Nvidia warm-water cooling: cuts on-site data center water use

Nvidia announced a warm-water, closed-loop cooling system that the company says can eliminate "pretty much all water usage" inside a data center, and its chief sustainability officer Josh Parker told Axios, "The water consumption challenge for data centers is largely solved." The system circulates a single fill of coolant for the life of the facility, and in favorable climates the company says that can amount to a 100% reduction in on-site water use.

What did Nvidia announce?

Nvidia announced a facility-level warm-water cooling system that pumps coolant into server racks at 45°C (113°F) and returns it at 55°C (131°F). The company describes the loop as closed, filled once and recirculated for the life of the facility, which it says eliminates on-site water consumption for chip cooling and can enable passive radiators to reject heat in many climates.

The announcement frames the change as an end to data center water consumption inside the data center walls. Nvidia’s messaging includes the claim that the system can remove "pretty much all water usage" inside the data center and Parker’s statement that "The water consumption challenge for data centers is largely solved."

How does the system work and what does it cover?

The coolant is injected into racks at 45°C and leaves at 55°C, carrying substantial heat away from hardware so outside air can draw heat off passive radiators without evaporative cooling or, in some cases, fans. That design reduces or eliminates the need for on-site evaporative cooling and chillers, removing their associated water consumption.

The system’s closed-loop is the key to the facility-level claim: because the coolant is recirculated and not continually topped up, Nvidia counts water consumed only inside the data center perimeter. That accounting excludes water tied to electricity generation and chip manufacturing, which the company’s blog and the reporting note sit outside the data center boundary.

How big is the water use outside the data center?

Water use outside the data center can be large. Fossil fuel power plants consume 2.7 billion gallons per day in the U.S., according to the U.S. Geological Survey, most of it for evaporative cooling. The International Energy Agency estimates fossil fuel plants currently generate about half of all data center power. Per-unit figures from recent studies cited in the coverage show natural gas power plants use 1.17 liters of water per kilowatt-hour and coal plants use 2.2 liters per kilowatt-hour. Hydropower’s reservoir evaporation is listed as 6.8 liters lost per kilowatt-hour.

Renewables use far less water per kilowatt-hour: wind about 0.01 liters and solar about 0.03 liters, the reporting states. Despite growth in renewables, the IEA projects natural gas and coal will provide more than 40% of the additional electricity needed to meet data center demand through 2030. The result is that facility-level reductions can address roughly a quarter to a third of an AI data center’s total water footprint, because electricity generation and chip manufacturing outside the site can double or triple total water use.

Why it matters

Nvidia’s warm-water system tackles a visible slice of data center water consumption and could eliminate on-site cooling water in many deployments. That reduces local stress on water supplies and can simplify facility operations. At the same time, the broader water footprint of AI infrastructure depends heavily on how electricity is generated and on chip manufacturing processes. If data centers continue to run largely on fossil-fueled power or rely on water-intensive manufacturing, total water consumption linked to AI operations will remain high despite facility-level savings.

What to watch

Watch how operators account for water across the full supply chain: a true reduction in an AI workload’s water footprint requires cleaner electricity mixes and lower-water chip manufacturing. Also watch deployment details: whether Nvidia’s system is paired with low-water or renewable electricity at scale will determine whether facility-level savings translate into meaningful reductions in total water use.