NVIDIA has launched a new line of AI servers that rely on liquid cooling at 45°C, a temperature the company says will dramatically reduce the energy and water needed to run hyperscale data centers. The systems are designed to handle the intense heat from training and deploying large AI models while lowering operating costs.
Warmer coolant, cooler costs
Most data centers cool servers with chilled air or liquid that runs at temperatures well below 45°C. NVIDIA's approach uses warmer coolant, which requires less energy to pump and chill. The system also cuts water consumption because it doesn't rely on evaporative cooling towers as heavily. A 45°C loop is easier to integrate with existing facility systems and avoids condensation issues that can damage hardware.
Why hyperscale operators are watching
Hyperscale data centers — the massive facilities run by cloud providers and big tech companies — consume huge amounts of electricity and water. As AI workloads multiply, that consumption is rising fast. NVIDIA's new servers aim to make those facilities more efficient without sacrificing performance. The company says the technology boosts overall efficiency of AI data centers, though it hasn't released specific numbers. Even modest improvements can translate into millions of dollars in savings for large operators.
A boost for AI performance
AI training and inference push GPUs to their limits, generating heat that can throttle performance if not managed properly. Liquid cooling allows for denser server configurations and more consistent operation. By keeping chips at optimal temperatures, NVIDIA's 45°C system helps GPUs run at full speed for longer periods. That means faster model training and more reliable inference, which is critical for applications like real-time language translation and autonomous driving.
The new servers are available through NVIDIA's partner network. The company didn't say which customers have already adopted the technology or when it will be deployed at scale. As AI data centers grow in size and number, the pressure to reduce their environmental footprint will only increase. Whether the 45°C approach becomes the new standard will depend on how easily it scales and how much it actually saves in real-world deployments.
