The Stanford 2026 AI Index, released this week, contained a data point that received less attention than the capability benchmarks but may matter more for infrastructure planning: local governments in the United States are beginning to embrace restrictions or outright bans on new data centre development.
This is not a fringe phenomenon. A Wisconsin town voted last week to restrict tax incentives for an OpenAI data centre project. Communities in Virginia, Texas, and Georgia - the three states with the highest concentration of US data centre capacity - have seen local opposition movements cite energy costs, water consumption, and noise levels as reasons to slow or stop new builds.
Why This Is Happening Now
AI data centres are meaningfully different from previous generations of cloud infrastructure in their resource intensity. A single modern hyperscale AI training facility can draw 100 to 500 megawatts of power - enough to supply a small city. Water consumption for cooling systems in hot climates runs into millions of gallons per day. The transformers and substations required can overload local grid infrastructure that was not designed for this demand.
For communities that agreed to data centre construction in previous decades with promises of jobs and tax revenue, the reality has often been disappointing. Modern hyperscale facilities employ relatively few local workers - the construction phase brings hundreds of jobs that disappear once operations begin, leaving a facility that draws enormous resources but contributes minimally to local employment.
The energy cost externality is the most acute trigger. When data centres absorb regional grid capacity, residential and commercial electricity prices rise. Several Virginia counties have seen measurable electricity price increases attributed in part to data centre density. That creates a directly visible cost to local residents that override abstract benefits of being part of the AI economy.
The Strategic Implication
For AI companies and infrastructure investors, local government opposition represents a new constraint that did not feature in most infrastructure build-out plans from 2024 and 2025. Sites with permits and willing local governments are becoming a scarce resource independent of land availability or energy supply. Companies that locked in favorable locations and permitting arrangements early - particularly in nuclear-powered regions like Northern Virginia and the Pacific Northwest - have advantages that cannot easily be replicated.
Scotland's renewable energy advantage and Ontario's nuclear grid, covered in our newsletter recently, look more valuable against this backdrop. The AI infrastructure race is not just about who can raise capital. It is increasingly about who can build.
