AI Is Draining Water From Areas That Need It Most

Bloomberg

08-05-2025

By Michelle Ma Dina Bass
Each time you ask an AI chatbot to summarize a lengthy legal document or conjure up a cartoon squirrel wearing glasses, it sends a request to a data center and strains an increasingly scarce resource: water.
The data centers that power artificial intelligence consume immense amounts of water to cool hot servers and, indirectly, from the electricity needed to run these facilities.
Even before ChatGPT launched in late 2022, communities complained about data centers guzzling up millions of gallons of water every day from cities that didn't have all that much to spare. The problem has only deepened in the years since ChatGPT kicked off an AI frenzy.
More than 160 new AI data centers have sprung up across the US in the past three years in places with high competition for scarce water resources, according to a Bloomberg News analysis of data from World Resources Institute, a nonprofit research organization, and market intelligence firm DC Byte. That's a 70% increase from the prior three-year period.
As Data Centers Become More Ubiquitous, Water-Stressed Areas See the Most Growth
Number of facilities built in the US by year and water stress zone
4008001,2001,600 data centersRelease of ChatGPT
A similar trend is unfolding in other countries. Arid regions like Saudi Arabia and the United Arab Emirates are welcoming more data centers than ever before to claim a piece of the AI boom. Meanwhile, in China and India, an even greater proportion of data centers are located in drier areas compared to the US, according to Bloomberg's analysis. That reflects both their growing data center footprint and heightened water scarcity issues caused by climate change.
As tech firms have pushed to develop new data centers to support cutting-edge AI systems, they've increasingly turned to states and countries with ample energy resources and favorable regulations, according to experts. What those places often lack, however, is an abundant supply of water. The result is that data centers threaten local water supplies, agriculture and energy production.
'It is very much a growing issue — and it is spreading everywhere,' said Newsha Ajami, chief strategy and development officer for research at Lawrence Berkeley National Laboratory and the founding director of the urban water policy program at Stanford University.
In recent years, there have been water-related protests against data centers in the Netherlands, Uruguay and Chile, the last of which temporarily revoked Google's authorization for a planned $200 million facility after people took to the streets. In the US, tech firms have opened more large data centers in drought-prone states like Arizona and Texas, adding to concerns about the local water supply.
Data Centers Proliferating Globally in Water-Stressed Areas
Number of facilities built in countries with the most data centers in high water risk zones
High to extremely high
Medium
Low
China'defined'defined1002003000
Germany0
France0
UK0
India0
Canada0
Australia0
Italy0
Spain0
Netherlands0
Russia0
Indonesia0
SaudiArabia0
Thailand0
UAE0
Chile0
Mexico0
Belgium0
Turkey0
SouthAfrica0
'Every part of the state is facing this water-energy nexus crisis,' said Amy Bush, a hydrologist for RMBJ Geo Inc. based in Abilene, the small Texas city where OpenAI is setting up a massive 1.2-gigawatt data center campus to anchor its $100 billion Stargate AI infrastructure venture.
Water is often the last consideration when making siting decisions for data centers because it's cheap compared to the cost of real estate and power, said Sharlene Leurig, a managing member of Fluid Advisors, a water consulting firm. 'Water has traditionally been far less consequential for industrial users than energy costs and availability,' she said.
At least one large tech company, Microsoft Corp., is rethinking the pace of data center expansion and pulling back on certain projects. But the number of data centers in dry areas is still poised to increase significantly in the coming years. In the US alone, there are at least 59 additional facilities planned in such regions by 2028, according to DC Byte, with more new projects seemingly getting announced every couple weeks.
A Growing Problem
More AI means more water. In the US, an average 100-megawatt data center, which uses more power than 75,000 homes combined, also consumes about 2 million liters of water per day, according to an April report on energy and AI from the International Energy Agency (IEA). That's equivalent to the water consumption of about 6,500 households, the report said.
Globally, the report estimates data centers consume about 560 billion liters of water annually and that could rise to about 1,200 billion liters by 2030, as tech firms push for bigger facilities stocked with more advanced AI computing chips that run hot.
Many data centers rely on evaporative cooling, or 'swamp cooling,' where warm air is drawn through wet pads. Data centers typically evaporate about 80% of the water they draw, discharging 20% back to a wastewater treatment facility, according to Shaolei Ren, an associate professor of electrical and computer engineering at the University of California, Riverside. Residential water usage, by comparison, loses just 10% to evaporation, discharging the other 90%, Ren said. (A spokesperson for Google said the company doesn't have a standard percentage because any data center would see some variation based on factors like location, temperature and humidity.)
Read More of Bloomberg's coverage on data centers:
AI Is Already Wreaking Havoc on Global Power Systems
AI Needs So Much Power, It's Making Yours Worse
The water footprint of data centers extends beyond cooling. A 2021 paper found nearly half of US data center servers are fully or partially powered by water-hungry power plants located within water-stressed regions. The IEA estimates 60% of data center water consumption is from indirect use.
The surging water demand now threatens two key environmental priorities for tech firms. Before the AI boom, the biggest US cloud computing companies — Amazon.com Inc., Microsoft and Google — pledged to cut their carbon footprint in the coming years and also be water positive by 2030, meaning they would add more water to the environment than they use. However, the regions with the most available renewable energy resources to support data centers — particularly solar — are sometimes the ones with the least water. Conversely, data centers set up to use less water in hotter regions end up requiring more power to run.
'When water has this power of reducing energy consumption and energy-related emissions like carbon, then we have to be able to discuss the role that it could, should, or should not play in different applications,' Ben Townsend, global head of infrastructure and sustainability at Google, said in an interview.
What Water-Stressed Sites for Data Centers Look Like
Satellite imagery showing the surrounding areas of planned and existing data centers in a selection of places where water stress is high or extremely high
Searching for a Solution
Tech firms have faced these opposing priorities before. Early in the history of data centers, companies relied on air conditioning to cool equipment. These systems used a lot of electricity, causing companies to look to water to create more energy-efficient cooling technology, said Steve Solomon, vice president of datacenter infrastructure engineering at Microsoft.
Now tech companies are trying new solutions, including data center and chip designs that let them use less water. Some are placing hot chips directly on cold plates that use water or else submerging chips and servers in liquid, a process known as immersion cooling. Businesses are also experimenting with synthetic liquids to cool data centers. But some coolants are being phased out from the market because they use so-called forever chemicals, which don't naturally break down and can persist in animals, people and the environment.
Recently, Microsoft said it developed a data center design that is closed so that water doesn't evaporate but rather is constantly circulated between servers and chillers, without the need for refilling. The design will be deployed first in facilities in Wisconsin and Arizona, planned for 2026.
Crusoe Energy Systems, a developer behind OpenAI's Stargate site in Abilene, also plans to use closed-loop cooling systems. But here, too, 'there is a tradeoff in energy,' said Ben Kortlang, a partner at G2 Venture Partners, an investor in Crusoe. These systems are more power-hungry than evaporative methods, he said.
For all the research into alternative approaches, a majority of AI-specialized data centers used evaporation-based cooling systems either 24/7 or during hot days in 2023, with more data centers expected to use water evaporation across the entire industry by 2028, according to a Lawrence Berkeley National Laboratory report released late last year. Amazon Web Services, for example, relies widely on direct evaporative cooling, the company said.
AWS does turn to different cooling methods based on the time of the year and uses sensors to react to changing weather. To minimize impacts on drinking water, Amazon's cloud unit cools 24 of its data centers using treated sewage. The company said it is hoping to dramatically expand that in coming years by working with local utilities to access the recycled water it needs.
The cloud computing firms also work with local nonprofits and officials on watershed restoration, replenishment and other projects to return water.
Lack of Transparency
As Silicon Valley mulls solutions, water advocates say tech companies need to be more transparent about the problem. Almost no information about data center water usage on an individual system level is publicly available.
In one case, The Dalles, a city in Oregon, sued Oregonian Media Group to prevent the release of Google's water usage records, arguing the information was a trade secret. The city agreed to disclose those records after a 13-month legal battle. (Google's Townsend said the company contributed to the available water in The Dalles by developing water resources at the site that were previously unavailable and giving some of that back to the community to 'increase the value available to everyone.')
Jennifer Walker, director of the Texas Coast and Water Program at the National Wildlife Federation, also said state officials need more information for water planning. But when the Texas Water Development Board sent a water use survey to data centers, it received a lackluster response, she said.
'We just had one of the hottest summers on record in Texas, and we've had several of those,' she said. 'I'm concerned about any super water-intensive industry that is going to come into our state.'
Related tickers:
1554630D:US (OpenAI Inc.)
AMZN:US (Amazon.com Inc.)
MSFT:US (Microsoft)
GOOGL:US (Alphabet Inc.)
Edited by Seth Fiegerman Chloe Whiteaker Brian Eckhouse With assistance from Rachel Metz Brody Ford Jeremy Diamond Leon Yin
Methodology
Bloomberg analyzed "Baseline Water Stress" data from the World Resources Institute's Water Risk Atlas. Baseline water stress is recorded in five categories: "low", "low - medium", "medium - high", "high", "extremely high", "arid and low water use". For the purposes of this analysis, Bloomberg combined "low-medium" and "medium-high" into one category ("medium"), and combined "high", "extremely high" and "arid and low water use" into one category ("high to extremely high").
Bloomberg analyzed data center locations and capacity — including active facilities, planned facilities, and those under construction — compiled by DCByte, a market intelligence firm. In order to account for data center that may be located at the edge of multiple water stress areas with different risk levels, Bloomberg used a CDKTree algorithm to find all of the water stress areas within a 30 mile radius of each data center location. For each data center, Bloomberg assigned the maximum water stress score found within this search radius. Bloomberg calculated the total number of data centers that fell within each water stress category.