NVIDIA introduces direct-to-chip liquid cooling technology, improving both performance density and energy efficiency.
Nvidia published a blog post that could reshape the artificial-intelligence industry. In it, Nvidia's head of corporate sustainability, Josh Parker, revealed that the company's new closed-loop liquid-cooling system may have solved one of AI's most pressing problems: reducing data-center water consumption to near zero while cutting cooling-related energy use by up to 40%.
The breakthrough works by allowing Nvidia's Rubin architecture servers to operate cooling liquid at temperatures up to 45 degrees Celsius (113 degrees Fahrenheit)—hotter than typical hot tubs at 38–40 degrees Celsius, yet more energy efficient. By eliminating most chillers and fans and reducing water dependence, the system frees up space and power for additional compute capacity while maintaining device temperatures within optimal operating ranges.
Parker noted that Rubin is the "world's first to achieve 100% liquid cooling—every chip, every networking component, cooled entirely by liquid in a closed loop with no fans anywhere in the system." The coolant is a mixture of 75% water and 25% propylene glycol, formulated similarly to antifreeze. Heat is captured directly at the chip and transported through liquid loops operating at higher temperatures, allowing outdoor dry coolers to reject heat efficiently while substantially reducing chiller, cooling-fan, and water requirements. The closed-loop design is filled with coolant once and remains sealed for the data center's operational life.
In moderate and cool climates, this architecture could reduce data-center water usage from roughly 2.6 million gallons per megawatt annually to nearly zero—a 100% savings. Ali Heydari, Nvidia's director of data-center cooling and infrastructure, stated: "With dry-cooler-based designs, it's a closed-loop system with no evaporative water cooling—outside of maybe 1% of the year when we might need chillers in some [warmer] climates."
The problem addressed is severe. Every AI chip running a workload generates heat. Graphics processing units can reach 105 degrees Celsius (221 degrees Fahrenheit) during heavy workloads. As Nvidia's Blackwell rack architecture scales toward 100 kilowatts per rack—with AMD and hyperscaler silicon pushing densities higher—cooling becomes the binding constraint on data-center operation. Air cooling cannot solve this at scale; liquid cooling is the only proven solution for required rack densities.
Large data centers consume up to 5 million gallons of water daily—equivalent to the consumption of a city of 50,000 people. This strains water supplies in drought-prone areas. Georgia recently experienced moderate-to-severe drought while a data-center campus in Fayetteville (20 miles outside Atlanta) drained nearly 30 million gallons from the area without payment or local awareness until residents experienced low water pressure. For perspective, 30 million gallons could fill 44 Olympic-sized swimming pools.
The power demands are equally staggering. The Environmental and Energy Study Institute estimates U.S. data centers could require 130 gigawatts of power annually by 2030—enough to power roughly 114 million homes, equivalent to a massive portion of America's 148 million housing units. Gartner projects electricity demand from AI data centers will double by 2030, rising from today's 448 terawatt-hours to 980 TWh by decade's end, representing roughly 3% of global power consumption. McKinsey data shows up to 40% of a data-center's electricity consumption currently comes from cooling alone.
Traditional air cooling creates additional operational challenges: cooling fans reaching 85 decibels, temperature variability across "cold aisles" at 64.4 degrees Fahrenheit and "hot aisles" at 95–110 degrees Fahrenheit, and significant spatial inefficiency. Nvidia's liquid cooling not only saves water and energy but also frees up physical space, with systems previously occupying six rack units now fitting in two.
A caveat: Rubin is still in volume production and won't ship until later this year. Years will pass before meaningful data-center adoption occurs, meaning existing infrastructure will continue using less-efficient cooling systems. Additionally, air-cooled data centers will require retrofit infrastructure to accommodate liquid cooling, necessitating direct-to-chip liquid-cooling installation.