The AI Water Problem: How Data Centers Use Water

AI is driving compute, and the heat that comes with it, to new highs. That makes water one of the industry’s hardest constraints. The AI water problem is real, but it is often described with one big number that hides how different one facility can be from another.

U.S. data centers directly consumed 17.4 billion gallons of water in 2023, according to EPRI, citing the 2024 Lawrence Berkeley National Laboratory report. At the facility level, cooling design, climate, load, utilization, and measurement boundaries change the answer. That is worth understanding in detail.

How much water does a data center actually use?

The water isn’t going into the servers. It is going into cooling. Many large facilities reject heat with evaporative cooling towers, which work by letting a portion of the water evaporate. That evaporation is what makes the cooling efficient, but it also means a continuous draw of makeup water.

There is no reliable universal gallons-per-megawatt number. EPRI estimates about 340,000 gallons per day for one 150 MW hyperscale scenario and about 2.7 million gallons per day for a water-cooled chiller scenario at the same load. Climate, utilization, cooling design, and Water Usage Effectiveness can change the result substantially.

For sourced national benchmarks, facility scenarios, and the WUE formula, read the data center water consumption guide.

Why cooling towers discharge blowdown

Evaporation leaves dissolved minerals behind. As the water cycles through the tower, those minerals concentrate, and left unchecked they can scale up equipment. To keep cycles of concentration within operating limits, operators deliberately discharge a portion of the liquid loop, called blowdown, and replace it with fresh makeup water.

The amount of blowdown is site-specific. It depends on makeup-water chemistry, treatment, controls, cycles of concentration, and operating conditions. It is not waste in the sense of a leak. It is an engineered discharge. But it is still water bought, used to carry minerals away, and then sent to the sewer.

Why the obvious fixes don’t scale everywhere

The two reflexive answers both run into walls:

“Just stop using evaporative cooling.” Dry and hybrid systems can reduce site water use, but the energy, footprint, cost, and climate tradeoffs vary. EPRI says water-cooled systems can be highly energy efficient while using more water on site, while air-cooled systems can reduce site water and require more electricity. Operators are not using evaporative cooling by accident.
“Just use reclaimed water.” Treated wastewater can reduce reliance on potable supply where source volume, infrastructure, quality, and treatment needs fit the site. It has geographic and operating limits. The supply is not automatically available where data center demand is concentrated.

Chasing new water supply can become a treadmill. The more compute you add, the more makeup water the site may need, and the harder each incremental gallon can be to source.

The fix: recover the water you already have

For the operating sequence, chemistry questions, and treatment choices, use the cooling tower blowdown recovery guide.

There’s another option: recover cooling tower blowdown on site instead of dumping it. If dissolved minerals can be continuously removed from that liquid stream, more water can return to the loop instead of leaving as discharge.

That’s what NONA RECOVER does. RECOVER uses Ion Concentration Polarization (ICP), an MIT-patented alternative to reverse osmosis, to continuously demineralize cooling water and recover 90% of cooling tower blowdown on site. It works alongside the cooling towers already on site. It treats liquid blowdown, not water already lost through evaporation.

ICP was developed at MIT, its development was supported by the U.S. Army for field water production, and RECOVER has been demonstrated with Tracy Renewable Energy.

What this means for site selection

Water is quietly becoming a gating factor for where data centers can be built. Once a project can recover more of its cooling tower blowdown on site, the calculus can change. Circularity beats scarcity. The facility can reduce its dependence on new makeup water and send less liquid blowdown to the sewer.

That does not make every site a fit, and it does not replace measured operating data. Start with cooling load, makeup water, blowdown, cycles of concentration, water chemistry, water and sewer rates, and local capacity constraints.

The AI water problem is real and growing. But the answer isn’t to abandon the cooling that works. It is to understand the whole system and stop throwing away the water that can be recovered.

Want to see the numbers for your site? Calculate your savings →

Technologies