Why AI Needs So Much Water
At its core, the issue is about heat. Artificial intelligence, especially training large language models, involves immense computational power. This processing generates a tremendous amount of heat from thousands of servers packed into data centres. To
prevent these expensive components from overheating and failing, they must be constantly cooled. For decades, the most energy-efficient way to do this on a large scale has been through evaporative cooling. This process uses water, which absorbs heat as it evaporates, much like how sweating cools the human body. This water vapour is then released into the atmosphere, meaning it is consumed and not returned to the local water source. While effective, this method creates a direct link between processing power and water consumption; as AI demand grows, so does the demand for cooling.
The Scale of Consumption
The numbers associated with data centre water use are staggering. A single large hyperscale facility can use between one and five million gallons of water per day, an amount comparable to a town of 30,000 to 50,000 people. While data centres still only account for a small fraction of total water use nationally, their impact is highly concentrated in specific regions. The problem is often local, where a new facility can place significant strain on municipal water supplies, especially in water-scarce areas. Google reported its data centre water consumption grew from 4.3 billion gallons in 2021 to 6.1 billion gallons in 2024. Projections show that direct water consumption by U.S. data centres could more than double by 2028, largely driven by the AI boom.
The Hidden Water Footprint
The water used directly for on-site cooling is only part of the story. There is also a massive indirect water footprint associated with the electricity required to power these facilities. Many power plants, including those that burn fossil fuels or use nuclear power, also rely on water for cooling in their electricity generation process. According to one federal report, the indirect water footprint of U.S. data centres in 2023 was nearly 800 billion liters, far exceeding the 66 billion liters used for direct on-site cooling that year. Additionally, the manufacturing of the semiconductor chips at the heart of these servers is itself a water-intensive process, requiring millions of gallons of ultrapure water daily for a typical factory.
A Shift Towards Waterless Solutions
In response to growing environmental concerns and community pushback, the tech industry is aggressively pursuing innovations to reduce water consumption. These solutions are moving away from traditional evaporative methods. Closed-loop cooling is one major trend; these systems circulate a sealed supply of water or other liquids to cool servers without evaporation, cutting water loss significantly. Microsoft has announced new data centre designs for AI workloads that use zero water for cooling by combining direct-to-chip liquid cooling with air-based heat exchangers. Other advanced methods include immersion cooling, where servers are fully submerged in a non-conductive dielectric fluid to dissipate heat. Companies are also increasingly using recycled or non-potable water instead of fresh, drinkable water.
Corporate Pledges and Future Outlook
The world's largest tech companies are making public commitments to address their water footprint. Both Google and Microsoft have pledged to become "water positive," meaning they aim to replenish more water than they consume through investments in water stewardship projects. Microsoft reported it reached this milestone in fiscal year 2025. However, transparency remains an issue, as many companies are not forthcoming about the water usage of individual facilities. While innovative technologies offer a path to greater sustainability, there is uncertainty about how quickly they can be deployed across the thousands of existing and planned data centres. The challenge for the industry will be to balance the explosive growth of AI with the urgent need to conserve a critical natural resource.
