The Hidden Environmental Cost of Artificial Intelligence
Thanks to recent advancements, artificial intelligence (AI) has made its way into the mainstream. The technology has seen a rapid rise, spreading into every industry and even integrating into the devices we use every day.
However, AI has an unspoken environmental cost. The technology is energy-intensive, requiring a vast amount of electricity to power, but that’s not all—AI is also draining water resources around the world.
Let’s explore the issue not enough people are talking about: the AI water problem.
AI Water Usage: How and Why Does AI Use Water
AI systems are complex, much more so than traditional computing programs. They need a vast amount of data to “learn” how to perform a task, and processing all that data is energy-intensive, requiring advanced computing equipment, such as servers in data centers. This is where the AI water problem originates: in the data center.
According to a research article in Communications of the ACM, the flagship magazine of the Association for Computing Machinery, there are three primary areas of AI water consumption and withdrawal: data center cooling, electricity generation, and server manufacturing.
Discover the full visual story The AI Water Problem in the first issue of Graphonaut Magazine. Get your free digital copy in PDF format, delivered straight to your inbox, by subscribing to our newsletter.
FAQ
AI runs on physical hardware, servers in massive data centers that generate enormous heat around the clock. To prevent that hardware from overheating, data centers rely heavily on cooling systems, and most of those systems depend on water. AI’s water footprint has three interconnected components: direct on-site consumption for evaporative cooling; indirect water use at power generation facilities supplying electricity to data centers; and embedded water in semiconductor manufacturing.
The numbers are staggering. A typical data center uses 300,000 gallons of water each day, equivalent to the demands of about 1,000 households, but large data centers can use an estimated 5 million gallons of water each day, equivalent to the needs of a town of up to 50,000 residents.
More than most people realize. According to scientists at the University of California, Riverside, each 100-word AI prompt is estimated to use roughly one bottle of water, around 519 milliliters. Broken down further, one paper estimated that generating a single GPT-3 text output of 150 to 300 words consumed a total of 16.9 milliliters of water in an average U.S. data center, 2.2 ml for on-site cooling and 14.7 ml for electricity generation.
Most people focus on the water used inside data centers, but the hidden cost is even larger. A significant amount of water is consumed indirectly through the production of electricity needed to run AI systems, and research indicates this indirect water use can be up to four times higher than the direct water consumption for server cooling alone. Many power generation technologies remain water-intensive, particularly thermal power plants fueled by coal, natural gas, or nuclear energy, which rely on water for steam generation and cooling.
Even before a data center switches on, building it consumes water. Producing a single microchip requires 2.1–2.6 gallons of water just to cool machinery and ensure wafer sheets are free of contaminants. At scale, with billions of chips being fabricated globally, this manufacturing footprint adds up significantly.
Projections suggest the issue will intensify sharply. Researchers at the University of California, Riverside estimate that by 2027, global AI demand could drive annual water withdrawals of 1.1 to 1.7 trillion gallons, more than four to six times Denmark’s total yearly water withdrawals. Making matters worse, about two-thirds of data centers built since 2022 are in areas already experiencing water stress.
Promising alternatives are emerging. In immersion cooling systems, servers are submerged in a fluid that carries heat away without evaporating water. Design improvements such as closed-loop cooling systems can reduce freshwater use by up to 70%. On the energy side, using renewable energy for data centers also saves water indirectly, since traditional power plants, coal, gas, and nuclear use vast amounts of water for cooling and steam generation.
Transparency remains a major problem. The water footprint of AI is even more difficult to assess than its carbon footprint because a company’s total water consumption comprises both direct and indirect use, and some major companies have explicitly chosen not to report indirect water consumption. Researchers suggest increasing transparency and mandatory reporting of water consumption, similar to how companies must report carbon emissions, since even direct water use is often currently omitted.
