How Much Water Does AI Use and Why?

local AI writing tool, private AI writing tool

Laptop writing workspace for local AI tools

AI's environmental footprint isn't just about electricity. Every data center running AI models also needs something less obvious: water, lots of it, to keep the servers from overheating.

Why Water Is Part of the Equation at All

AI chips run hot, and the servers housing them need to stay within a specific temperature range to function. Many data centers handle this with evaporative cooling: water absorbs heat from the servers, turns to vapor, and is vented out, the same basic principle as sweating (MOST Policy Initiative). It's effective, but it permanently uses up the water in the process, unlike air-based cooling systems, which use electricity instead of water.

The Numbers, Straight from the Companies Themselves

Google's own 2025 Environmental Report discloses that its data centers consumed about 8.1 billion gallons of water in 2024, a 28% increase from the year before. The company itself frames that figure as equivalent to the annual watering needs of 54 golf courses in the arid U.S. Southwest (RCR Wireless News, reporting on Google's 2025 Environmental Report).

That's just one company. Looking at the U.S. as a whole, a 2024 Lawrence Berkeley National Laboratory report estimated that data centers in the country directly consumed about 17 billion gallons of water through cooling in 2023, with that figure projected to double or even quadruple by 2028 (The Current).

The Hidden Number: Indirect Water Use Is Even Bigger

Direct cooling water is actually the smaller part of the story. The same Lawrence Berkeley National Laboratory analysis estimated that U.S. data centers used an additional 211 billion gallons of water indirectly in 2023, water consumed not by the data center itself, but by the power plants generating the electricity that runs it (The Current). That indirect figure was roughly 12 times larger than the direct cooling water use the same year.

This is an important distinction, because most public conversation focuses on the visible, direct number, while the larger, less visible number rarely gets mentioned at all.

Why It's Hard to Get a Clear Picture

A genuine obstacle here is transparency. Sustainability reports from major tech companies are voluntary, and they don't follow a consistent format. A research team from the University of Wisconsin-Milwaukee that reviewed water disclosures from six major tech companies found that the reports varied widely in both the amount of water used and how much detail companies actually provided, making it difficult to compare companies or track trends over time with any precision (The Current).

Individual facility numbers, when they are disclosed, also vary enormously based on location and cooling method. Google has reported some of its air-cooled data centers using as little as 10,000 gallons a year, while some of its more water-intensive facilities consume well over a million gallons (The Current).

What Companies Say They're Doing About It

To their credit, several major tech companies have made public water-positive commitments. Google reports that its water stewardship projects replenished roughly 4.5 billion gallons in 2024, about 64% of what its data centers consumed, with a stated goal of replenishing 120% of freshwater consumption by 2030 (Data Centre Magazine). Microsoft has made a similar water-positive pledge for 2030 and is deploying closed-loop cooling systems in some new facilities that lose far less water to evaporation (TechTarget).

The Bottom Line

AI's water footprint is real, growing, and only partially visible in public reporting. The direct cooling water that gets most of the attention is, by most estimates, the smaller piece of a much larger indirect footprint tied to electricity generation. Both numbers are worth knowing if you want an honest picture of what running AI at scale actually costs.

Sources

My Private AI

Take control of your security content.

© Copyright 2026 My Private AI - All rights reserved.