AI data center liquid cooling technology developments

The industry is seeing a shift from traditional air cooling to liquid cooling strategies to manage increasing rack densities caused by artificial intelligence and high-performance computing.

A central component in these direct-to-chip systems is the coolant distribution unit (CDU), which manages heat transfer through pumps, heat exchangers, valves, and sensors. Effective implementation requires a system-level approach that considers heat loads, coolant temperatures, and the physical placement of the CDU to manage piping and space requirements. The CDU acts as an interface between facility infrastructure and the technology cooling system, protecting sensitive components from variations in pressure or water chemistry.

In line with these technological shifts, Midea Building Technologies (MBT) has introduced a power-cooling hyperconvergence architecture. This integrated infrastructure stack combines power systems, energy storage, and liquid cooling CDUs to manage rising power densities and improve Power Usage Effectiveness (PUE) in AI data centers.

Nvidia has expanded its DSX platform to further increase efficiency through a closed-loop liquid cooling system combined with dry coolers. Unlike traditional cooling towers that rely on evaporation, this system transfers heat to the surrounding air, which could potentially reduce facility water consumption to nearly zero. For a 2-gigawatt deployment, Nvidia estimates this could save up to 54 million liters of water per day. Additionally, the DSX MaxLPS software aims to optimize electrical efficiency, potentially using up to 30% less power for the same AI output. Companies including Google Cloud, CoreWeave, and Nebius are planning implementations, supported by partners such as Siemens, ABB, and Schneider Electric.