DOE flags 16 federal AI sites

The thing to understand is that this is not just a data-center story. It is a land-and-power story. The Department of Energy decided in April 2025 that 16 federally owned or managed sites could be turned into AI infrastructure hubs, with the pitch that developers could build faster there because the land is already controlled and the energy hookups are closer than usual. The goal was blunt — get selected projects under way by the end of 2025 and have them operating by the end of 2027. (energy.gov) ### What did DOE actually do? DOE put out a request for information — basically a call for developers, utilities, grid operators, and other interested groups to say how they would use DOE land for AI infrastructure. The department said 16 sites were already identified as candidates f(energy.gov)e, or transmission upgrades. This was framed as a public-private buildout, not a government-owned supercomputer program. (energy.gov) ### Why use federal sites? Because the slow part of building giant data centers is often not the building. It is land control, interconnection, environmental review, and power delivery. DOE is trying to collapse some of that timeline by offering secure sites that already sit near lab(energy.gov)ap dirt, but fewer bottlenecks. (energy.gov) ### Why is AI tied to energy now? Because modern AI training and inference chew through electricity at a scale that looks more like heavy industry than software. DOE’s own framing says AI data centers need not just servers and cooling, but generation, transmission, and storage planne(energy.gov)wn. (energy.gov) ### How big could these sites get? Very big. Public summaries tied to the RFI described some candidate sites in the 750-megawatt to 1-gigawatt range. Brookhaven, for example, was described in local coverage of the federal filing as a roughly 90-a(energy.gov) 1,000-megawatt AI data park. That is utility-scale, not campus-scale. (federalregister.gov) ### So where does Fort Bliss fit? Fort Bliss is separate from the DOE’s 16-site list, but it shows the same logic pushed to an extreme. The Army’s proposed Fort Bliss complex in El Paso would ramp from about 100 megawatts of compute capacity nex(federalregister.gov)rest example of what “AI infrastructure” really means in practice — not just chips, but power plants, water, wires, and local backlash. (usnews.com) ### Why are people worried? Because these projects do not land on empty spreadsheets. They land in real grids and real communities. The Fort Bliss proposal has already triggered questions about water use, air pollution, and whether regional generation and tr(usnews.com) and who pays for the grid upgrades. (usnews.com) ### What is the catch for DOE? The catch is that faster federal siting does not magically create power. Even if land and permitting move quicker, the hard constraint is still electricity delivery. DOE can make sites available and attract developers, but util(usnews.com)f asked grid operators for input, which tells you DOE knows the real bottleneck sits there. (energy.gov) ### Bottom line? DOE’s 16-site move was an early signal that Washington had stopped treating AI as a pure software race. It is now treating AI as physical infrastructure — more like ports, pipelines, or semiconductor fabs. And once the conversati(energy.gov) to run them. (energy.gov)