Cloud computing is drying out our rivers, lakes, and oceans. The average 100-megawatt data center consumes about 2 million liters of water a day, equivalent to the water consumption of 6,500 American households. Globally, data centers consume around 560 billion liters of water annually, or the equivalent of 224,000 Olympic-sized swimming pools.
Not only do they consume massive amounts of water, but two-thirds of new data centers built or in development globally since 2022 are located in places already plagued by water stress. Even before the launch of OpenAI’s ChatGPT, communities complained about data centers guzzling millions of gallons of water from cities without much to spare (Bass et al., 2025).
(Barrat & Gambarini, 2025)
How do Artificial Intelligence (AI) data centers work?
Data centers are physical spaces that house computing equipment, like servers, storage systems, and network equipment. These servers store everything from emails to google searches, and this information is distributed via underground cables. Thus, the so-called “cloud” is just a collection of data centers. When information is stored in the cloud, it’s stored in the data centers of the cloud provider (Cisco, n.d.).
While the specific needs of AI data centers depend on the kind of AI that they support, they typically require higher performing infrastructure than non-AI data centers. The advanced storage, networking, energy, and cooling capabilities required to train, use, and deliver AI applications and services also often require AI data centers to occupy significantly more square footage (Gomstyn & Jonker, 2025).
(International Energy Agency, 2025)
Different Types of AI: Generative, Predictive, Extractive, & Agentic
Although we can’t aggregate data centers based on the type of AI they support, it is useful to understand what types of AI exist and what they are able to do.
Generative AI (gen AI):
- generates original content—like video, audio, images, code or text—from a user’s prompt or request
- trained on massive amounts of raw data (McKinsey & Company, 2024).
Predictive AI:
- performs predictive analytics by combining statistical analysis with machine learning to find patterns and make forecasts
- mostly used in business forecasting, management, and detection (IBM, 2024)
Extractive AI:
- retrieves information from documents, tables, charts, and graphs (Dolfi et al., 2024)
- similar to using control-F to find information in a document
Agentic AI:
- the newest form of AI
- autonomously performs tasks on behalf of a user or another system
- can look like autonomous vehicles or virtual assistants (Downie & Finn, n.d.)
This range of powerful capabilities is becoming essential in both public and private use, which is why AI is expected to have a compound annual growth rate of nearly 30% between 2025 and 2030 (Statista, n.d.).
Why Is Water Relevant to Data Centers?
To power these data centers, water is used in two ways:
1) indirectly, to generate the electricity that data centers need to operate
2) directly, to dissipate the heat generated by servers and other equipment
Computer equipment needs to stay cool to function properly. Continued exposure to high temperatures can cause sometimes irreversible loss of functionality. High temperatures can also cause electrical fires.
Direct water consumption depends on a number of factors: the data center’s location, the number and placement of servers, the fire suppression system, the quality of available water, as well as the needs and number of workers in the building. Of course, data centers in hot regions require more cooling.
Cooling techniques vary between data centers, but the most common method is evaporative cooling (also known as swamp cooling), where warm air is drawn through wet pads. Data centers evaporate 80% of the water drawn and discharge 20% to wastewater treatment. Residential evaporative cooling, on the other hand, loses 10% of water to evaporation, discharging the other 90%. The majority of AI specialised data centers use evaporation-based cooling 24/7 or, at minimum, during hot days. Even more data centers are expected to do so by 2028.
It is important to consider where water is being sourced to power data centers. Some use treated wastewater or reclaimed/recycled water. By far the most common, however, is potable water. Servers and other electrical equipment are very sensitive. The quality of the cooling water can affect the equipment’s performance and longevity. For example, reclaimed water can cause more corrosion, scaling, and microbial growth in the equipment than potable water due to its content. Dry areas also reduce the risk of metal corrosion because they have low humidity (Ahmad, 2024).
In terms of indirect water consumption, nearly half of US data centers are fully or partially powered by water-hungry power plants in water-stressed regions. The International Energy Association estimates that 60% of data center water consumption is from indirect use (Bass et al., 2025).
On direct water consumption alone, data centers rank in the top 10 highest water-consuming American industrial or commercial industries. Yet, a 2021 survey found that only 51% of data center operators track their water usage. Out of those who do monitor their usage, monitoring is mainly at individual data centers, not across facilities. Only 10% of data center operators track water use across all their facilities (Ahmad, 2024).
(Bass et al., 2025)
Water as an Afterthought
Across the board, AI data centers require more resources than non-AI data centers: more energy, more land, and more water. Before the widespread use of AI, the biggest American cloud computing companies pledged to cut their carbon footprints in the next few years and be “water positive” by 2030. Being “water positive” means that companies compensate for their water usage by producing high-quality water to return to water systems, while reducing their water consumption (Bass et al., 2025). Google’s data centers, for example, used 6.1 billion gallons of water in 2023 (Google, 2024).
(Google, 2024)
However, there’s a trade-off when locating new data centers: the regions with the most renewable energy usually have the least available water. Hotter climates generally tend to have more availability for renewable energy and less water. Data centers can either consume less water and use more electricity, or use less electricity and consume more water. Traditionally, water has been cheaper and more abundant than energy and land, leading businesses to consider water availability as more of an afterthought (Bass et al., 2025).
Where Are Data Centers Anyway?
Due to the availability of energy and land, data centers are increasingly located in arid regions. In the Middle East, Saudi Arabia and the United Arab Emirates are welcoming more data centers than ever before. Globally, Asia has some of the greatest proportions of data centers located in dry areas, especially in China and India. In Latin America, data centers are particularly common in Chile, Uruguay, and Mexico. The Netherlands, Spain, and Ireland house the most data centers in Europe (Bass et al., 2025).
(Bass et al., 2025)
Case Study: Chile
Latin America and the Middle East account for 5% of global electricity consumption by data centers, which experts highlight as an opportunity to strengthen connectivity, increase employment in technology, and boost the integration of renewable energy into the electricity matrix. The countries with the highest growth potential for AI infrastructure have been identified as: Brazil, Mexico, Chile, Colombia, Panama, Peru, Uruguay, and Costa Rica.
The question of AI and its associated infrastructure as a new form of digital extractivism is central to the conversation around Latin American data centers (Recessary, 2025). Could AI data centers become another technology at the “exclusive service of industrialized countries, at the expense of ecosystems” in the Global South (Recessary, 2025)? Data centers are already putting stress on electricity and water systems in Mexico, Chile, and Uruguay.
(Barrat & Gambarini, 2025)
In 2024, Google paused its plans for a $200 million data center in the Cerrillos area because of environmental concerns. A Chilean environmental court partially reversed Google’s 2020 permit, saying that the company needed to revise their application to reflect the effects of climate change, local pushback, and water concerns. The project would use the Central Santiago Aquifer, the strained aquifer of Chile’s capital. The Chilean government began water rationing in 2022, 15 years into an unprecedented drought (Swinhoe, 2024).
Satellite Image of an Existing Data Center Site in Chile
(Bass et al., 2025)
In March of this year, Google signed four partnerships as part of its water stewardship program, one of which is in Chile. The Chilean partnership is meant to support the rehabilitation of a storage reservoir and the lining of an irrigation canal in the Maipo River basin, as well as improve water security in the semi-arid region (Skidmore, 2025).
Case Study: The United States
The United States has the highest number of data centers in the world. Maricopa County, Arizona is facing extreme drought, according to the National Oceanic and Atmospheric Administration (NOAA). It is also a fast-growing data center hub.
In January 2025, Meta opened a $1 billion data center in Mesa, a city in Maricopa County. Google has one data center in the area and is developing two more. Microsoft has two active data centers in the county. Yet, in June 2023, the state of Arizona revoked construction permits for some new homes in Maricopa County citing a lack of groundwater. Google’s Mesa data center has a permit to use 5.5 million cubic meters of water a year, the same amount used by 23,000 Arizonans. Google is going ahead with plans for a second data center in Mesa, but is not planning to use the full amount of water in the permit as it plans to use an air cooling system (Barrat & Gambarini, 2025).
Satellite Image of an Existing Data Center Site in Arizona
(Bass et al., 2025)
In the U.S., Google is the most likely of the big three tech companies—Microsoft, Amazon, and Google—to build data centers in dry areas. In parts of the US facing water scarcity, they have seven active data centers and are building 3 more.
Case Study: Spain
As one of the main producers of renewable energy in Europe, Spain is a leading destination for data centers of the big three tech companies. By 2030, Madrid forecasts that the sector will be worth 58 billion euros. Renewable energy combined with its underwater cable connections with other countries and thousands of hectares of available land make the country very appealing for data centers (France 24, 2025). At the same time, 75% of Spain is already at risk of desertification.
Satellite Image of an Existing Data Center Site in Aragon, Spain
(Bass et al., 2025)
In the Aragon region of northern Spain, Amazon has three existing data centers and is proposing at least three more. These new data centers are licensed to use an estimated 755,720 cubic meters of water a year, which is enough to irrigate 233 hectares (576 acres) of one of the region’s major crops: corn. However, this figure doesn’t take into account the indirect water usage, meaning that the real figure is much higher (Barrat & Gambarini, 2025).
Aragon houses the biggest data center in Europe at 190 hectares (470 acres) large. 500 million liters (132,086,026 gallons) of drinking water a year are required to cool this one data center (France 24, 2025).
In December of 2024, Amazon asked Aragon’s government to increase its water consumption permit at its three existing data centers by 48% (Barrat & Gambarini, 2025). Amazon’s application stated that “climate change will lead to an increase in global temperatures and the frequency of extreme weather events, including heat waves,” meaning that data centers need increasingly large volumes of water as the planet gets hotter (Barrat & Gambarini, 2025). In March of 2025, Aragon asked for European Union aid to deal with its drought. Groups like Tu Nube Seca Mi Río (“Your cloud is drying my river”) are calling for a moratorium on new data centers in Spain due to water scarcity.
Being “Water Positive”
Amazon claims that it will be water positive by 2030, offsetting its water consumption by providing water to communities and ecosystems in areas of scarcity elsewhere. However, water offsetting does not and cannot work on the same principles as carbon offsetting. Pollutants removed from the atmosphere can cancel out pollutants emitted elsewhere.
Improving access to water somewhere far away does nothing to help the local community that has lost access to its water. This idea has caused significant controversy, including within Amazon itself. Spokespeople for Amazon say the projects are in communities within which the company operates (Barrat & Gambarini, 2025). However, former employees and critics of the company alike are skeptical.
Despite being the biggest owner of data centers in the world, Amazon states that “many of our facilities do not require the ongoing use of water to cool operations” (Barrat & Gambarini, 2025).
(Amazon, 2023)
In Amazon’s 2023 sustainability report, they note that Spain is one of the most water-stressed industrialized countries in the world, having experienced a long term drought since 2022. They also state that they have reduced water loss by an estimated 33 million liters per year, through their partnership with FIDO tech. FIDO tech is an AI water leak detection company that uses Amazon’s technology to identify and reduce leakages in water systems. In 2023, Amazon returned 3.5 billion liters of water, 41% of their water positive goal. This means that their 2023 goal was approximately 6.54 billion liters of water (Amazon, 2023).
Implications
As of 2024, there were 38 active data centers and 24 more under development owned by Microsoft, Amazon or Google in parts of the world already facing water scarcity. There are 632 data centers active or owned by Amazon, Microsoft or Google globally, but their locations are often industry secrets (Barrat & Gambarini, 2025).
Furthermore, big tech companies often rent a large portion of their data center capacity from third-party data center operators (or “colocation” data centers). Large tech companies (or “hyperscalers”) represented 37% of worldwide data center capacity in 2022, with half of that (18.5% of worldwide data center capacity) being third-party contracts (O’Brien, 2024). Through third-party operators, companies can claim smaller footprints—reduced emissions, water use, and climate toll.
Everything we do on the internet travels through data centers. Every search we make and every email we send requires water. This goes double if AI is involved. Without urgent action, clear regulation, and genuine accountability, we will never know how much water is being used until we feel its effects.
References
Ahmad, R. (2024, March 4). Engineers often need a lot of water to keep data centers cool.
American Society of Civil Engineers.
Amazon. (2023). 2023 Amazon sustainability report.
Barrat, L. & Gambarini, C. (2025, April 9). Revealed: Big tech’s new data centres will take water from the world’s driest areas. The Guardian.
Bartz, N. (2025, June 24). NREL Maps Out US Data Infrastructure. Hackaday.
Bass, D., Ma, M., & Nicoletti, L. (2025, May 8). AI is Draining Water from Areas that Need it
Most. Bloomberg.
Cisco. (n.d.). What Is a Data Center?
Downie, A. & Finn, T. (n.d.). Agentic AI vs. generative AI. IBM.
Dolfi, M., Staar, P., Ramis, C. B., Mishra, L. & Vagenas, P. (2024, February 22). AI is making extracting key information from reports easier than ever. IBM.
France 24. (2025, April 18). Thirsty giants: Big Tech’s data centres multiply in drought-hit Spain [Video].
Gomstyn, A. & Jonker, A. (2025, February 21). What is an AI data center? IBM.
Google. (2024). Google 2024 environmental report.
IBM. (2024, August 9). Generative AI vs. predictive AI: What’s the difference?
International Energy Agency. (2025). Energy and AI.
McKinsey & Company. (2024, April 2). What is generative AI?
O’Brien, I. (2024, September 15). Data center emissions probably 662% higher than big tech
claims. Can it keep up the ruse? The Guardian.
Recessary. (2025, April 7). As Demand for AI Data Centers Grows, Expert Warns Against Digital Extractivism.
Skidmore, Z. (2025, March 28). Google signs four partnerships as part of its water stewardship program. Data Center Dynamics.
Statista. (n.d.) Artificial Intelligence – Worldwide.
Swinhoe, D. (2024, September 20). Google pauses Chilean data center project to rethink water use. Data Center Dynamics.
