Data centers supporting AI operations have become notorious for their massive water consumption, sometimes using as much water daily as a small town. Nvidia has introduced a new server platform that could transform this reality by nearly eliminating the need for water in on-site cooling systems, thanks to an innovative approach to thermal management.
The company’s Vera Rubin servers operate at significantly higher temperatures than traditional designs, enabling a shift from evaporative cooling—which wastes millions of gallons of water—to a closed-loop liquid-cooling system. This system circulates a water-and-propylene-glycol mixture heated from around 113°F to 131°F, which then cools without losing water through evaporation. This technology could reduce fresh water use for cooling to almost zero in many climates.
Such advancements come at a critical time, given the rapid growth of AI infrastructure that strains both energy grids and water resources. While Nvidia’s innovation addresses cooling water demands, it does not reduce the extensive electricity consumption tied to AI training and operation, nor does it alleviate other environmental pressures from expanding data center footprints. Still, reducing water stress presents a notable environmental benefit amid concerns about drought and rising utility costs in certain regions.
Josh Parker, Nvidia’s sustainability lead, described the ability to intake and operate at higher temperatures as a transformative development. This breakthrough could ease one of the major environmental challenges linked to AI expansion, though the sector’s broader resource footprint remains a pressing issue as demand for AI-powered services continues to rise.