Artificial Intelligence Every time you use ChatGPT to write a 100-word email, approximately 519 milliliters of water are consumed, about the size of a regular water bottle. This figure comes from a 2025 peer-reviewed paper, published in Communications of the ACM, by Pengfei Li, Shuli Ren, and their colleagues at the University of California, Riverside.
It represents both the direct water used to cool the data center servers and the indirect water needed to generate the electricity those servers run on.
Scale that to the millions of users who have dozens of exchanges daily, and the numbers become staggering. By 2027, global infrastructure processing AI queries are expected to consume between 4.2 and 6.6 billion cubic meters of water per year, equivalent to half of the UK’s entire annual water withdrawal.
Much of this water is drawn from already dry areas.
Why do AI data centers consume so much water and where does it go?
Data centers generate huge amounts of heat. The chips at the heart of modern high-end AI GPUs, which can each dissipate between 300 and 700 watts, operate under intense load as long as queries persist. The most common way to manage that heat is evaporative cooling: water is pumped through the facility, absorbs heat from the servers, and then some of it is released as water vapor into the atmosphere.
About 80% of the water drawn into an evaporative cooling system is permanently lost to evaporation. The rest of the cycles return, sometimes at higher temperatures and with chemical residues.The latest generation of hyperscale data centers for AI is larger, denser, and more thermally dense than the general-purpose cloud infrastructure built in the 2000s. A large university campus can now consume more water in one day than a city with 10,000 drinking, sanitation, cooking and agricultural spaces combined. The Lawrence Berkeley National Laboratory’s 2024 U.S. Data Center Energy Use Report, produced for the U.S. Department of Energy, estimates that data centers consumed approximately 17.4 billion gallons of water directly through cooling in 2023 with an additional 211 billion gallons consumed indirectly through electricity generation to power the same facilities.
Data center load growth has tripled over the past decade, and is expected to double or triple again by 2028.
Google , Microsoft and dead : What the numbers actually show
The largest technology companies have begun disclosing water consumption numbers in annual sustainability reports, and the trend remains consistent across all of them.Google’s 2024 Environmental Report put total water consumption for the year at 8.1 billion gallons, with about 95 percent used in data centers.
This number represents an 8 percent increase from 2023, which itself was a 17 percent increase from 2022, and 2022 represented a 20 percent increase from 2021. In three years, Google’s water consumption has nearly doubled, with the company naming AI workload growth as a key driver in successive reports.Microsoft’s consumption numbers are smaller in size but similar in shape. The company announced production of nearly 1.7 billion gallons in 2022, an increase of 34 percent year over year.
Independent reports on the Microsoft data center cluster in West Des Moines, Iowa, where GPT-4 training operations were conducted in 2022, documented that a single training operation consumed 11.5 million gallons of water in July 2022 alone, and 13.4 million gallons in August.
This same group has since expanded to include five facilities, drawing 68.5 million gallons annually from the local municipal water system. Meta consumed approximately 813 million gallons globally in 2023. Amazon, which operates the world’s largest cloud infrastructure, does not publish total water consumption figures.
AI is being built in the world’s most water-stressed regions
Lee Wren’s study predicts that by 2027, global demand for AI could represent water withdrawals equivalent to more than four Danish countries, or close to half of the UK’s total annual withdrawals. The problem is not just the size, but where that size comes from.Microsoft acknowledged in its 2023 Sustainability Report that nearly 42 percent of its water consumption that year came from areas classified as “water-deficient” under the World Resources Institute’s classification system.
Google’s equivalent figure for 2023 was 15 percent of freshwater withdrawals from areas with high water scarcity.The consequences on the ground are already clear. And in Chile, Google temporarily halted construction of a planned $200 million data center near Santiago after an environmental court ruled that the company did not adequately take into account the impact on Santiago’s central aquifer in a country that has been suffering from persistent drought for fifteen years and began rationing residential water in 2022.
In Queretaro, Mexico, where 32 new data centers are planned, the state suffered the worst drought in a century in 2024.
Microsoft has acquired the rights to approximately 25 million liters of water per year from a local aquifer that currently has an annual deficit of 60 million litres. In Arizona, a $14 billion data center project was withdrawn in 2024 after local residents successfully opposed the rezoning.
What’s not revealed and why it matters
The above numbers are what companies have chosen to make public.
The true water footprint of the AI industry, by all available assessments, is much larger.There are three disclosure gaps that keep recurring. The first is the difference between water withdrawal and water consumption, i.e. the volume permanently lost to evaporation versus the volume returned to local systems. Most reports only list one number, and choosing between them can change the apparent footprint by a factor of three or more.
The second is the gap between direct cooling water and indirect electricity generation water, a number that Li and Ren’s research estimates is about twelve times the direct number, a number that is not included in almost any corporate report.
The third is the absence of facility-level data: annual company-level totals tell communities nothing about whether their aquifer is under pressure.The primary contribution of the UC Riverside study is that it produces reliable estimates of these gaps using publicly available proxies.
The numbers that the AI industry has refused to publish are increasingly numbers that independent academic researchers can now estimate within reasonable limits, making avoiding voluntary disclosure more difficult over time.
The barter industry has not yet answered
The global AI infrastructure is being built faster than any similar technology in modern history. The physical buildings produced by the trillion-dollar investment are, at their most basic level, large-scale evaporative cooling systems with computing equipment inside them.Every little query. The whole is not like that. Whether the technologies being developed within these facilities, such as better climate modelling, more efficient irrigation, and more accurate drought forecasting, will begin to contribute to large-scale solutions faster than water consumption accelerates is an open question that will determine the actual environmental legacy of this moment. On the current course, this question remains unanswered.
