What happened
RAND publishes a new site-screening rubric and quantitative energy-cost model for siting gigawatt-scale AI data centers on US federal land, driven by the 2025 executive order opening federal leaseholds. RAND evaluated 12 US Department of War sites offered for leases against 17 Department of Energy sites and modelled present-value energy infrastructure costs across 31 candidate locations under four generation scenarios, then stress-tested findings in a 14-participant federal/industry workshop. Central finding: 'Most sites will require major energy investment' — nearly every candidate site lacks existing generation capacity for gigawatt-scale demand, and 'at the gigawatt scale, energy infrastructure expenses are comparable in magnitude to data center construction costs.' The report also warns that unresolved community opposition, litigation, and infrastructure disputes at a federal site 'could escalate into overlapping legal, political, and security crises.' It recommends treating energy infrastructure as a parallel coequal workstream in all lease solicitations and issuing formal PPP policy guidelines covering risk allocation, security responsibilities, and decommissioning.
Why it matters
For executives planning AI infrastructure and energy procurement, this provides the first public quantitative benchmark on how binding energy — not acreage — is as a constraint on federal AI data center builds, and where the cost pressure sits relative to construction.
Action needed
Use the RAND siting rubric (energy supply, grid interconnection, infrastructure readiness) to model your own build sites ahead of federal solicitations, before lease terms lock in energy obligations.