Nvidia is presenting a new cooling approach as a major step toward reducing data center water use. The company says its warm-water system can cut on-site consumption sharply, and in some climates may eliminate it entirely.
That is a meaningful facility-level change. But it does not settle the broader water question around AI infrastructure, because much of the footprint sits outside the building itself.
What Nvidia says its cooling system changes
Nvidia announced a warm-water cooling system designed to reduce the water a data center uses. In a press release, an Nvidia executive said it could eliminate “pretty much all water usage” inside the data center itself.
Josh Parker, chief sustainability officer at Nvidia, recently told Axios: “The water consumption challenge for data centers is largely solved,” according to the source article.
The technical idea is straightforward: the coolant runs in a closed loop. It is filled once and then recirculated for the life of the facility, so no new water is consumed to cool the chips. Nvidia says that, in favorable climates, this can amount to a 100% reduction in on-site water use.
The system moves coolant into racks at 45˚ C (113˚ F). After passing through a server, Nvidia said, it comes out at 55˚ C (131˚ F), carrying heat away from the hardware.
At that temperature, outside air in most climates can remove heat from passive radiators without evaporative cooling. In some cases, it may also avoid the need for fans. A data center without fans or chillers would use less water, consume energy more efficiently, and operate more quietly.
The measurement boundary matters
The key limitation is not whether Nvidia’s cooling system works inside the facility. The issue is how the company defines the water footprint.
According to the source article, Nvidia’s blog post effectively draws a boundary around the data center. Water used inside that boundary is counted. Water used outside it is not.
That distinction changes the meaning of the claim. A facility can cut on-site cooling water dramatically while still depending on outside systems that consume large amounts of water.
The source identifies two major areas beyond the data center walls:
- Electricity generation, especially when data centers run on fossil fuels.
- Chip manufacturing, which is part of the wider infrastructure footprint.
Water use outside the data center can double or triple the total water footprint of a facility. On that basis, Nvidia’s approach addresses about a quarter to a third of AI data centers’ total water consumption.
TechCrunch said it asked Nvidia to clarify the matter and would update its article if it received a reply.
Power choices shape the bigger water footprint
No data center can run without electricity. That means the source of power matters as much as the cooling system inside the building.
Many types of power plants use substantial amounts of water. Fossil fuel power plants are one of the largest water users in the U.S., consuming 2.7 billion gallons per day, according to the U.S. Geological Survey. Most of that use is for evaporative cooling.
Natural gas power plants use 1.17 liters of water for every kilowatt-hour of electricity they generate, according to a recent study cited in the source article. Coal plants use 2.2 liters per kilowatt-hour.
Fossil fuel power plants collectively generate about half of all data center power today, according to the IEA. That matters because tech companies are increasingly making choices that keep AI data centers tied to fossil fuels.
Hydropower is different. Dams do not consume water in the same direct way, but reservoir evaporation leads to 6.8 liters lost per kilowatt-hour generated. Hydropower supplies around 10% of data center power.
Geothermal power varies widely, depending on the specific technology. Some enhanced geothermal startups, including Fervo, have pledged to use mostly “degraded” water that would otherwise go unused.
Why wind and solar change the equation
Wind and solar power have far smaller water footprints than fossil fuel plants. Wind uses about 0.01 liters per kilowatt-hour. Solar uses about 0.03 liters per kilowatt-hour, including the water needed for manufacturing and cleaning solar panels.
That contrast shows why an AI water strategy cannot stop at server racks. Cooling hardware can reduce one part of the problem, but the electricity mix determines how much water is consumed elsewhere to keep data centers running.
Renewables are providing a growing share of new electricity capacity. Still, natural gas and coal are expected to provide more than 40% of the new electricity needed to meet data center demand through 2030, according to IEA projections cited in the source article.
If that trajectory does not change, data centers will continue consuming large amounts of water. Nvidia’s cooling system may reduce water use inside the facility, but it cannot erase water consumption tied to the power plants and manufacturing systems that support AI infrastructure.
The practical takeaway
Nvidia’s warm-water cooling system appears to be a real improvement for on-site data center water use. Closed-loop coolant, higher operating temperatures, passive radiators, and reduced reliance on evaporative cooling all point to a more efficient facility design.
But the broader AI water problem is not contained inside the data center. It extends into the energy system and the supply chain. Any serious accounting of AI data center water use has to include those external sources, especially when fossil fuel power remains a large part of the electricity supply.
The result is a more nuanced picture: Nvidia may help solve an important cooling problem, but that is not the same as solving AI’s total water footprint.