NVIDIA has announced that its newest AI servers can run cooling liquid at temperatures up to 45 degrees Celsius — hotter than a typical hot tub. This higher threshold is designed to improve energy efficiency in large-scale AI data centers, known as AI factories.
The breakthrough addresses a growing challenge: as AI models grow, the machines that train them generate immense heat. Traditional cooling systems consume substantial power, but NVIDIA’s approach allows coolant to absorb more heat before needing to be chilled again.
By running coolant at 45°C rather than lower temperatures, the system reduces the energy required for cooling. The company did not disclose exact efficiency gains, but emphasized that the higher temperature limit is “precisely what makes them more energy efficient.”
This innovation could lower operational costs for hyperscale data centers and support the deployment of more powerful AI hardware without proportional increases in cooling energy. It may also ease pressure on water and electricity resources in regions hosting large compute clusters.
Some experts caution that liquid cooling systems remain expensive to retrofit into existing facilities, and the full efficiency gains depend on ambient conditions and server room design.