, and what technical and economic models make sense. (energy.gov) ### Which sites made the list? The 16 sites span national labs, defense-related facilities, and former uranium-enrichment complexes: Idaho National Laboratory, Paducah, Portsmouth, Argonne, Brookhaven, Fermi, NE(energy.gov)ia, Savannah River, Pantex, and the Kansas City National Security Campus. That mix tells you DOE is not just offering empty land — it is offering places already wired into serious science and energy systems. (world-nuclear-news.org) ### Why do former nuclear sites matter? Because they already look a lot like the kind of industrial campuses hypersca(energy.gov)n the MISO market, with major water access and existing transport links. Portsmouth was described as designed for 2.2 GW in PJM. Basically, these are not greenfield cornfields waiting years for substations — they are heavy industrial sites with unusual power potential. (rcrwireless.com) ### Why involve the national labs? The labs bring more than acreage. DOE explicitly pitched co-location with research facilities as a way to imp(world-nuclear-news.org)n attempt to turn federal science campuses into AI-energy testbeds. That could matter for advanced nuclear, geothermal, storage, cooling, and grid integration. (energy.gov) ### How fast is the government trying to move? Faster than you might expect. DOE said it wants construction at selected sites to begin by the end of 2025 and operations to start by the end of 2027. That is aggressive for projec(rcrwireless.com)ase award or signed power deal. (energy.gov) ### What changed versus the earlier federal approach? The Biden administration’s late-stage plan had focused on at least three frontier-AI sites and a separate list of 10 nuclear-capable sites, with some timelines stretching mu(energy.gov)ely, and pushed a shorter operating target. Same basic idea — federal land for AI — but more concrete and faster. (rcrwireless.com) ### Bottom line? This was the moment the federal AI-infrastructure conversation stopped being abstract. DOE did not just say America needs more compute. It put actual places on the table — with land, wires, water, and a clock running toward 2027. (energy.gov)