The constraint just went kinetic. Dominion Energy filed suit this week against the Trump administration over a stop work order that froze five offshore wind farms mid-construction, including its $11.2 billion Coastal Virginia Offshore Wind project already 18 months into build. The timing isn't incidental. Hyperscalers are committing $50B+ annually to AI data center buildout, and Virginia's data center corridor—already the world's largest concentration—faces electricity demand doubling over the next 36 months. For the first time, power supply just became the bottleneck, not cost or real estate.

The lawsuit that landed Tuesday exposes what Dominion Energy's December 22 press release spelled out with unusual bluntness: "Virginia needs every electron we can get as our demand for electricity doubles. These electrons will power the data centers that will win the AI race." That's the inflection point stated plainly—energy is no longer a utility variable, it's a competitive advantage constraint.

Let's ground this in numbers. Dominion has already committed $8.9 billion to the Coastal Virginia Offshore Wind farm out of its $11.2 billion budget. The project broke ground in 2024. All federal, state, and local approvals were obtained and certified. The farm was on schedule to generate power next year. Then Monday came the stop work order from the Bureau of Ocean Energy Management (BOEM), halting not just this project but five offshore wind farms collectively.

The administration's justification centers on a 90-day "national security review" citing classified threat reports and concerns about turbine radar interference. Former Navy Commander Kirk Lippold, a recognized national security voice, immediately challenged the framing publicly: "I want to know what's changed? To my knowledge, nothing has changed in the threat environment that would drive us to stop any offshore wind programs." The federal complaint Dominion filed alleges the move is both "unlawful" and "arbitrary and capricious"—the precise language a federal judge just used to strike down Trump's earlier offshore wind memorandum earlier this month.

But here's where the timing becomes structurally important. This isn't just renewable energy policy clash. This is the moment when AI infrastructure expansion collides with power grid reality. Virginia hosts the world's largest concentration of data centers. Meta, Google, Amazon, and other hyperscalers collectively spend $50 billion annually on data center capex. That number is accelerating. Every major AI model requires exponentially more electricity during training—a single large language model can consume as much power in a training run as 130 US homes use in a year, multiplied across dozens of active training programs per company.

The power grid that serves Virginia's data center alley is already stressed. Rising electricity costs have become a flashpoint in Virginia elections and in communities near data center projects across the country. New supply was supposed to ease that pressure. The Coastal Virginia Offshore Wind farm, once operational, would produce 9.5 million megawatt-hours per year—enough to power roughly 660,000 homes. That's not insignificant. It's foundational infrastructure for the very AI race the Trump administration says it's trying to win.

This creates a strange circularity. The policy justification—winning the AI race—directly undermines the infrastructure needed for that race. And unlike previous offshore wind halts that got lifted (Revolution Wind off Rhode Island, Empire Wind off New York both had stops reversed by federal judges before being suspended again this month), this action comes with explicit 90-day uncertainty now layered onto 18-month construction delays that will push costs skyward. Dominion warns that every delay raises project costs that "customers ultimately pay for." That customer in this case is every enterprise buying compute power in the region.

What makes this an inflection point rather than a routine policy dispute is the constraint it reveals. For two decades, tech infrastructure planning treated power as a commodity—available, scalable, subordinate to other variables like real estate and connectivity. The shift to massive AI training workloads and the simultaneous regulatory uncertainty around power generation means that assumption just inverted. Power is now the variable that limits deployment velocity.

For builders currently planning data center expansions, the window on 2026 timelines just closed. Dominion's $8.9 billion commitment represents a binding constraint on regional power availability. A 90-day pause becomes a 12-18 month delay once construction ramp-up requirements are calculated. The federal court challenge will likely succeed—the "arbitrary and capricious" language from the earlier judicial ruling suggests legal precedent is running against the administration. But even if the pause is lifted in 90 days, the uncertainty itself becomes a cost that builders price into infrastructure decisions. That's how policy inflection points work at scale.

The precedent this sets matters too. If executive action can halt renewable infrastructure projects mid-approval despite completed environmental reviews, every major infrastructure project—transmission line upgrades, nuclear plant licensing, hydro expansions—now carries a different regulatory risk premium. That risk ultimately flows to enterprises buying power and ultimately to the consumers of AI services paying for compute time.

What to watch: The federal court will likely rule on Dominion's "arbitrary and capricious" challenge within 30 days. If the judge lifts the stop work order again, the real test becomes whether the administration uses other tools (national security reviews that actually drag past 90 days, expanded definitions of radar risk) to achieve the same delay. If the court upholds the pause, expect immediate challenges to similar infrastructure projects and a cascade of regulatory uncertainty that extends far beyond wind.

The energy supply constraint on AI infrastructure just transitioned from theoretical to litigated. Dominion Energy's lawsuit forces a federal court to rule on whether policy can unwind completed infrastructure approvals mid-construction, with direct implications for every enterprise buying power in the region and every hyperscaler planning regional data center expansion. For investors, this materializes infrastructure cost risk that wasn't priced into AI capex models. For decision-makers, alternative power procurement strategies (long-term contracts, regional diversification, load shifting) become immediate priorities. For builders, the 2026 timeline window just compressed. Watch the federal court decision in 30 days—that ruling will determine whether policy uncertainty becomes a structural feature of infrastructure planning or a recoverable delay.