Rural America starts voting against the AI boom's power-hungry datacenters

Information Technology 3 min 54 sources

In Coweta County, Georgia, residents are gathering signatures to force a public vote on an 831-acre AI datacenter — part of a fast-spreading revolt against the power, water, and land the AI boom needs, landing in towns that get the costs but none of the upside.

Key takeaways

Residents of Coweta County, Georgia are gathering signatures to force a public vote on an 831-acre AI datacenter, part of a fast-spreading revolt that already saw Monterey Park, California ban datacenters outright.
The fight is physical: datacenters drink water and pull power on a scale that strains local grids — one Georgia utility gave a developer 30 million gallons of water free before residents noticed their pressure dropping.
The same AI boom that just made SpaceX's IPO the largest in history needs to land as power and cooling somewhere, and the rural towns that get the costs but none of the upside are starting to vote it down.

The AI boom runs on buildings. Behind every chatbot answer and every image a model spits out sits a datacenter — a warehouse packed with computer chips that draws enormous amounts of electricity and water to stay cool. This week, the people who live next to those buildings started fighting back at the ballot box.

What happened

In Coweta County, Georgia — about an hour south-west of Atlanta — hundreds of residents turned out after church on a recent Sunday to sign a petition. The goal: force a public vote on Project Sail, an 831-acre datacenter, and block others like it [26]. Organizers had collected roughly 6,500 signatures by Friday and need about 14,000 to trigger a referendum, a rare tool that lets voters overturn a decision their county commission already made [26].

If it works, Coweta would be only the third county in Georgia’s history to hold such a vote [26]. It comes weeks after Monterey Park, California, became the first US city to pass a referendum banning datacenters outright [26]. This is not one angry town. Recent polling found seven in ten Americans would oppose a datacenter being built near their home [26].

Why now

The AI industry is in a building frenzy, and the buildings have to go somewhere. Developers favour rural counties with cheap land and “loose zoning laws,” as one Coweta resident put it [26]. The county commission rezoned the 831-acre Project Sail site from “rural conservation” to “industrial” — and at least five datacenters are now planned for Coweta alone [26].

The friction is physical, not abstract. Datacenters drink water to cool their chips and pull power on a scale that can strain a local grid. In nearby Fayetteville, Georgia, residents noticed their water pressure dropping; the utility then discovered it had handed a datacenter developer 30 million gallons of water — free of charge [26]. Loudoun County, Virginia, now nicknamed “Datacenter Alley,” is so dense with these warehouses that one Coweta organizer moved away from it to escape them [26].

What’s new is the organizing. “I’ve never seen grassroots opposition as I am seeing with datacenters,” said Chris Manganiello, a water-policy director who has tracked Georgia’s rivers for years. “Everything about datacenters in Georgia is unprecedented” [26]. The people doing it are mostly first-timers — a graphic designer, a retired teacher, a lifelong resident who’d never been to a commission meeting before this one [26].

What changes

A successful referendum doesn’t just stop one project — it sets a template. Coweta’s organizers say they want other towns to copy the playbook, and rural counties across Georgia are already passing moratoria, temporary bans on new construction [26]. For an industry that assumed empty rural land was the path of least resistance, the math is shifting: the cheapest place to build is becoming the place most likely to vote you out.

The bigger picture: the boom needs power somewhere

The datacenter fight connects to the week’s loudest tech story. SpaceX’s record stock-market debut — the largest in history, briefly making Elon Musk the world’s first trillionaire — was sold partly on its AI ambitions, including orbital datacenters [7][6]. With OpenAI and Anthropic both filing to go public, Wall Street is pricing in years more of AI build-out [23][7]. SpaceX closed Friday worth $2.1 trillion, sixth among US-listed companies [6].

All that capital has to land as physical infrastructure — chips, power, cooling, and the land to put them on. The investors buying the boom and the residents of Coweta County are looking at the same machine from opposite ends. One sees a balance sheet. The other sees a bald eagle’s nesting ground rezoned to industrial.

The angle

If your work touches AI infrastructure — siting, energy procurement, cloud architecture — the cheap-rural-land assumption is no longer safe to bake in. Local referendums and moratoria are becoming a real schedule and cost risk, not a fringe one. The number to watch isn’t model benchmarks; it’s how many counties pass a moratorium this year.

02 · Lesson · why it matters

The bill for a system rarely lands where the decision was made

Whoever decides to build a thing and whoever pays its real costs are often two different groups — and the gap between them is where the costs hide.

A retired teacher, a graphic designer, and a lifelong farmer walk into a market after church to sign a petition. None of them had been to a county commission meeting before this year. None of them set out to fight an industry. They are now trying to stop an 831-acre datacenter — and what they’ve stumbled into is one of the oldest shapes in how systems work.

The decision and the cost happen in different rooms

A datacenter gets approved in a commission chamber. The people in that room weigh the tax revenue, the jobs, the developer’s promises. They are choosing.

The water it drinks, the power it pulls, the night sky it floods with light — those land somewhere else entirely. In Coweta County, they land on a graphic designer two miles away who worries about her groundwater, and on a man who’s watched bald eagles nest there all his life.

The decision and the cost are split across two different rooms. That split isn’t a glitch. It is the normal way large systems get built. The people who hold the upside sit at the table. The people who hold the downside are usually somewhere else — at work, at home, not yet aware there’s a table at all.

When the bearer can’t object, the cost looks free

In nearby Fayetteville, the utility handed a datacenter developer 30 million gallons of water — for free — and nobody noticed until residents’ taps lost pressure. That’s the tell.

A cost only counts when someone with a voice is made to pay it. The water was never free; it was just charged to people who weren’t asked and didn’t see the bill. To the developer’s spreadsheet, water that nobody invoices is water that costs nothing. The cheapness was an illusion built out of someone else’s silence.

This is why the industry favours rural counties with, as one resident put it, “loose zoning laws.” Not because the land is empty — it isn’t, people live there — but because the people there are the least equipped to send back an invoice. The cheapest place to build is the place where the cost stays quietest.

The same machine, seen from two ends

This week, SpaceX’s stock-market debut became the largest in history, sold partly on AI. Investors are pricing in years more of this build-out. From that end of the machine, AI is a balance sheet that keeps getting bigger.

From the Coweta end, the same machine is a rezoning notice that turned “rural conservation” land into “industrial” land near your house. Both are looking at the identical thing. Neither is lying. The investor genuinely sees growth; the resident genuinely sees loss. The machine is large enough that no single seat can see both halves at once — and the halves don’t naturally talk to each other.

The boom needs to become real somewhere. The capital raised in New York lands as concrete and cooling towers in Georgia. The line between the two is invisible until you trace it — and almost nobody is paid to trace it.

The cost becomes visible only when someone makes it

Here’s the part worth carrying past today. The water, the power, the rezoning — all of it was happening for over a year while commission meetings drew “less than a dozen” people. The cost was real the whole time. It just wasn’t seen.

What changed wasn’t the datacenter. What changed was that one neighbor said, “It’s two miles from our house. I want to be aware of what’s happening,” and started showing up. Attendance went from a dozen to over a hundred. A cost that had been silent acquired a voice, and the moment it did, it stopped looking free.

That’s the mechanism, stripped bare: a system’s hidden costs don’t announce themselves. They sit quietly on whoever can’t yet object, and they stay off the books until someone decides to read them out loud.

What the whole looks like from here

It would be easy to leave this as a story about good rural people versus greedy billionaires. It isn’t, quite. The investor isn’t a villain for seeing a balance sheet; that’s the only thing visible from where they sit. The commissioner weighing tax revenue isn’t lying about the jobs. Each seat sees a true and partial slice.

The harder thing to hold is that you are probably at the table end of some system right now — getting a fast answer, a cheap service, a convenience — whose real bill is being charged, quietly, to a room you’ll never be in. The chatbot that answered your question this morning drew its water somewhere. Someone near that somewhere may not know yet.

Seeing the whole here doesn’t make you clever about datacenters. It makes you slower to trust the word “free,” and quicker to ask the only question that ever surfaces a hidden cost: who’s paying for this who wasn’t asked?

03 · Lab · your turn

Where the Bill Lands

Rehearse siting a datacenter and watch how a system's real cost stays "free" until the people paying it can object.