Virginia Senate fails to pass a ratepayer-protection bill, effectively blocking AI data centers' permission to build in the state.

Virginia Governor Abigail Spanberger signed Executive Order 22 on September 18, transforming "Data Center Alley" — home to more AI server capacity than any other place on the planet — from the world's most permissive data center jurisdiction into one that requires community approval for every new project over 25 megawatts. Four days later, Prince William County voted unanimously, 8 to 0, to strip data centers of their remaining by-right development status across the entire county, requiring special-use permits — with public hearings and potential denial — for all future builds. Both actions arrived the same week the U.S. Senate blocked a bipartisan bill that would have required data centers to pay for their own grid upgrades, leaving American ratepayers in every state with no federal protection against absorbing the cost of electricity infrastructure that AI companies need.

These are not isolated events. In the first half of 2026, community opposition blocked, stalled, or indefinitely delayed approximately 120 AI data center projects across the United States, representing roughly $198 billion in planned investment — a figure that exceeds what the entire US data center industry spent in all of 2024. The communities doing the blocking span 49 states and have organized into 843 distinct opposition groups, driving bipartisan legislation from statehouses to Congress with no sign of slowing.

Prince William and Loudoun Counties — the two Northern Virginia jurisdictions comprising "Data Center Alley" — have historically operated as the global benchmark for permissive data center development. Loudoun County alone hosts more than 230 data centers and generated $1.2 billion in local tax revenue in fiscal year 2026. That revenue argument, repeated for a decade as the industry's primary public justification for new builds, is no longer sufficient to guarantee approval.

Spanberger's executive order does not impose a moratorium. It bans state agencies from entering nondisclosure agreements on projects — ending a practice that shielded deal terms from public scrutiny — and designates Eastern Virginia as a "cooling-water-scarcity area" subject to enhanced water-use oversight. It directs expedited development of state noise regulations, orders a review of diesel backup-generation emissions, and establishes Virginia's first Artificial Intelligence Task Force to address workforce displacement, data privacy, and cybersecurity risks. The governor called it "most comprehensive and aggressive data center accountability effort in the country."

The broader framework unveiled alongside the order outlines proposed legislation for Virginia's 2027 General Assembly — legislation that would eliminate by-right approval for any data center over 25 megawatts, end state subsidies for data centers in Virginia's site development programs, and remove large data centers from the state's fast-track permitting process. These remain proposals. The Prince William County zoning shift is already effective: developers who do not file applications under the old rules within a 90-day grace period will face discretionary review requirements.

Loudoun County voted on September 15 to move toward pausing new data center and substation approvals for up to 12 months, with a formal resolution expected in October. The county is also expected to revisit grandfathering provisions adopted for pending applications in 2025. Developers with projects in the pipeline are racing to confirm vested rights before those windows close. Law firm McGuireWoods, tracking the Virginia regulatory landscape, concluded that project timelines are at risk, entitlement strategies require revision, vested-rights questions are urgent, and the cost of doing business in Virginia will rise. "Developers should not assume that today's regulatory environment represents the floor," the firm stated.

Virginia's tightening matters more than other states. Northern Virginia hosts a disproportionate share of the world's AI inference infrastructure. When AWS, Microsoft Azure, Google Cloud, and Meta can no longer build new hyperscale campuses in their primary US market on existing terms, they must find alternatives. They have begun redirecting investment toward Spain's Aragon region, Norway's hydropower corridor, and Central and Eastern Europe — but no secondary market can absorb the displacement on a timeline matching hyperscaler demand.

The scale of organized opposition warrants specificity. Data Center Watch, a project of AI intelligence firm 10a Labs, counted 75 blocked or delayed projects in the first quarter of 2026 alone, worth approximately $130 billion — a single quarter matching the total disruption of all of 2025. The second quarter added 45 more projects worth $68 billion, representing more than half of all large-scale data center developments the organization tracked during the period. Together, those two quarters stalled roughly $198 billion — approaching the combined annual capital expenditure of the five largest US technology companies.

Carbon Direct's analysis of 46 cancellations since January 2024 — totaling more than $170 billion in announced investment — revealed a pattern differing from how industry spokespeople describe the opposition: communities blocking these projects cluster near the national demographic median, not wealthy suburbs or historically underserved communities. The concern driving the fastest-forming, most durable opposition is not primarily power or water, but process.

Non-disclosure agreements between developers and local governments, shell LLC ownership structures obscuring the actual end-user (typically an unnamed hyperscaler), and closed-door pre-application negotiations appear more consistently in opposition narratives than any other concern. "Developers who treat community engagement as a permitting formality are burdening communities and breaking trust, and it is costing them billions in capital and future opportunities," said Grant Gutierrez, head of community impacts at Carbon Direct. Virginia Governor Spanberger's NDA ban directly targets the practice Carbon Direct identified as the most reliable opposition trigger.

The opposition's political character is its most striking feature to investors who expected the debate to follow partisan lines. A Gallup poll conducted in March 2026 found 71% of Americans opposed constructing AI data centers near their homes, including 63% of Republican voters, 75% of Democratic voters, and 74% of independents. A February 2026 Marquette University Law School poll found 70% of Wisconsin voters believe data center costs outweigh benefits. Data Center Watch's analysis found that nearly two-thirds of blocked investment sits in counties that voted for Donald Trump in 2024.

President Trump has publicly encouraged communities to accept data centers, calling them "money machines" at a summer 2026 event. His framing has not moved public sentiment.

The dispute over electricity costs is not a communications problem but a direct consequence of the physics of AI-era data centers. Modern hyperscale AI facilities training and running large language models require 100 to 500 megawatts of continuous power, excluding cooling overhead. A single Nvidia H100 GPU draws approximately 700 watts; a cluster of 100,000 H100s — roughly the scale needed for major AI training — consumes approximately 70 megawatts before accounting for cooling infrastructure. Cooling creates a second thermodynamic constraint: minimizing water requires substantially more electricity, while near-zero-electricity water-based evaporative cooling competes directly with residential and agricultural water supply. There is no engineering solution eliminating both costs simultaneously.

When a new 300-megawatt campus connects to a regional grid, utilities must build the generation, transmission, and distribution infrastructure to serve it. Under existing law — specifically the Public Utility Regulatory Policies Act (PURPA), the 1978 federal statute governing large-load cost allocation — states determine who bears those costs.