AI’s Every time you use ChatGPT to write a 100-word email, roughly 519 millilitres of water is consumed, nearly the volume of a standard water bottle. That figure comes from a peer-reviewed 2025 paper published in Communications of the ACM by Pengfei Li, Shaolei Ren, and colleagues at the University of California, Riverside. It accounts for both the direct water used to cool data centre servers and the indirect water required to generate the electricity those servers run on. Scale that to millions of users having dozens of exchanges daily, and the numbers become staggering. By 2027, the global infrastructure processing AI queries is projected to consume between 4.2 and 6.6 billion cubic metres of water annually the equivalent of half the United Kingdom’s entire yearly water withdrawal. Much of that water is being drawn from regions already running dry.
Why AI data centres consume so much water and where it goes
Data centres generate enormous quantities of heat. The chips at the core of modern AI high-end graphics processing units that can each dissipate between 300 and 700 watts run at intense load for as long as queries keep coming in. The most common method for managing that heat is evaporative cooling: water is pumped through the facility, absorbs heat from the servers, and a portion of it is then released as water vapour into the atmosphere. Around 80 per cent of the water drawn into an evaporative cooling system is permanently lost to evaporation. The rest cycles back, sometimes at higher temperatures and with chemical residues.The newer generation of AI-specific hyperscale data centres are larger, denser, and more thermally intense than the general-purpose cloud infrastructure built in the 2010s. A single large campus can now consume more water in a day than a town of 10,000 uses for drinking, sanitation, cooking, and agriculture combined. The 2024 US Data Centre Energy Usage Report by Lawrence Berkeley National Laboratory, produced for the US Department of Energy, estimated that data centres 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 centre load growth has tripled over the past decade and is projected to double or triple again by 2028.
Google, Microsoft, and Meta: What the numbers actually show
The biggest technology companies have begun disclosing water consumption figures in annual sustainability reports, and the trajectory is consistent across all of them.Google’s 2024 Environmental Report set out total water consumption for the year at approximately 8.1 billion gallons, with around 95 per cent used at data centres. That figure represented an 8 per cent increase on 2023, which had itself been a 17 per cent increase on 2022, and 2022 had been a 20 per cent increase on 2021. In three years, Google’s water consumption nearly doubled, with the company naming AI workload growth as the primary driver in successive reports.Microsoft’s consumption figures are smaller in scale but similar in shape. The company reported approximately 1.7 billion gallons in 2022, a 34 per cent year-on-year increase. Independent reporting on Microsoft’s data centre cluster in West Des Moines, Iowa, where GPT-4 training runs were conducted in 2022, documented that a single training run consumed 11.5 million gallons of water in July 2022 alone, and 13.4 million gallons in August. That same cluster has since expanded to 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 aggregate water consumption figures.
AI is being built in the world’s most water-stressed regions
The Li and Ren paper projects that by 2027, global AI demand could account for water withdrawal equivalent to more than four Denmarks, or approaching half the United Kingdom’s total annual withdrawal. The problem is not just the volume it is where that volume is coming from.Microsoft acknowledged in its 2023 sustainability report that approximately 42 per cent of its water consumption that year came from regions classified as “water-stressed” under the World Resources Institute’s rating system. Google’s equivalent figure for 2023 was 15 per cent of freshwater withdrawals from areas of high water scarcity.The on-the-ground consequences are already visible. In Chile, Google paused a planned $200-million data centre near Santiago after an environmental court ruled the company had not adequately accounted for the impact on the Central Santiago Aquifer in a country that had been in continuous drought for fifteen years and had begun rationing residential water in 2022. In Querétaro, Mexico, where 32 new data centres are planned, the state suffered its worst drought in a century in 2024. Microsoft has secured rights to approximately 25 million litres of water annually from a local aquifer currently running a 60-million-litre annual deficit. In Arizona, a $14-billion data centre project was withdrawn in 2024 after local residents successfully opposed the rezoning.
What is not being disclosed and why it matters
The figures above are what companies have chosen to make public. The true water footprint of the AI industry is, by every available assessment, considerably larger.Three disclosure gaps recur persistently. The first is the difference between water withdrawal and water consumption, the volume permanently lost to evaporation versus the volume returned to local systems. Most reports name only one figure, and the choice between them can shift 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 figure the Li and Ren research estimates to be roughly twelve times the direct figure, and one that almost no corporate report includes. The third is the absence of facility-level data: a company-wide annual total tells local communities nothing about whether their specific aquifer is under pressure.The UC Riverside paper’s core contribution is that it produces credible estimates of these gaps using publicly available proxies. The figures the AI industry has declined to publish are increasingly ones that independent academic researchers can now estimate within reasonable bounds, which makes voluntary disclosure harder to avoid over time.
The trade-off the industry has not yet answered
The global infrastructure for AI is being built faster than any comparable technology build-out in modern history. The physical buildings that the trillion-dollar investment is producing are, at their most basic level, large industrial-scale evaporative cooling systems with computing equipment inside them.Each query is small. The aggregate is not. Whether the technologies being developed inside these facilities, such as better climate modelling, more efficient irrigation, and more accurate drought forecasting, begin contributing to solutions at scale faster than the water consumption accelerates is the open question that will define the actual environmental legacy of this moment. On the present trajectory, that question remains unanswered.