Artificial Intelligence (AI) is rapidly spreading across industries, but it also comes with a hidden cost: it’s a water-hungry technology. AI systems rely heavily on electricity, which often requires significant amounts of water for cooling and power generation.
According to the United Nations, half of the world’s population suffers from water scarcity, and the situation is expected to worsen due to rising demand and climate change. So, could the explosive growth of AI make the global water crisis even worse?
How Much Water Does AI Consume?
Sam Altman, CEO of OpenAI, claims that every query on ChatGPT uses about one-fifth of a teaspoon of water. But a U.S. academic study from California and Texas estimates that 10 to 50 responses from GPT-3 consume around half a liter of water — that’s 2 to 10 teaspoons per response.
These differences in water estimates depend on the query type, response time, location of processing, and infrastructure used. The U.S. study includes water used to generate electricity — such as steam powering coal, gas, or nuclear plants — which Altman may not have accounted for in his figures.
Given ChatGPT reportedly handles one billion queries per day, and it's just one of many AI tools, total water use is substantial. The study estimates that by 2027, the AI sector could consume 4 to 6 times more water annually than the entire country of Denmark.
Professor Shaolei Ren of the University of California, Riverside, one of the study’s authors, puts it simply: “The more we use AI, the more water we consume.”
Why Does AI Need Water?
AI processes — from emails and streaming to image or video generation — are handled in massive data centers housing rows of servers, sometimes the size of football fields. As electricity runs through these servers, they heat up, requiring constant cooling.
Water, often clean and drinkable, is a key component in cooling systems. Many systems evaporate up to 80% of the water used, releasing it into the atmosphere. Since AI tasks require significantly more computational power than traditional online activity (like shopping or Googling), they use far more electricity — and thus, more water.
A single ChatGPT query, according to the International Energy Agency (IEA), can use up to 10 times the electricity of a Google search. That means more heat, more cooling, and more water.
How Fast Is AI's Water Use Growing?
While major tech companies don’t break down water use specifically for AI, their overall water consumption is increasing. According to their environmental reports, Google, Meta, and Microsoft — major investors in OpenAI — have seen sharp rises in water use since 2020. Google’s water use, for example, has nearly doubled.
Amazon Web Services (AWS) has not provided water usage data. The IEA predicts that data center water use — including for energy generation and chip production — will double by 2030.
Google reported withdrawing 37 billion liters of water globally in 2024, consuming 29 billion of that (mostly through evaporation). That’s enough water to supply the UN-recommended minimum of 50 liters per person per day to 1.6 million people for an entire year, or to irrigate 51 golf courses in the southwestern U.S.
Why Build Data Centers in Dry Regions?
Despite rising awareness, data centers continue to pop up in water-stressed areas — including parts of Europe, Latin America, and U.S. states like Arizona. In Spain, an environmental group named "Your Cloud Dries My River" was formed to fight new data center developments.
In Chile and Uruguay, where droughts are severe, Google halted or revised projects after water access protests. Abhijit Dubey, CEO of NTT Data, which operates 150+ data centers globally, says that land availability, access to renewable energy, and favorable regulations make dry areas appealing, despite their water issues.
Low humidity in dry areas is also attractive, as it reduces corrosion risks — and, ironically, lowers cooling costs.
Alternatives to Water-Based Cooling?
Dry or air-based cooling systems are an option, but they tend to use more electricity. Microsoft, Meta, and Amazon are developing “closed-loop” cooling systems that recycle water or alternative liquids without evaporation or replacement.
However, Dubey notes these are still in early development, especially for dry climates.
In Europe, projects in Germany, Finland, and Denmark are exploring how to reuse waste heat from data centers to warm homes.
Despite environmental efforts, most companies still prefer using clean freshwater for cooling, as it's less likely to cause bacterial growth, clogs, or corrosion. Some use non-potable water sources like seawater or industrial wastewater, but this is less common.
Is the Environmental Cost Worth It?
AI can contribute to sustainability — detecting methane leaks, rerouting traffic to save fuel, or advancing education and healthcare for children, according to UNICEF’s Thomas Davin.
Still, Davin calls for a shift in focus from building ever-larger, more powerful models to efficiency and transparency. He urges companies to open-source their models to reduce the need for extensive training — a process that consumes massive energy and water.
Lorena Jaume-Palasí, an AI ethics researcher, warns that the environmental impact of AI is ultimately unsustainable. “We can improve efficiency, but that just increases usage,” she says. “In the long run, we don’t have enough raw materials to keep growing AI models.”
What Are Tech Companies Doing About It?
Tech giants like Google, Microsoft, Meta, and Amazon Web Services say they carefully choose cooling technologies based on location. All have pledged to become “water positive” by 2030 — meaning they aim to replenish more water than they consume.
To meet these goals, they support local projects like forest restoration, wetland preservation, leak detection, and smart irrigation.
Amazon says it’s 41% of the way toward its target. Microsoft claims to be “on track.” Google and Meta both report progress in recycling more water than ever.
Yet, Thomas Davin says there's still “a long way to go.”
OpenAI, meanwhile, says it is “working hard to improve water and energy efficiency” and that “rethinking how we use computing power remains essential.”
But Professor Ren insists that the tech industry needs more consistent and unified reporting on water usage: “If we can't measure it, we can't manage it.”
