NVIDIA advances direct-to-chip liquid cooling technology, enabling higher density GPU deployments with improved thermal management.

Data centers face mounting criticism for their energy-intensive operations. Rising compute demands generate significant heat, requiring substantial amounts of air and water to maintain optimal operating temperatures. This cooling challenge has been compounded by land scarcity and financial incentives that are pushing new data center developments into drought-prone regions, further straining local water resources.

Traditional air cooling has reached its practical limits as AI hardware becomes increasingly dense. Recognizing this constraint, Nvidia has developed a closed-loop liquid cooling system designed to address both sustainability and performance concerns in AI infrastructure.

The system operates by running coolant at 45°C (113°F)—approximately 5-7°C higher than hot tub temperatures—enabling simpler cooling systems and reduced electricity consumption. Nvidia's proprietary coolant formulation uses 75% water and 25% glycol. In this closed-loop design, coolant continuously circulates through servers, directly removing excess heat from chips. The warm coolant then cycles through external dry coolers, leaving virtually no water evaporation.

"The Nvidia DSX reference design for AI factories has zero water consumption—we have eliminated massive amounts of power usage and pretty much all water usage," said Ali Heydari, Nvidia's Data Center Cooling and Infrastructure Director. Compared with traditional evaporative cooling towers, this approach can reduce cooling-related water consumption by as much as 100% in suitable climates, even eliminating cooling towers entirely on typical days.

The system's primary advantage stems from direct-to-chip liquid cooling, where liquid flows directly through cold plates attached to CPUs and GPUs, capturing heat at its source. This method transfers heat thousands of times more effectively than air cooling, delivering superior thermal efficiency compared to room-level cooling approaches.

The performance benefits extend beyond energy and water savings. Independent testing shows that Nvidia's H100 systems delivered approximately 17% higher performance when water cooled versus air cooled. Under sustained AI workloads, GPU temperatures ranged between 41-50°C with water cooling compared to 54-72°C with air cooling. Greater thermal efficiency may also extend hardware longevity.

Equally significant are the density improvements. Nvidia's Rubin systems now fit inside two racks instead of six, allowing data center operators to add more GPUs per rack, increase rack power levels, and deploy larger AI clusters within existing building footprints. "Once the watts per chip crossed a certain level, liquid cooling became mandatory," noted Motivair CEO Richard Whitmore.

Nvidia plans to deploy these higher-temperature closed-loop water cooling systems in upcoming Rubin installations beginning this year.