With the rise in the use of data and the adoption of artificial intelligence (AI), data centres increasingly need to be able to support more power usage.
This provides significant challenges for data centre cooling systems (traditionally an operational consideration) to now be able to balance the competing needs of more power and heat generating software while remaining energy efficient. With the requirements of data centres rapidly changing, what is the best approach to take when choosing a solution?
Air to air cooling is historically the most widely used cooling system in data centres, whereby computer room air conditioning (CRAH) or computer room air handlers provide air conditioning systems, which cool down the warmed air, and circulate it back through the room. Vents are then placed around the room which make sure that cool air is distributed evenly throughout the space, and that hot air can be expelled effectively.
This approach is commonplace in northern Europe, the UK and the Nordic regions, taking advantage of cooler temperatures. There are multiple benefits too, with lower upfront costs and a well-recognised understanding across the industry, meaning repairs, if needed, are faster and simpler.
However, air-to-air cooling has higher operational power requirements compared with newer liquid-based solutions. As rack densities rise, particularly with AI and HPC hardware generating 20–80 kW per rack (and in some cases exceeding 100 kW), the efficiency of traditional air-based cooling diminishes significantly. Large volumes of air must be moved and conditioned, increasing fan and chiller loads, which drives up Power Usage Effectiveness (PUE).
While it remains viable for many enterprise or colocation environments at moderate densities, questions remain as to whether air-to-air cooling can continue to meet the power and thermal demands of next-generation workloads.
Liquid to liquid cooling
This is an emerging system, which uses liquids to remove heat by circulating fluids through complex pipework, and supporting greater equipment densities. There are three types:
- Immersion cooling: servers are placed and completely covered in a non-conductive liquid allowing them to cool down
- Direct to chip cooling: liquid is put in a direct contact with heat generating equipment, absorbing the heat generated and cooling it down. The disadvantage it that it must be combined with air cooling because the technology only cools the components that are major heat sources
- Rear door cooling: A rear door heat exchanger (RDHx) is a cooling system mounted onto the rear door of a server rack. The hot air from the server passes through the RDHx coils and the liquid in these coils cool the air before it re-enters the room or is removed entirely
What are the pros and cons of using this form of cooling? Importantly, liquid to liquid uses less energy, as water is more efficient than air at conducting heat. Using less energy translates into a lower operating cost. Additionally, this style of cooling is quieter, creating a better environment for staff working at data centres. However, although generally there may be cost savings when measured over the longer term, initial expenditure to set this up can be much higher. There is also an increased risk of leaks, which have the potential to damage equipment and cause delays to service for data centre clients.
Note on implementation
The adoption of liquid-to-liquid cooling should be assessed on a case-by-case and site-specific basis, supported by detailed feasibility studies which we can support. These studies aim to help data centre operators and clients evaluate technical requirements, cost implications, integration with existing infrastructure, and long-term energy savings, ensuring that the chosen solution is both practical and commercially viable with the aim of mitigating additional risks.
The importance of a sustainable approach
While these methods vary in a lot of ways, their common ground is a drive towards sustainability. This is more than a nice-to-have, but a vital goal against the backdrop of various international net zero targets, and data centre operators are under increasing pressure to reduce environmental impact. For example, the UK government is aiming to generate enough clean power to meet 100% of its electricity demand by 2030, and Germany will require new data centres to achieve a power usage efficiency of 1.2 PUE or less from 2026.
For data centres operational before July 2026, the law mandates a PUE of 1.5 or lower by July 2027, and a further reduction to 1.3 or lower by July 2030. For comparison, the current average PUE in Germany is 1.46. New data centres must achieve a PUE of 1.2 or less. Data centres with a nominal power connection of 300 kW or more were required to cover 50% of their electricity consumption from renewable sources as of January 2024. This requirement increases to 100% by January 2027.
As I’ve outlined, there is a great number of factors to take into account when deciding on the best method for temperature maintenance for data centres facilities. For this and other common questions issues related to delivering a data centre of all shapes and sizes, Gleeds has a wealth of experience.
Read more about our offering: Data centres.
