Nvidia targets data center cooling as AI infrastructure strain grows

Nvidia is positioning cooling as a central part of the next phase of AI infrastructure. As AI factories expand into large computing campuses, operators face constraints beyond GPUs, including electricity, land, memory and water. Traditional data centers often use evaporative cooling towers, which can create resource pressure in regions with limited water supply.

The company’s DSX AI factory architecture uses closed-loop liquid cooling instead of water towers. Coolant moves directly to chips, absorbs heat and recirculates through a sealed system without evaporating. Nvidia CEO Jensen Huang has said future systems built around the Vera Rubin architecture can operate with coolant entering at approximately 45°C, allowing facilities to use outside air for heat removal rather than energy-intensive chillers.

Lower cooling overhead could let AI factories devote more electricity to computation. The Energy Dept. has estimated cooling can account for roughly 40% of a facility’s electricity consumption depending on design. Nvidia’s systems can also produce warmer coolant output, around 54°C, which could potentially be reused for nearby buildings, industrial processes or district heating.

The strategy extends Nvidia’s role beyond selling GPUs into networking, data center design, cooling technology and software ecosystems. AMD and Intel continue to develop AI accelerators, but Nvidia’s integrated infrastructure approach could help address bottlenecks that limit large-scale deployment.