The AI Data Center Backlash

1 · Moratoriums and the state-level revolt

The politics flipped in 2026. After two years of states competing to attract data centers with tax breaks, the pendulum swung hard toward restriction. New York's legislature passed a one-year moratorium on new hyperscale data centers in June 2026 (awaiting the governor's signature). More than 300 data-center bills were introduced across 30+ states — rate-protection clauses, mandatory water disclosure, tax-incentive clawbacks, and stricter siting rules.

Local opposition is now material to project economics: community resistance blocked or delayed an estimated $98B of projects between March and June 2025 alone. The era of frictionless siting is over.

2 · Electricity bills — the political flashpoint

The most potent grievance is the monthly bill. US utilities filed more than $30B in rate-increase requests in 2025, affecting roughly 81 million people, and average residential power bills are up about 40% since 2021. In the PJM market — the largest US grid — the independent monitor attributed roughly $13.8B of consumer bill increases to data-center demand, and households near some large campuses saw bills climb up to 267%.

3 · Water — the quieter cost

Evaporative cooling trades electricity for water. A 100 MW campus consumes roughly 200–400 million gallons per year — comparable to a town of 2,000–4,000 homes — and the strain lands hardest in the hot, dry regions (Arizona, Texas) where cheap power and land pull data centers in the first place.

The trajectory is toward closed-loop and direct-to-chip cooling that recirculates the same water, plus designs that consume essentially zero on-site water. But the existing fleet — and the gigawatts being built right now — still leans on evaporative towers.

4 · Who pays for stranded capacity

The deepest structural fight is about risk transfer. When a regulated utility builds gigawatts of new generation for a single hyperscale customer holding a lease it can exit every few years, the risk that those long-lived assets become "stranded rate-base" — if AI demand softens — falls on everyone else on the system. Louisiana's largest industrial power users formally opposed parts of the Meta Hyperion deal on exactly these grounds, and similar fights are live in Virginia, Ohio, and Mississippi.

This is the policy question that will define the 2026–2030 buildout: how to let AI infrastructure scale without quietly socializing its downside onto residential and industrial ratepayers.

5 · Frequently asked