NVIDIA says its Rubin AI servers can run liquid coolant at 45°C, aiming to cut data-center power and water use

NVIDIA is pitching a key piece of the cooling puzzle behind its biggest AI systems: running liquid coolant hotter to make hyperscale data centers more efficient, with the goal of cutting both electricity and water use. In a June 22 blog post, the company said its newest AI server cooling architecture can operate with coolant entering at up to 45 degrees Celsius, or 113 degrees Fahrenheit, compared with typical lower-temperature approaches used in conventional designs.

The company framed the higher-temperature limit as a practical efficiency lever. NVIDIA wrote that historically, cooling has accounted for up to 40% of a data center’s electricity consumption, making cooling one of the largest controllable drivers of operating cost and total energy demand at large scale.

At the system level, NVIDIA said the Rubin generation of its AI infrastructure is designed to be “100% liquid cooled,” meaning every chip and networking component is cooled entirely by liquid in a closed loop with no fans anywhere in the system. NVIDIA tied this approach to its DSX AI factory reference design, described as a guide for designing, building, and operating the broader AI factory infrastructure stack, not just the servers.

To explain the energy and water impact, NVIDIA said it aims to reduce the need for energy-intensive cooling infrastructure such as evaporative cooling. The post quoted Ali Heydari, NVIDIA’s director of data center cooling and infrastructure, saying the DSX reference design for AI factories has “zero water consumption,” which the company said it has enabled by using dry-cooler-based designs. NVIDIA added that outside of roughly the “1% of the year” when chillers might be needed in some climates, the system is intended to avoid evaporative water cooling.

NVIDIA also provided a climate-and-scale argument for higher coolant temperatures. The company said industry estimates suggest that raising chiller plant temperatures by 1 degree can cut cooling energy costs by about 4%. Using a 50-megawatt hyperscale facility example, NVIDIA wrote that the change could save more than $4 million annually in cooling-related energy and water costs. In favorable climates, NVIDIA said its 45-degree liquid-cooling architecture could allow “chiller-less” operation with dry coolers, reducing facility cooling water use from roughly 2.6 million gallons per megawatt per year for conventional cooling-tower-based systems to near zero, up to a 100% reduction.

The underlying physical idea is straightforward, at least as NVIDIA presented it: instead of relying on large volumes of cooled air to remove heat from IT equipment, heat is captured at the chip and transported through liquid loops operating at higher temperatures. NVIDIA said data center ambient temperature becomes more flexible because “nothing in the server depends on cool air,” and the same liquid can be recirculated in a closed loop so no new water is consumed for cooling the chips.

NVIDIA described a practical example of the thermal pathway. It said the coolant used is 75% water and 25% propylene glycol, flowing through cold plates mounted directly on processors. In its description, coolant entering at 45°C exits at roughly 55°C after absorbing heat across the chip surface, and it said performance does not degrade because the liquid-cooled cold plates keep device temperatures within validated operating limits.

In addition to operational cost, NVIDIA argued that moving to fully liquid cooling can change the physical and acoustic footprint of AI racks. The company said walk-in-style assumptions about “cold data centers” are misleading, noting that silicon generates substantial internal heat and that liquid cooling keeps component temperatures in range. It also claimed that fully liquid-cooled systems can reduce noise by eliminating cooling fans, which it said can contribute to overall noise levels at or above 85 decibels in traditional air-cooled layouts.

NVIDIA tied those design choices to a broader cooling ecosystem. The post said Motivair, an advanced cooling division of Schneider Electric, has worked with NVIDIA’s product roadmap for nearly a decade, and quoted Motivair CEO Richard Whitmore on the shift to liquid cooling as power densities crossed a threshold. Whitmore said liquid cooling became mandatory once the watts per chip rose beyond what air cooling could sustain, and he added that in the right geography with appropriate system design, operators can use outdoor radiator-coil-style dry coolers without refrigeration equipment.

Looking ahead, NVIDIA also floated the possibility of waste-heat recovery, where residual heat from AI factory operations is repurposed to heat nearby buildings. That is a potential second-order benefit, but NVIDIA did not provide site-specific performance numbers in the post. The company also acknowledged an important caveat: geography matters, and it described differences between colder regions and places like Arizona, where chillers may run for a small number of days depending on outdoor conditions. For now, operators will likely focus on whether dry-cooler-based, 45-degree designs can achieve the promised near-zero water use and cooling-energy reductions at their own sites, while managing how much backup refrigeration is needed during peak weather.