Data Centers Use Almost None of America's Water. So Why Is Your Aquifer Dropping?

Updated October 26, 2026

So you looked it up. Data centers use some tiny sliver of the country's water — less than one percent, the kind of number that's supposed to make you exhale and move on. And that number is basically right. So why is the well level down the road from you dropping, and why did the county start talking about restrictions? I went digging, because "it's a rounding error" and "our aquifer is in trouble" can't both be the whole story. Turns out they're both true at once, and the gap between them is the whole thing.

The national number is real. It just isn't about you.

Here's the first thing I had to get through my head. When somebody says data centers use a sliver of one percent of the nation's water, they're adding up every drop in the whole country — every cornfield in Iowa, every lawn in Seattle, every snowmelt river — and setting the data centers against all of it. Against that pile, yeah, the servers barely register.

But nobody drinks the national average. You drink out of your county. And that's where the number falls apart like a cheap umbrella. An average that folds your dry county in with the entire rainy Midwest isn't a measurement. It's a card trick. It's like being told the average house on your street didn't flood — cold comfort if yours is the one with water in the basement.

And even the per-prompt version is slipperier than it sounds. Researchers at the University of California, Riverside figured a single hundred-word request to an AI model can burn through roughly 519 milliliters of water — about a bottle. One bottle sounds like nothing. Then you remember these things answer millions of requests an hour, every hour, forever.

The draw doesn't spread out. It piles up.

This is the part that flipped the whole question for me. The water isn't sprinkled evenly across the country like fertilizer. It gets dumped, on purpose, in a handful of places — and a lot of those places are already short on water.

Bloomberg ran the map and found that about two-thirds of the data centers built or in the works since 2022 landed in areas already under high water stress. Not spread thin. Concentrated, right where there's the least to spare. Five states soak up most of that build-out.

Then there's the power plants, which almost nobody counts when they tell you the comforting number. The servers themselves sip water for cooling, but the electricity feeding them is thirsty too — a lot of that juice comes from plants that boil and evaporate water to make steam. One analysis pegged that indirect, power-side water consumption at around 211 billion gallons in 2023. A separate study found the electric generators keeping data centers running in just seven states were drinking more water than all the cities in California combined — California's generators alone pulled about 1.4 trillion gallons in a single year. So even a data center that recycles every drop on-site can still be draining a reservoir two hundred miles away through the wall socket.

What that looks like from somebody's front porch

Averages don't knock on your door. People do. In the Texas Panhandle, part of the city of Amarillo's water comes out of the Ogallala Aquifer — the same aquifer the region's farmers live on, and it's already being pulled down faster than rain can refill it. Then a developer, a company co-founded by former governor Rick Perry, asked the city to sell it two and a half million gallons a day, with talk of pushing that to ten million. For reference, the mayor said the whole city of Amarillo uses about fifty million gallons a day. One campus, asking for a fifth of a city. Residents showed up at the county courthouse to push back, and one of them put the real question plainly: what are we going to have to give up so these things can run?

Out in Arizona, around Buckeye west of Phoenix, the state got worried enough about groundwater that it blocked some new development that leans on aquifers. And here's the tell that stuck with me. A guy in the comments under one of these explainer videos said an Amazon data center was going up near him, people kept asking what kind of cooling it would use — closed-loop that barely touches water, or the evaporative kind that drinks it — and nobody would tell them. That's the whole fight in one anecdote. Half the people in the thread insisted "the new ones don't use water." The guy living next to one couldn't get a straight answer about the one actually being poured next to his house.

That's the human on the other side of the spreadsheet. Not a percentage. A homeowner wondering if his well and his resale value just became somebody else's cooling budget.

So is it YOUR aquifer? Here's where I landed.

I'll level with you about what I couldn't nail down: whether your specific aquifer is dropping because of a data center is a question I can't answer from a national chart, and honestly, a lot of the time neither can your town. In Texas, most data centers aren't even required to report how much water they use. The disclosure just isn't there, which is exactly why the reassuring national number gets to stand in for an answer it was never measuring.

What I can tell you is where to point your attention: not the country-wide percentage, but your own county's water stress, your own utility's willingness to say what's being built and how it's cooled, and whether the power feeding it comes from plants that evaporate water upstream. That's the version of this that actually touches your tap.

This is the kind of rabbit hole we climb all the way down over at Byte Bungalow — the turbines, the transformers, the cooling, and who ends up paying for the water and the power when the servers move to town. If you want to see how the whole machine actually fits together instead of getting handed a tidy number that dodges the question, come watch the video and subscribe on YouTube. That's where the digging lives.

Common questions

Can't data centers just recycle the water in a closed loop so none gets used up?

Some do, and a closed-loop system really does recirculate its cooling water instead of evaporating it. But two catches. First, closed-loop cooling usually leans harder on electricity, and that power often comes from plants that evaporate water to make steam — so the water use moves upstream rather than disappearing. Second, plenty of large facilities still use open evaporative cooling, which consumes water by design. "Closed-loop" is a real thing; it's just not automatically the thing being built near you.

If the new data centers supposedly don't use much water, why won't the one near me say what cooling it uses?

That's the gap that drives most of the local anger. The low-water claim describes some newer closed-loop designs, but whether a specific campus uses that or an evaporative system is often not disclosed — and in states like Texas, data centers generally aren't required to report water use at all. No report, no straight answer, and the national average gets quoted instead of the local facts you actually asked for.

Does a data center hurt my well if it pulls from a different, deeper aquifer?

Developers sometimes argue a deep aquifer isn't connected to the shallower one your well taps, so your water is safe. Sometimes that's true, sometimes the two are linked and nobody's proven which. It's a site-by-site geology question, not a blanket guarantee — treat "different aquifer, no impact" as a claim to verify locally, not a settled fact.

How far from a data center do I need to live for it to affect my water?

There's no fixed distance, because the impact runs through shared water sources, not through the fence line. If a facility draws from the same aquifer or municipal supply your home uses, it can matter miles away; if it's on a separate source, a neighbor across the street might not feel it. What matters is whether you share the straw, not how close the building is.

If data centers are under one percent of national water use, isn't the whole worry overblown?

The national percentage is accurate and nearly useless for your situation, because the draw is concentrated, not spread out. Bloomberg found about two-thirds of recent data centers went into high-water-stress areas. A tiny share of a huge national total can still be a massive share of one dry county's supply — which is the only total your faucet cares about.

Keep reading

My Electric Bill Doubled and I Can't Move. Here's the First Lever I Pulled.

Retired, bill doubled from the data center boom, stuck in your house? The cheapest real lever isn't using less power. It's a phone call. Here's what I found.

A packed room made the developer walk — so why did the people who won tell everyone not to cheer yet?

Does showing up to a town hall actually kill a data center, or just push the vote to a quieter night? I dug into what really happens after the room clears.

To Escape Data-Center Rate Hikes With Solar — Is Rooftop a Trap and Ground-Mount the Smarter Buy?

Data-center rate hikes push homeowners to solar. Is rooftop a trap and ground-mount smarter? What leaks, insurance, and resale do to the real math.

They Already Voted Yes on the Data Center — So Why Are There Still Two Decisions Left?

The data center vote already passed? It's not the last decision. Here's the air permit and utility rate case you can still weigh in on.

Why Don't Data Centers Cool With Treated Sewage Instead of Your Drinking Water?

A lot of data centers already cool with treated wastewater. The ones that don't usually can't — and the reason comes down to a purple pipe.

If the Ocean's Right There, Why Cool Data Centers With Drinking Water?

The ocean's right there — so why cool data centers with drinking water instead of seawater? A few already do. Here's why it's rarer than you'd think.

Sources

Bloomberg — About two-thirds of the data centers built or in development since 2022 are in places already gripped by high levels of water stress; five states account for 72% of new centers in high-stress areas.
The Texas Tribune — A developer (Fermi America, co-founded by former Gov. Rick Perry) asked Amarillo for 2.5 million gallons of water per day, with talk of raising it to 10 million, while the mayor said the city uses about 50 million gallons per day; part of Amarillo's water comes from the Ogallala Aquifer, which is being drained faster than it recharges, and Texas does not require most data centers to report water use.
Environmental and Energy Study Institute (EESI) — Indirect data center water consumption through electricity generation was about 211 billion gallons in 2023 (per the LBNL 2024 U.S. Data Center Energy Usage Report); a 100-word AI prompt can use roughly 519 milliliters of water (University of California, Riverside, via The Washington Post).
Bloomberg Law (citing Ceres) — Electric generators powering data centers in seven states used more water than all the cities in California; California's generators withdrew about 1.4 trillion gallons and Virginia's about 753 billion gallons in 2024, and 66% of water-using plants in those states faced medium-to-high water stress.

Byte Bungalow

I do the homework out loud.

If this is the kind of thing you want checked against the documents instead of the hype, the video's above — and there's a new one every week.

Subscribe on YouTube →

By Byte Bungalow. Home power and home tech, checked against the documents instead of the hype. Independent commentary; not affiliated with any manufacturer, utility, or builder named here. Not professional electrical advice.