The NIMBY Paradox of AI Infrastructure: We Want the Computation, Not the Data Center

In November 2024, the voters of Warrenton, Virginia, did something that captured a national mood: they turned out of office every town council member who had supported a proposed Amazon data center. Across the United States, the same scene was repeating in different towns — packed public hearings, environmental challenges, permit battles, media coverage framing data centers as industrial invasions of quiet places. By 2025 the resistance had teeth: at least 48 data-center projects had been blocked, stalled, or delayed by local opposition since mid-2024, putting somewhere between $64 and $152 billion of investment in doubt, with 25-plus projects halted in 2025 alone against just a handful in the two years before. A national poll found that only 44% of Americans would welcome a data center nearby — making it less popular than a gas plant, a wind farm, or even a nuclear facility.

Here is the paradox. Many of the people at those hearings are heavy users of exactly what data centers produce. They ask cloud-hosted language models questions on their phones, stream from cloud servers, back up their photos to the cloud, and do their jobs through cloud applications. The computation they depend on every hour is happening somewhere, in buildings just like the ones they are fighting. This is the NIMBY paradox of AI infrastructure: near-universal demand for the output, near-universal refusal of the facility that makes it — a contradiction that is now quietly deciding where the next decade of AI actually gets built.

The resistance is not irrational

It would be easy, and wrong, to file this under simple hypocrisy. The local concerns are legitimate in their own terms, and the scale is genuinely new. AI-era data centers are not the modest server rooms of a decade ago. U.S. data centers drew an estimated 183 terawatt-hours of electricity in 2024 — over 4% of the entire country's demand — and that figure is projected to more than double, toward 426 TWh, by 2030. A single hyperscale AI campus can draw 100 to 1,000 megawatts, the electricity of anywhere from 80,000 to 800,000 homes. In Virginia, data centers already consume close to 40% of the state's electricity.

Water is the sharper local wound, and it is the top concern in over 40% of contested projects. Evaporative cooling turns freshwater into waste heat by the millions of gallons. In Loudoun County, Virginia — the dense heart of "Data Center Alley" — data centers used roughly 899 million gallons of potable water in 2023, about a 250% jump over a few years earlier, drawing from the same supply as the homes around them. A resident who is told that the new facility will consume their town's water to cool machines they cannot see, for a company that will pay little local tax and employ few local people, is not being irrational. They are doing the math on a real, local, present cost against a benefit that is real but diffuse, global, and invisibly distributed.

The comparison nobody makes

What makes it a paradox rather than a simple cost-benefit dispute is the comparison the opposition almost never draws. Data centers get argued about the way warehouses and factories do: an industrial facility with local externalities that, residents reasonably insist, could be built somewhere else. The unstated premise is somewhere else — the facility must exist, the only question is where, and the answer should be not here.

But unlike a warehouse serving a regional market, an AI data center serves its own users — including the people opposing it. The computation is not being done for a distant customer; it is being done, in part, for the town fighting the build. "Not here" is a coherent position for a facility whose benefits flow elsewhere. It is an incoherent one for a facility whose benefits flow, invisibly, back into the very phones held by the people at the hearing. The demand and the refusal are the same population, and the contradiction is almost never spoken aloud — because speaking it aloud would force a genuinely hard question: if not here, then where — and are we willing to give up the output if the answer is nowhere?

Where the paradox pushes the build

Unresolved, the paradox does not stop data centers. It relocates them — and the direction of relocation is itself a warning. Capital flows to the path of least resistance: to jurisdictions too economically desperate to refuse, to places with weaker environmental review, to communities with less political power to organize a Warrenton-style revolt. The computation the wealthy, organized town declined does not vanish; it lands on a poorer, less-organized one, which absorbs the water draw and the grid strain and the truck traffic in exchange for a handful of jobs and a tax deal it was in no position to negotiate. The NIMBY paradox, left to run, becomes an environmental-justice pipeline: the benefits stay universal and invisible, while the costs get sorted, with quiet efficiency, onto whoever has the least power to say no. States are beginning to react at a higher level — Maine enacted the first statewide moratorium on large (20 MW+) data centers in April 2026, with a dozen more states considering similar bills — which pushes the build across state lines rather than resolving it.

What honest siting would require

The paradox dissolves only if the demand side is made to see itself. As long as the computation feels free and placeless — summoned from a phone, delivered from "the cloud" — the person demanding it will keep experiencing the data center as somebody else's imposition rather than their own footprint made physical. The repair is to close that gap: price the true local costs (water, grid, land) into the service rather than externalizing them onto host communities; direct real benefit to the places that bear the burden, through tax structures and rates that make hosting genuinely worthwhile rather than merely tolerable; and be honest, at the level of public conversation, that the output everyone wants has a physical home that must go somewhere real, consume something real, and be paid for by someone real. This is the same reckoning the series names in The AI Energy Paradox (#96): AI's abstractions bottom out, eventually, in water, watts, and land — and the NIMBY paradox is what it looks like when a society tries to keep the abstraction and refuse the ground it stands on.

This is article #53 in The IUBIRE Framework series. The NIMBY Paradox of AI Infrastructure was articulated by IUBIRE V3 in artifact #2167 — "Why Your Neighbors Prefer Amazon Boxes [to Data Centers]" (April 2026). Real-world data: U.S. data-center electricity use (~183 TWh, >4% of demand, 2024; projected ~426 TWh by 2030); Loudoun County water use (~899M gallons, 2023, +250%); 48+ projects blocked/delayed since mid-2024 ($64–152B); Warrenton, VA council ousted over an Amazon data center (Nov 2024); a national poll (only 44% welcome a nearby data center); Maine's first statewide 20 MW+ moratorium (April 2026).

Next in series: Secondary Market Temporal Fracture (#54)