Air cooling, direct-to-chip, and immersion
The cooling technologies for data centers provide options for server and rack level cooling and room and facility level cooling, varying by degree of commonality and efficiency.
Three main options are available for server level cooling, including air cooling as the traditional method, which uses fans installed on a server board to provide air flow to cool down heat sources. IDTechEx states that with the increasing thermal design power of graphics processing units (GPUs) and central processing units (CPUs), standalone air cooling could be insufficient, which is where liquid cooling can take over.
Direct-to-chip cooling is the most common type of liquid cooling as of 2024 and 2025, where a cold plate is installed on top of a heat source (typically GPU or CPU). Depending on the type of liquid used inside the cold plate, direct-to-chip cooling can be split into single-phase and two-phase cooling. Single-phase cooling typically uses a water-glycol mixture such as PG25 and two-phase refrigerants. Immersion cooling is the third method for server level cooling and involves immersing an entire server board into a fluid tank. Much like direct-to-chip cooling, there is the option of either single-phase or two-phase immersion, with two-phase offering a significantly higher cooling capacity. However, regulatory concerns around PFAS are still hindering the mass adoption of two-phase as of 2025.
Comparison of cooling methods. Source: IDTechEx
Computer room air handler units and liquid cooling
In addition to the server and rack level cooling, cooling also happens in data center facilities at the room and facility levels. Room and facility level cooling also traditionally employs air cooling as a main approach. Methods include a rear door heat exchanger at the back of a server rack, computer room air conditioning, computer room air handlers, and a sidecar heat exchanger. Computer room air handler units are a hybrid of air and liquid cooling that use cold water from pump rooms to cool down the perimeter of a computer room. They work by absorbing hot air from a computer room and releasing cooled air back into the room to maintain optimum temperatures.
Liquid-to-air cooling is currently a common approach, as liquid-cooled server racks are beginning to replace air-cooled server racks. Liquid-to-air cooling in this context indicates that liquid (typically direct-to-chip) cooling happens on the server level, and air cooling happens on the facility level, where cold air is used to cool down the hot water. However, this is not the most efficient method. Therefore, there has been a trend towards liquid-to-liquid cooling, with liquid-to-liquid cooled data centers predicted to grow from 1% in 2024 to 30% by 2026, demonstrating much interest and scope for developments within the sector.
IDTechEx’s future outlooks
As GPU TDPs increase, advanced cooling systems become more necessary to avoid overheating and to maintain system efficiency, which can provide challenges for thermal management. Two-phase cooling is likely to rise, as this approach could provide greater success and efficiency for better managing increasing temperatures. For data centers, efficient heat dissipation is vital to see long lifetimes and well-maintained GPU performance.
IDTechEx predicts that GPUs with a TDP of over 1500W could be expected over the next year in 2025, with the management of energy costs becoming slightly trickier. Thermal management and cooling systems will therefore need to become a main consideration in data center designs to enable efficient cooling, enhanced thermal efficiency, and reduced energy waste.
For more information, visit IDTechEx’s report, www.IDTechEx.com/TMDC.
For the full portfolio of thermal management market research available from IDTechEx, please see www.IDTechEx.com/Research/TM.