Frontier AI is a software story with a heavy physical base. Analysts estimate that the largest cloud companies will spend roughly $600–700 billion on capital projects in 2026, about four times their 2023 level, most of it on data centres and the chips inside them.1 The International Energy Agency projects global data-centre electricity use rising from about 415 terawatt-hours in 2024 to about 945 by 2030, roughly 3% of the world's electricity, with the United States and China accounting for almost 80% of the increase.2
Those plants have to go somewhere, and the places they go are increasingly saying no.
The scale of the opposition
Data Center Watch, which tracks local resistance in the US, counted at least 75 projects worth about $130 billion blocked or delayed in the first quarter of 2026, the largest single-quarter total it has recorded. The number of active opposition groups more than doubled, from 396 to 833, spread across 49 states.3 The pace eased in the second quarter, to about 45 projects worth roughly $68 billion.4 The tracker cautions that its figures are estimates and that delay rarely has a single cause.3
Why people object
Power bills. The clearest grievance is electricity. In PJM, the grid that serves 13 US states from Illinois to New Jersey, the price of capacity rose from $28.92 per megawatt-day for 2024/25 to the capped maximum of $329.17 for 2026/27. The grid's independent market monitor attributed 63% of the increase in the 2025/26 auction, about $9.3 billion of $14.7 billion, to data-centre demand, and concluded that conditions in the capacity market were "almost entirely the result" of data-centre load.5 Not every analyst agrees that households are already paying: a Rutgers policy lab found no large, detectable increase in local residential bills yet, and some argue the auction's design is the real problem.6
Water. Large facilities can draw heavily on local water for cooling. In Malaysia, Johor state stopped approving the most water-intensive classes of data centre after protests in February.7 In Britain, the Environment Agency has said data-centre growth cannot be considered separately from local water supply.8
Land, noise and say. Residents object to rural land turned into industrial campuses, to round-the-clock noise, and to deals agreed, often with tax breaks, before the public hears of them.
From town halls to statehouses
- Local pauses. Nearly a third of Indiana's counties have moved to restrict data centres.9 Santa Fe County became New Mexico's first to adopt a pre-emptive moratorium, for 18 months.4 In California, at least 26 resolutions, moratoriums or bans have been passed or proposed since June, according to Food & Water Watch.10
- States. Texas paused environmental permits for new data centres while it reviews their water and energy use.11 A New York bill that would make it the first state with a statewide moratorium awaits the governor's decision.12 Illinois has suspended tax incentives for new projects, and Pennsylvania now requires binding commitments from projects seeking state permits, according to policy trackers.13
- Congress. Senator Bernie Sanders and Representative Alexandria Ocasio-Cortez have introduced an AI Data Center Moratorium Act. It has little chance of passing, but it marks the issue's arrival in national politics.13
- Elections. Ahead of the 3 November midterms, at least 20 candidates across 36 governors' races have promised some form of moratorium, and 38 data-centre-related measures are on ballots in eight states, according to research cited by WIRED and Ballotpedia.14
- Abroad. In Britain the Green Party has called for a moratorium on most new data centres; the government has rejected it, citing jobs.8
The case for building
Supporters point to construction jobs, local tax revenue, and the national stakes. Governments in the US and Britain treat AI capacity as a matter of economic and strategic competition, chiefly with China, and argue that what is not built at home will be built elsewhere. Many of the grievances are also about how projects are approved and who pays for new power, which can be fixed through rate design and planning rather than bans. Oregon, for example, has created a separate electricity rate class for data centres.6
Why it matters here
The physical build-out is the one part of frontier AI that ordinary democratic processes can see and touch. Model training happens in a handful of companies; the buildings, power lines and water permits it requires go through county boards, state utility commissions and voters. That makes data centres the place where public consent for the AI race is actually being tested, and, so far, often withheld. It also ties the AI story to the climate one: the extra electricity has to come from somewhere.
Whether the backlash slows frontier AI meaningfully is not yet clear. The second-quarter dip in blocked projects, and the scale of capital still flowing, suggest that builders are mostly finding somewhere else to go. What the opposition has already changed is the politics: AI infrastructure is now an electoral issue, and the terms on which it is built are being rewritten state by state.