NVIDIA's Warm-Water Cooling System Cuts On-Site Data Center Water Use, but Broader Water Footprint Remains

NVIDIA has unveiled a warm-water cooling system for its Rubin generation AI servers that the company claims can reduce on-site water consumption by up to 100% in favorable climates. The system operates at temperatures up to 45 degrees Celsius (113 degrees Fahrenheit) — hotter than a typical hot tub — using 100% liquid cooling in a closed loop with no evaporative cooling or fans. According to NVIDIA, this approach can cut facility cooling water consumption from approximately 2.6 million gallons per megawatt per year for conventional cooling-tower-based systems to near zero, while also reducing cooling energy costs by millions annually at hyperscale facilities.

The technical innovation addresses a major inefficiency in traditional data center design. Historically, cooling alone accounts for up to 40% of a data center's electricity consumption. By capturing heat directly at the chip and transporting it through high-temperature liquid loops, NVIDIA's system allows outdoor dry coolers to reject heat efficiently without requiring energy-intensive mechanical cooling or chiller systems during most of the year. The coolant is filled once and recirculated throughout the facility's lifespan, eliminating new water consumption for cooling. NVIDIA's leadership, including Ali Heydari, director of data center cooling and infrastructure, and Josh Parker, chief sustainability officer, assert that the Rubin platform reference design has "eliminated massive amounts of power usage and pretty much all water usage."

However, critics and analysts note significant limitations to NVIDIA's claims. The company's accounting draws a boundary around the data center facility itself, excluding water consumption from upstream sources. The broader water footprint of AI operations includes electricity generation and chip manufacturing — areas where NVIDIA's solution provides no mitigation. Water consumption in power generation can be substantial: natural gas power plants use 1.17 liters of water per kilowatt-hour, while coal plants consume 2.2 liters per kilowatt-hour. Hydropower generation results in 6.8 liters of water lost per kilowatt-hour through reservoir evaporation. Industry analysis suggests that water use outside the data center — primarily in electricity generation and manufacturing — can double or triple the total water footprint, meaning NVIDIA's solution addresses approximately one-quarter to one-third of AI data centers' total water consumption.

Additionally, the Rubin platform does not address water consumption during data center construction or the broader environmental concerns surrounding AI infrastructure expansion. While NVIDIA states that "every cloud provider and data center operator building for Rubin is making the transition," the company has not publicly disclosed the cost differential between fully liquid-cooled systems and conventional air-cooled designs. Industry momentum appears strong, with cooling equipment manufacturers such as Motivair, the advanced cooling division of Schneider Electric, working closely with NVIDIA's product roadmap. Nevertheless, experts emphasize that without coordinated changes to electricity generation sources — shifting away from fossil fuels toward renewable energy like wind and solar, which consume minimal water — the broader water crisis facing AI data centers will persist despite on-site efficiency gains.