Federal policy is pushing AI infrastructure through the approval process as fast as the executive branch can manage. An executive order issued in July 2025 accelerates federal permitting for large data centres, covering facilities that add more than 100 megawatts of new electricity load, cost at least 500 million dollars, or serve national security purposes. It streamlines environmental review, expands FAST-41 coverage and promotes the use of federal and contaminated land. Executive Order 14318 remained in force as of September 2026.
A second order, issued in December 2025, directed the Commerce Secretary to publish a list of state AI laws considered invalid. That list has not been published.
The limits of the federal lever are structural. None of these orders preempt state permitting, zoning or energy regulation. That is where the real constraint sits, and twenty seven states are now advancing measures aimed at large load customers.
Three have already enacted them: California with SB 57, Ohio with SB 103 and Utah with HB 507. The common core is cost allocation. Where the industry signed a voluntary Ratepayer Protection Pledge on 4 March 2026, in which major developers committed to cover the full cost of new generation needed to serve their load, the state statutes make the same allocation legally enforceable.
Reporting thresholds are also being set far below the federal line. State bills reach facilities as low as 10 megawatts, against the 100 megawatt trigger in the federal order, and several states are advancing water usage reporting alongside energy reporting.
The sharpest version is in Maine, which is positioned to become the first state to enact a moratorium pausing new data centre construction until November 2027.
For anyone modelling compute capacity in the United States, the federal timeline is no longer the binding constraint. Interconnection queues, state cost allocation rules and local land use decisions are. A project can clear an accelerated federal environmental review and still wait on a utility commission that now has a statute telling it who pays for the new generation.
The political economy is straightforward. AI load growth arrives as a visible increase in residential electricity bills, and elected officials at the state level respond to that faster than they respond to national competitiveness arguments. Federal acceleration and state cost shifting are not contradictory policies. They are two different constituencies being answered at the same time.
The practical planning assumption for 2027 is that siting decisions will be driven by which state utility regime is settled, not by which site clears federal review first.