The Truth About Water Use in Modern AI Data Centers

The Concern

One of the biggest concerns I see raised by those opposing the construction of AI data centers is that of water.

Some claim that they use millions of gallons a year.

AI CEOs claim the water use is almost nonexistent.

So, should we, as citizens, be worried about water consumption from AI data centers?

Where do the claims come from that data centers use a lot of water?

Cooling Is the Key Distinction

First, data centers use water to cool the chips. But how they accomplish that cooling is important.

Evaporative Cooling

Many early data centers used what is called evaporative cooling. This is where water is used to cool the data center in a way that it then evaporates into the ecosystem, usually by spraying the water to cool the air or the radiators. Evaporative cooling requires a constant source of water and should only be used in areas where water is abundant.

But not everywhere has an abundance of water. Even in places that do, having to continually pay for water consumption is a burden on the companies that manage these data centers.

Closed-Loop Cooling

The good thing is, there are other options.

In fact, the newest AI-focused facilities are increasingly moving toward closed-loop, direct-to-chip, non-evaporative cooling.

Closed-loop dry cooling still uses water. Instead of cooling the air or cooling towers with water, the water is pumped to the chips themselves. The water acts as a heat sink, absorbing the heat from the chip and is pumped away from the chip so there's always a fresh supply of cool water at the chip. The warm water is pushed through radiators which are then cooled by fans. The cool water from the radiators is then pumped back to the chips.

In normal operation, the same coolant is reused again and again. That means little to no routine cooling-water consumption.

Initial Fill Is Not Annual Consumption

The only water usage here is the initial filling of the system and incidental needs of the data center, like bathrooms and employee break rooms. Yes, there are maintenance needs as well, but there is no permanent water draw to meet its cooling needs.

This is the current standard for data center development. One initial water draw followed by negligible water usage for basic employee needs and maintenance.

If you live in an area like Utah where water is not abundant and the data center proposal does not state it is using closed-loop cooling, then your concerns are warranted and you should use your voice to affect change here so that the data center makes responsible use of the water. But, if the plan includes closed-loop cooling, then the water usage concerns are largely unwarranted.

The Data

While that's nice and all, let's look at the math.

Understanding Acre-Feet

Before we do, it's important to understand the units.

Water usage is often measured in acre-feet. One acre-foot is about 325,850 gallons. That number is useful for math, but it's not a helpful visual. So think of it this way: one acre-foot is enough water to cover an acre of land with one foot of water. Picture something close to a football field, endzone to endzone, covered with water one foot deep. Another way to think about it is that one acre-foot is enough to fill roughly 16 typical backyard swimming pools.

Now let's compare a few land uses.

Putting the Numbers in Perspective

Agriculture

In Utah, irrigated agriculture can require multiple acre-feet of water per acre per year. Alfalfa is a good example. If 100 acres of alfalfa requires around 4 acre-feet per acre in a season, that is roughly 400 acre-feet of water in a year. That's 130,340,571 gallons or roughly 6,500 swimming pools. That's some serious water usage!

Golf Courses

Golf is another helpful comparison because courses are very visible. Utah golf courses used tens of thousands of acre-feet of water in 2022. One course may use more than 100 acre-feet per year, and some use much more depending on size, climate, turf area, and irrigation practices. To put that in perspective, 100 acre-feet of water is 32,585,142 gallons or over 1,600 swimming pools.

Evaporative Data Centers

Evaporative-cooled data centers belong in that conversation. A large data center using evaporative cooling can consume hundreds of acre-feet of water per year. That is real water consumption. In a desert climate, that deserves scrutiny.

Closed-Loop Data Centers

But that is not the same thing as closed-loop dry cooling.

A closed-loop system may need a meaningful initial fill. The pipes, pumps, heat exchangers, and cold plates have to be charged with coolant. For a large facility, that initial fill might be measured in tens or hundreds of thousands of gallons. For a very large campus, especially one built in phases, the total could be larger.

But that is the key difference: initial fill is not the same thing as annual consumption.

An evaporative system needs a continuing water supply because water is consumed as part of the cooling process. A closed-loop dry-cooled system recirculates the same coolant over and over. There may still be water used for commissioning, maintenance, restrooms, break rooms, landscaping, fire systems, or other site needs. But the cooling system itself is not consuming water year after year in the same way.

That distinction matters.

A data center that consumes hundreds of acre-feet every year through evaporative cooling is a legitimate concern in Utah.

A data center that uses closed-loop dry cooling, has a one-time system fill, and then has very low ongoing water consumption is a different conversation.

The Questions Citizens Should Ask

That does not mean citizens should stop asking questions. They should ask better questions.

• What cooling system is being used?
• Is the commitment binding?
• Where is the water coming from?
• Is it culinary water, agricultural water, reclaimed water, or brackish groundwater?
• How much water is needed for construction and commissioning?
• How much water is needed during normal operation?
• What happens during heat waves?
• Is there an on-site power plant, and does that plant require water?
• What monitoring will be public?
• What happens if the system does not perform as promised?

Those are the questions that matter.

So when someone says, "AI data centers use millions of gallons of water," the right response is not to blindly accept or reject the claim. The right response is, "What kind of cooling system are we talking about?"

Because the difference between evaporative cooling and closed-loop dry cooling is the difference between a facility that has an ongoing water draw every year and a facility that mostly just contains water.

Conclusion

Putting all this another way, the data center water problem is not imaginary, but it is technically solvable.

If a company wants to build a hyperscale AI data center in an arid place like Utah, closed-loop dry cooling should not be treated as a bonus feature or a nice sustainability talking point. It should be the baseline expectation.

We should still be vigilant as citizens. We should still demand clear numbers, binding commitments, public monitoring, and serious community benefits. But we should also be precise. A data center that relies on evaporative cooling is not the same thing as a data center that uses closed-loop dry cooling. Treating them as the same makes the public debate worse, not better.

And yet, even with all that, I agree with Nathaniel Whittemore when he said, "If you don't like what the water's used for, even one gallon is too many." For some people, the actual water number will never be the deciding issue. If they see AI data centers as a threat, then any water use will feel unacceptable.

But I also agree with Anne Davis Vaughn when she said, "Two things can be true: data centers have drawbacks, but communities can win far more than they realize."

That is where I think the real civic opportunity is. Communities should not simply ask, "Should we allow data centers?" They should ask, "If these are coming, what standards do we require, and what do our communities get in return?"

Closed-loop cooling. Responsible power. Real infrastructure investment. Roads. Schools. Parks. Trails. Public transparency. Long-term accountability.

Those are the terms communities should be negotiating.

The water use of modern AI data centers can be very low if the right engineering choices are required from the beginning. But "can be" is doing important work there. Citizens should not accept vague promises. They should require the design, the numbers, and the commitments to match the ecology of the place where the data center is being built.