Nvidia’s warm-water cooling system reduces on-site data center water use but does not address broader AI water footprint

Nvidia announced a warm-water cooling system designed to eliminate on-site water use in data centers by circulating coolant in a closed loop, which the company says can reduce facility-level water consumption by up to 100% in favorable climates. The system pumps coolant into server racks at 45°C, extracting heat so the warmed coolant exits at 55°C. At this temperature, passive radiators can dissipate heat without evaporative cooling or fans in many climates, reducing both water use and energy consumption.

However, analysts and the company acknowledge that the solution addresses only a portion of AI’s total water footprint. According to TechCrunch’s reporting, the on-site reduction represents roughly a quarter to a third of the total water consumed by AI data centers, with the remainder tied to electricity generation and semiconductor manufacturing.

Fossil fuel power plants, which supply about half of current data center power according to the International Energy Agency, are major water consumers. The U.S. Geological Survey estimates fossil fuel plants use 2.7 billion gallons of water per day, primarily for evaporative cooling. Natural gas plants consume 1.17 liters of water per kilowatt-hour generated, while coal plants use 2.2 liters per kilowatt-hour.

Renewable energy sources like wind and solar have significantly lower water footprints—about 0.01 liters and 0.03 liters per kilowatt-hour, respectively—though manufacturing and cleaning processes contribute small amounts. The IEA projects that natural gas and coal will still supply over 40% of new electricity capacity needed to meet data center demand through 2030, even as renewables grow.

Nvidia’s system does not address water use in power generation or chip production. The company’s sustainability executive stated in an interview that the water consumption challenge is ‘largely solved’ at the facility level, but critics argue this framing ignores the broader system-wide impacts of AI’s energy mix and supply chain.