AI Data Center Water Use Could Hit 1.125 Trillion Liters by 2030
US data centers may use up to 1.125 trillion liters of water by 2030, but clean power, smart siting and liquid cooling could sharply cut future demand.
Summary
Protests across Texas, New Mexico and Arizona are targeting AI data centers over water, energy and pollution. Processors can reach 176°F, and many facilities evaporate water to remove heat. US data center cooling consumed an estimated 66 billion liters in 2023, under 1 percent of national consumption, but Fengqi You’s Cornell team projects 731 billion to 1.125 trillion liters annually by 2030. The high estimate roughly equals New York City’s annual drinking water supply. Totals include cooling and water used by coal and gas generation, while company disclosures remain incomplete and inconsistent. Shaolei Ren’s 2024 GPT-4 estimate of 500 milliliters for drafting a short email is already outdated. Google estimated five drops for a median Gemini query in 2025.
You says renewable power, efficient processors and siting in water richer Montana, Nebraska, parts of Texas and South Dakota, rather than Arizona, New Mexico and Southern California, could cut the future footprint by up to 86 percent. Amazon, Microsoft and Google use closed loop liquid cooling, although extreme heat or humidity can require water misting. NVIDIA sustainability chief Josh Parker says its latest processors can use 113°F cooling water, often allowing fans to reject heat, but Ren warns higher temperatures can hurt performance or incompatible non AI servers. Vaibhav Bahadur estimates Texas data centers could rise from under 1 percent of state water demand to 3 to 9 percent by 2040. Immersion cooling and ocean installations offer alternatives but can complicate maintenance. AWS aims to become water positive by 2030 and was more than halfway there in 2024. Google targets replenishing 120 percent of freshwater consumption by 2030, OpenAI has outlined conservation plans, and European facilities are supplying waste heat to district heating systems.
Positives
- Strategic siting, renewable power and efficient processors could reduce AI’s future water footprint by up to 86 percent.
- Google estimated that a median Gemini query consumed only five drops of water in 2025.
- NVIDIA’s latest processors can use 113°F cooling water, reducing reliance on evaporation outside severe heat waves.
- AWS was more than halfway toward its 2030 goal of returning more water than its direct operations consume by 2024.
- Google plans to replenish 120 percent of the freshwater consumed by its data centers by 2030.
Risks & concerns
- US data center water consumption could reach 1.125 trillion liters annually by 2030, roughly New York City’s yearly drinking water supply.
- Texas data centers could grow from under 1 percent of state water demand to 3 to 9 percent by 2040.
- New construction in Arizona, New Mexico and Southern California could intensify water stress in drought prone communities.
- Extreme heat and humidity can force closed loop cooling systems to activate additional water consuming methods.
- Incomplete and inconsistent company disclosures make AI’s actual water footprint difficult to measure or compare.