Nearly 150 protests against AI data centers happened across 42 states in July alone. That’s not a fringe number. It’s also not really about AI. It’s about water and power, and what happens to a town when a building that consumes both like an industry moves in next door.
What a data center actually costs a town
A large AI-focused data center can draw somewhere between one and five million gallons of water a day for cooling, with some running higher at peak. Power draw dwarfs a traditional data center of the same footprint.
That demand comes from somewhere. In rural Georgia, residents near Meta’s data centers have reported wells running dry or contaminated. One resident’s line, “I can’t drink the water,” has been quoted across most of the coverage of the buildout. In South Carolina, a permit letting Google draw 1.5 million gallons a day drew criticism for putting a data center’s cooling ahead of the water supply everyone else in the area depends on. Arizona’s data centers pull from the same aquifers as the state’s agriculture, in a state that already didn’t have water to spare.
The pushback is working, some of the time
In the first quarter of 2026, local opposition got twenty proposed projects canceled outright. Roughly $41 billion in planned investment that didn’t happen, at least not there. Virginia and Minnesota have both started drafting legislation requiring real reporting on water usage instead of leaving it a negotiated secret between a company and a utility.
This isn’t one political side’s project. Virginia and Minnesota aren’t the same kind of state, and the people showing up to these hearings are doing it because the well ran dry, not because of who they voted for.
Where we land
We run a NAS in a closet instead of renting space in someone else’s warehouse-sized one, and the gap between those two things is most of the point of this site.
That obviously doesn’t scale to training a frontier model. Nobody’s proposing you do that in a closet. But the current default is building infrastructure faster than the water table and the power grid underneath it can absorb, then treating the shortage as a surprise when it shows up.
A town finding out its water allocation got renegotiated without much say in it isn’t an AI problem specifically. It’s what happens when infrastructure gets built faster than anyone local gets a real chance to ask about it. Same instinct we’d apply to anything else quietly claiming a shared resource without much accounting for it.
Sources: Gizmodo on data center water use, Wikipedia’s overview of the opposition movement, the Lincoln Institute on land and water impacts.