Data-centers siting concentrates noise, water, air and cost burdens on host communities while the benefits are dispersed

Observable evidence

The aggregate energy footprint of computing is a separate, already-catalogued problem. This issue is about a different asymmetry: a data centre is a point load whose costs land inside a few kilometres and whose benefits do not.

The concentration is measurable. Roughly 300 data centres operate in Northern Virginia, about 14% of the world total, and about one third sit within 200 feet of residentially zoned land, some within 50 feet (EESI). A typical 250,000 sq ft facility employs around 50 full-time workers, roughly half of them contractors (Good Jobs First), so the local labour benefit is small relative to the physical footprint.

Resistance is now scaling faster than the sector's ability to absorb it. Organised opposition groups grew from 396 at the end of 2025 to 833 by March 2026 across 49 US states, with roughly 75 projects worth about $130 billion blocked or delayed in Q1 2026 alone, matching the whole of 2025. More than 300 data centre bills were introduced in the first half of 2026, and 14 states saw statewide moratorium proposals (NBC News). Note that the underlying tracker, Data Center Watch, is industry-adjacent and counts pre-filing opposition as "delay", so the dollar figure is best read as an upper bound.

Who is affected

Households within roughly 300 metres of cooling plant and generator yards; ratepayers on the same utility; water users in the same basin or on the same municipal system; and school districts and local budgets where the tax base was abated. The burden is not evenly distributed within a community either: the sharpest US air-quality disputes are in neighbourhoods already failing ozone standards.

What is genuinely contested

An honest statement of this problem has to include the evidence that runs the other way. A hedonic regression on 2023 Northern Virginia home sales found properties closer to data centres sold higher, and concluded that the claim of depressed values "lacks statistical support" (George Mason CRA). Virginia's legislative audit commission found the sector's sales tax exemption returns $6.10 in personal income per $1 forgone, the state's second-best incentive (Cardinal News), while Good Jobs First reads the same recovery rate inversely and puts the K-12 cost at $267.4 million in FY2024. Those two figures measure different things, gross economic activity and fiscal transfer, and are routinely used interchangeably in public debate when they should not be.

Scope boundaries

This issue covers the local, place-based externalities of siting and operating a data centre, and the process by which those trade-offs are decided. It excludes the global electricity and carbon footprint of computing, which is catalogued separately, and it excludes the question of whether particular AI applications are worthwhile. The specific facets (noise, water, ratepayer cost, air quality, waste heat, and disclosure) are handled as sub-issues.

Current state versus desired state

Today the sequence is typically: a non-disclosure agreement, a rezoning, then measurement and mitigation after residents complain. The desired state is that the noise limit, the water source, the generator standard and the cost allocation are all specified and enforceable before the permit issues.

Impact if unaddressed

Two failure modes, in opposite directions. Communities absorb harms that were avoidable at the design stage, because the design was fixed before they could see it. And genuinely useful compute capacity gets blocked wholesale, because a blunt moratorium is the only instrument a community has when the specific instruments (a noise ordinance that measures the right frequencies, a tariff that allocates grid cost, a permit condition on water source) do not exist.