NVIDIA Rubin Liquid Cooling: Reducing Data Center Water Use to Near Zero

The 45°C Breakthrough in AI Infrastructure Cooling

NVIDIA's Rubin generation of AI infrastructure achieves 100% liquid cooling, allowing servers to operate with coolant temperatures up to 45°C (113°F). This higher temperature threshold enables data centers to transition from energy-intensive evaporative cooling to closed-loop dry coolers, effectively reducing facility water consumption from approximately 2.6 million gallons per megawatt per year to near zero.

Transitioning to 100% Liquid Cooling

While previous liquid-cooled servers were hybrid systems—where GPUs and CPUs were liquid-cooled but other components remained air-cooled—the Rubin architecture is fully liquid-cooled. Every chip and networking component is cooled by liquid in a closed loop, eliminating the need for fans and the traditional "hot aisle/cold aisle" physical layout of data centers.

Engineering Improvements

Component Redesign: NVIDIA's thermal engineering team redesigned cooling loops to route liquid to multiple high-power chips on a single board using one inlet and outlet, simplifying the tray-level architecture.
Space Efficiency: Fully liquid-cooled servers enable higher rack density. Systems that previously occupied six rack units can now fit into two, allowing for more compute power in less physical space.
Noise Reduction: By eliminating cooling fans, which typically contribute to noise levels of 85 decibels or higher, the environment becomes significantly quieter.

Energy and Water Efficiency Gains

Raising the cooling setpoint allows heat to be rejected more efficiently into the outdoor air. In favorable climates, this enables "chiller-less" operation, where heat is transported via coolant distribution units to outdoor radiator coils (dry coolers) without the need for mechanical refrigeration.

Economic and Environmental Impact

Energy Costs: Industry estimates suggest that raising chiller plant temperatures by just one degree can reduce cooling energy costs by 4%. A 50-megawatt hyperscale facility can save over $4 million annually by moving to liquid-cooled infrastructure.
Water Consumption: Conventional cooling towers rely on evaporation to remove heat. By using a closed-loop system with a coolant mixture of 75% water and 25% propylene glycol, NVIDIA's design eliminates the need for evaporative water loss.
Waste Heat Recovery: The higher operating temperature of the coolant makes it more viable to repurpose residual heat for district heating in commercial or residential buildings.

Technical Constraints and Industry Perspectives

Despite the efficiency gains, the implementation of this technology faces geographic and technical challenges. The effectiveness of "dry cooling" depends heavily on the ambient outdoor temperature; in extremely hot climates, mechanical chillers may still be required for a small percentage of the year.

Community Insights and Counterpoints

Industry practitioners and observers have raised several points regarding the scalability and practicality of this design:

"In the right geography — somewhere with reliably cool outdoor air — a liquid-cooled data center can reject its heat through coolant distribution units... The loop is filled once and runs closed for the life of the facility."

However, some critics argue that the water savings are a matter of perception, noting that the initial fill of the system requires significant water volume and that the heat is simply being moved from water evaporation to air pollution (waste heat).

Additionally, some engineers have pointed out that while 45°C is an improvement, high-temperature liquid cooling is not a new concept in specialized supercomputing. For example, the NASA Ames Research Center Modular Supercomputing Facility operates with inlet water temperatures around 90°F (approx 32°C), avoiding traditional air conditioning.