Nvidia Claims Its New Liquid Cooling Design Eliminates Data Center Water Use Entirely
At London Climate Week, Nvidia unveiled a closed-loop liquid cooling reference design for its Rubin AI infrastructure that it says achieves 100% reduction in evaporative water consumption. The system uses a water-propylene glycol coolant operating at 45°C — hot enough to shed heat without external chillers.
Nvidia says it has solved AI’s water problem. That’s a bold claim — and it deserves scrutiny.
At London Climate Week on June 22, the company unveiled a closed-loop liquid cooling reference design for its Rubin AI infrastructure that it claims achieves a 100% reduction in evaporative water consumption compared to conventional data center cooling. Josh Parker, Nvidia’s chief sustainability officer, said plainly: “The water consumption challenge for data centers is largely solved.”
How the System Works
Conventional data centers use evaporative cooling towers — essentially giant outdoor swamp coolers — to reject heat. They consume enormous amounts of water in the process. A 50-megawatt facility can use around 2.6 million gallons of water per megawatt per year, a figure the UN has used to project that AI-related water consumption could equal the annual needs of 1.3 billion people by the end of the decade.
Nvidia’s approach eliminates that external water loop. The system circulates a closed-loop coolant made of three-quarters water and one-quarter propylene glycol directly over the chips. The critical innovation is the operating temperature: the coolant runs at 45°C (113°F), roughly 15 degrees hotter than industry-standard liquid cooling systems.
That higher temperature matters because it means the coolant can shed its heat to the outdoor environment — through dry coolers or ambient air — without requiring additional chilling equipment in most climates. No evaporation, no water consumption.
Ali Heydari, Nvidia’s director of data center cooling and infrastructure, said the design has “eliminated massive amounts of power usage and pretty much all water usage.”
The Financial Case
Nvidia estimates a 50-megawatt hyperscale facility could save over $4 million annually in cooling-related energy and water costs by switching to this design. At scale — and hyperscale facilities are often 100 to 500 MW — that’s a meaningful operational number, not just a sustainability talking point.
The Caveats
Not everyone is convinced the claim holds universally.
Professor Andrew A. Chien, a researcher who studies data center environmental impact, told Fortune that “‘zero water use’ is unrealistic” in all climates. Dry cooling at 45°C works well in cool or temperate environments. In hot climates — Phoenix, Singapore, Saudi Arabia — where ambient temperatures regularly exceed the coolant’s operating temperature, some supplemental cooling is still required.
There’s also a cost problem. Nvidia did not respond to questions about the capital cost of the system or plans to retrofit existing data centers. Closed-loop liquid cooling requires significant infrastructure investment, and the world’s existing data center stock was not built for it.
The Context
AI data centers are under environmental scrutiny from multiple directions at once. Power consumption is the headline issue — models like GPT-5.x and Gemini 3.5 Pro require training runs measured in gigawatt-hours. Water is the quieter companion problem.
If Nvidia’s Rubin reference design holds up at scale and in warm climates, it removes one of the most politically charged objections to new AI data center construction. Municipalities blocking new data centers over water use would have fewer grounds to do so.
That’s a strategic advantage for Nvidia beyond the hardware sale. The company selling the chips also shapes the narrative about whether those chips can be deployed responsibly.