In this module, we’ll explore the most commonly used liquid cooling systems and how they meet different cooling demands; let’s get started.

Several liquid cooling systems are used in data centers, but their underlying infrastructure is similar: most systems operate with two connected loops that exchange heat via a heat exchanger inside a coolant distribution unit (CDU).

The Technology cooling system which captures heat directly from the IT equipment.

The Facility water system (FWS) which carries the heat away from the CDU to release it into the atmosphere or transfer it to a heat reuse system.

But if the infrastructure is similar, why are there different systems? Each system is designed to suit different cooling demands depending on factors like heat density, data center layout, and the amount of heat to be removed.

First there’s direct liquid cooling, also known as direct-to-chip cooling, which is the most widely used method. It cools components like CPUs and GPUs directly by pumping coolant through cold plates made from materials such as copper or aluminium and mounted on the processors.

These plates contain fine microchannels to maximise surface area and heat absorption. A thermal interface material helps conduct heat from the chip into the plate, and the design compensates for uneven heat distribution.

The coolant is usually 25% propylene glycol and 75% demineralised water, with additives to prevent corrosion and biological growth.

It circulates in a closed loop between the chips and a CDU. While this method mainly cools components with cold plates, such as CPUs and GPUs, other parts, like network interface cards, power supply units (PSUs), memory, and storage devices, still rely on air cooling.

Most systems remove 75-85% of the heat via liquid, the rest by air. This hybrid solution improves efficiency and protects sensitive components.

Immersion cooling works differently. It submerges servers in a dielectric, non-conductive liquid that absorbs heat directly from components, providing uniform cooling and eliminating the need for air.

Although immersion cooling is one of the most efficient systems we have, it does have some limitations. The fluids used, often special oils or refrigerants, have lower heat transfer properties than those in direct-to-chip systems. Without precision cooling or forced flow, high-TDP components cannot be cooled as efficiently.

This makes them less effective… for very high heat loads, like GPUs with high thermal design power. Immersion systems also require major infrastructure changes due to the complexity of liquid handling and increased weight. Because of these factors, it hasn’t gained much traction in the industry.

Both direct-to-chip and immersion cooling can be designed as single-phase or two-phase systems.

In single-phase systems, the fluid remains in a liquid state. It absorbs heat from the components and is then pumped through a heat exchanger where it’s cooled before being recirculated.

In two-phase systems, the liquid boils into vapour when heated, rises, condenses on a cool surface, and returns to the system. This phase change boosts heat transfer efficiency.

Two-phase systems can be up to 10 times more efficient than single-phase, but they often use fluorochemical refrigerants, many of which contain per- and polyfluoroalkyl (PFAS) substances. This raises environmental concerns and widespread adoption has been slow as a result.

Liquid cooling systems are becoming more common as data centers aim to support high power densities, reduce energy use, and recover heat. But liquid cooling isn’t one-size-fits-all, choosing the right one depends on the specific needs of the data center.

In the next module, we’ll explore how liquid cooling improves uptime, reduces water and energy use, and how smart control strategies can boost operational efficiency in data centers.