This is the fourth article in a series exploring the facts behind the data center debate. You can read parts one, two, and three, or download our full eBook, The Truth About Data Centers, for a closer look at the issues covered in this series.

Water consumption is one of the most emotionally charged concerns raised about data centers today, and it’s one of the most misunderstood. A single statistic can circulate for months before anyone checks where it came from, and by then it may have already shaped local debates and headlines.

The picture looks very different once you look at how these figures actually get built.

Cooling Isn’t One Thing

Lumping every data center into the same water conversation is where most confusion starts. There are three completely different approaches to cooling, and each comes with its own tradeoffs.

1. Air-based cooling uses no water at all, relying entirely on traditional air conditioning. It’s common in older and smaller facilities, although it becomes less practical as computing density rises.
2. Evaporative cooling trades water for efficiency. It’s a continuous process: Water goes in and doesn’t come back, and many large legacy facilities run this way because the energy savings are real.
3. Closed-loop cooling sidesteps that tradeoff entirely. Water (or in some systems, a glycol-based fluid) circulates inside a sealed system rather than getting released, so once the system is charged, ongoing consumption approaches zero.

That distinction is measured through Water Usage Effectiveness, or WUE, the industry’s standard for tracking how much water a facility uses relative to its size and output. DataBank’s standard facility design runs on closed-loop cooling, so new builds start service with a near-zero water footprint from day one, not something the facility works toward over time.

One Prompt, One Bottle? Not Quite.

A claim that keeps resurfacing is that a single ChatGPT prompt costs about a bottle of water. It traces back to a 2024 Washington Post estimate, and independent researcher Andy Masley later found the number was likely 50 to 250 times too high.

The inflated version counted water lost to evaporation at hydroelectric reservoirs as part of a data center’s footprint, water that evaporates off dammed rivers regardless of who’s drawing power from the dam. It’s a bit like arguing that a retail store uses water every time its cash register rings up a sale, since generating the electricity behind that transaction required water somewhere upstream.

Strip that out and the numbers change dramatically: Masley’s corrected estimate for onsite water use lands around 2.2 milliliters per prompt across the U.S. grid, or about 1/227th of a standard bottle. It’s the same underlying data, but just with a different definition of what actually counts as data center water use.

Real Numbers, Real Context

Compare data center water use against everyday consumption and even the larger figures shrink fast.

• California’s almond industry consumes over 1.3 trillion gallons of water a year. By comparison, all the data centers in the U.S. combined consume less than 1.5% of that amount.
• S. golf courses used roughly 531 billion gallons in 2024, more than 30 times the direct cooling water of every U.S. data center combined.
• Leaky municipal pipes lose an estimated 2,500 billion gallons a year before that water ever reaches anyone.
• A study from Arizona State University’s Kyl Center for Water Policy found that data centers in Arizona use less water than power generation, semiconductor manufacturing, sand and gravel mining, golf courses, and beverage production in the same state.
• Virginia’s own legislative audit commission found that 83% of the state’s data centers, home to the largest data center market in the world, use no more water than a large office building.

Every comparison points the same direction: Data centers are a small, and shrinking, line item in the country’s water use.

The Industry is Moving in the Right Direction

Cooling technology is shifting fast, and the water story is shifting with it. About 22% of data centers worldwide currently use liquid cooling technologies, a figure expected to approach 50% of new hyperscale builds within two to three years, driven largely by the thermal demands of AI workloads. Major operators have made firm commitments in this direction, including Microsoft’s announcement of zero-water closed-loop cooling for all future builds.

A less-discussed piece of this story is the power behind the water. Fossil fuel generation is water-intensive; yet, wind and solar are not. As the grid shifts toward renewables, a data center’s indirect water footprint shrinks along with its carbon footprint, even before accounting for cooling design. That’s a benefit that rarely makes it into the figures critics cite, even though it compounds every year renewable capacity grows.

Drought-prone communities are right to ask hard questions here, and those concerns reasonably extend beyond onsite cooling to the water used generating the electricity in the first place. The honest answer is that the industry isn’t moving at a uniform pace, which is exactly why specifics matter more than pledges.

The Bottom Line

Water figures that sound alarming often turn out to be measuring the wrong thing, not the wrong facility. Once the boundaries get examined, a modern, closed-loop data center tends to look a lot more like an office building than an industrial plant.

Data centers are essential infrastructure, and concerns about their footprint deserve real answers, not headlines that never checked their own math.

Want to stay ahead of the trends shaping digital infrastructure? DataBank Digest delivers analysis, insights, and the latest news from the world of colocation and digital infrastructure.