No products in the cart.
As artificial intelligence (AI) and cloud-based services grow, so too does the environmental concern over the amount of water needed to keep the digital world cool. Data centres use vast volumes of clean water for cooling; most of this water evaporates and is not returned to the watershed. Studies suggest that medium-sized data centres withdraw tens of millions of gallons per year, while the largest facilities may use over a billion gallons annuallyeesi.org. In 2021, U.S. data centres collectively consumed about 163.7 billion gallons of watereesi.org.
Data Centre Water Consumption Overview
Clean water is pumped through cooling systems to maintain safe temperatures for servers. Large hyperscale facilities may withdraw up to 5 million gallons per day (1.8 billion gallons per year)eesi.org. The Environmental and Energy Study Institute (EESI) estimates that fewer than one-third of operators track water useeesi.org, and roughly 80 percent of the water withdraws is lost to evaporationeesi.org.
Indirect water use
Besides on-site cooling, data centres indirectly consume water for electricity generation. EESI calculated that U.S. data centres indirectly used about 211 billion gallons of water in 2023 through power-plant coolingeesi.org. This indirect footprint often goes unreported, despite being larger than direct withdrawals.
Water footprint per AI prompt
The water footprint of AI appears modest at first glance—researchers estimate that processing a 100‑word query on a large model uses roughly 519 millilitres of watertechrepublic.com. However, millions of users and constant training quickly multiply consumption: training GPT‑3 consumed about 700,000 litrestechrepublic.com, while training ChatGPT reportedly required 11.5 million gallons at Microsoft’s Iowa facilitiesfuturism.com.
Company‑Specific Water Use
The table below summarises reported water withdrawals for major technology firms. Note that many companies do not fully disclose water use, so several values are marked “Unknown.”
| Company | Water Use (Billion gallons) | Key Info |
|---|---|---|
| 6.4 | 6.4B gallons; 95% for cooling | |
| Microsoft | 1.7 | 1.7B gallons; 68.5M at Des Moines |
| Meta | 0.813 | 0.813B gallons; 95% for cooling |
| Amazon | Unknown | Unknown; uses reclaimed water at 20 sites |
| Apple | Unknown | Unknown; aims for 100% water positive |
| OpenAI | Unknown | Unknown; training pumped 11.5M gallons |
| DeepSeek | Unknown | Unknown; 2k chips; low energy |
Google reported withdrawing 6.4 billion gallons of water in 2023, with 95 % used to cool its data centresthecurrentga.org. The company’s water use increased by about 20 % from 2022e360.yale.edu. Google’s Iowa campus in Council Bluffs alone used roughly 1 billion gallons in 2024thecurrentga.org. Though Google says around 25 % of its cooling water comes from reclaimed sourcesasianometry.com, critics argue that total withdrawals remain high.
Microsoft
Microsoft increased its water consumption by 34 % in 2022, reaching nearly 1.7 billion gallonsfuturism.com. A training run for GPT‑4 in West Des Moines, Iowa consumed 11.5 million gallons in July 2022 and 13.4 million gallonsin August 2022futurism.com. Across its five facilities in the city, Microsoft used 68.5 million gallons of water in 2024, making it the region’s largest userkcur.org. In response, the company is piloting closed‑loop cooling that uses chip‑level heat exchange and avoids water except for an initial fill—each such data centre could save 125 million litres per yearinstituteofsustainabilitystudies.com.
Meta
Meta (Facebook) consumed 813 million gallons of water globally in 2023, with 95 % used in data centresthecurrentga.org. The company emphasises using non‑potable and recycled water where possible but has not provided site‑level detailsdigitalinformationworld.com.
Amazon
Amazon Web Services (AWS) does not disclose total water consumption but claims to reuse water multiple times. The company said its water recycling program preserved 530 million gallons of drinking water annually and that it aims to expand the number of data centres using treated wastewater from 20 to 120 sites by 2030datacenterdynamics.com. Nonetheless, new AWS facilities may still withdraw large volumes: Amazon’s planned centres in Aragon, Spain are licensed to use 755,720 m^3 (about 200 million gallons) per yeartheguardian.com.
Apple
Apple has not publicly reported its data centre water use for yearsaquatechtrade.com. However, the company pledged to replenish 100 % of the fresh water used in high‑stress regions by 2030 and to achieve a 50 % water reuse rate across its suppliersaquatechtrade.com. Apple has committed nearly 7 billion gallons of water benefits over the next two decadesaquatechtrade.com.
OpenAI
OpenAI does not operate its own data centres; instead, its models run on cloud infrastructure from Microsoft and Google. During training, however, water demand is intense: Microsoft’s Iowa data centres pumped 11.5 million gallonsover a single month for GPT‑4futurism.com. Researchers estimate that a 100‑word query processed by ChatGPT uses around 519 millilitres of watertechrepublic.com, and training GPT‑3 consumed 700,000 litrestechrepublic.com.
DeepSeek
Chinese firm DeepSeek claims its language model uses roughly 2,000 chips for training—far fewer than the ~16,000typically employed by Western rivalsmarylandmatters.org. This hardware efficiency means lower energy and water demand. DeepSeek’s R1 model reportedly uses such a small amount of energy that it challenges the need for new gas‑fired power plantstriplepundit.com.
Transparency and Hidden Footprints
Many tech firms provide incomplete or aggregated data on water use. Amazon, Apple, and OpenAI do not publish comprehensive metricsthecurrentga.org, while Microsoft and Meta offer only overall figures without site detailsthecurrentga.org. Moreover, corporate sustainability reports rarely address indirect water consumption from electricity generation. According to EESI, only one‑third of data centre operators track water useeesi.org, making accurate comparisons difficult.
Long‑Term Implications
If unchecked, data centre water demand could exacerbate water scarcity, particularly in arid regions. Amazon’s planned site in Arizona is licensed to withdraw 5.5 million m^3 (~1.45 billion gallons) of water annually, comparable to the consumption of 23,000 residentstheguardian.com. In 2025 researchers warned that current AI data centres use six times the amount of water as Denmark and as much energy as Japanmarylandmatters.org. Excessive withdrawals may lower aquifers, cause land subsidence, and spur conflicts with local communities.
Solutions and Emerging Innovations
1. Advanced cooling technologies. Companies are exploring direct‑to‑chip, immersion, and closed‑loop cooling systems that drastically reduce or eliminate water use. Microsoft’s new designs avoid water entirely by circulating a coolant through chips and radiatorsinstituteofsustainabilitystudies.com.
2. Water reuse and non‑potable sources. AWS recycles water at 20 data centres and plans to expand to 120 sites by 2030, preserving 530 million gallons of drinking water annuallydatacenterdynamics.com. Google uses reclaimed or seawater at some locationsasianometry.com, and Digital Realty obtains 36 % of its cooling water from non‑drinkable sourcesasianometry.com.
3. Waste‑heat recovery. Microsoft’s Finland data centre channels waste heat to a district heating system, reducing overall energy and water consumptiontriplepundit.com.
4. Efficient AI models. DeepSeek’s energy‑efficient training reduces the number of chips required, lowering water demandmarylandmatters.org. Optimised algorithms and smaller models can deliver similar performance with less resource use.
5. Siting strategies. Building data centres in cooler climates reduces reliance on evaporative cooling. Meta’s facility in Sweden uses natural cold airtheguardian.com, while some researchers propose using polar regions or even orbiting servers to take advantage of near‑free cooling—ideas that remain largely conceptual due to logistical and environmental challenges.
6. Policy and transparency. Governments can require companies to disclose water withdrawals, adopt water‑positive commitments, and avoid constructing hyperscale facilities in water‑scarce areas. Transparent metrics such as Water Usage Effectiveness (WUE)—litres used per kilowatt‑hour of energy—allow stakeholders to compare facilities. WUE values have improved to 0.2–0.5 L/kWh for modern sitesdatacenterknowledge.com, but total demand continues to rise.
Charts and Flow Diagram
Figure 1: Water Use by Company (2023 or most recent)
Figure 2: U.S. Data Centre Water Use Over Time
Figure 3: Data Centre Cooling Water Cycle
Conclusion
Clean water is emerging as a critical resource for the digital economy. The thirst of data centres—driven by AI training and cloud services—competes with human and ecological needs. While companies are beginning to adopt innovations such as closed‑loop cooling, reclaimed water, and efficient models, overall demand is still rising. To ensure a sustainable future, transparency, regulation, and investment in water‑efficient technologies are essential.
