07 October 2025

Lauri Salmia, Applied Portfolio Manager, Trane

When it comes to preparing for AI-ready data centres, there is no one-size-fits-all cooling solution. The ongoing revolution in AI, machine learning, and high-performance computing is rapidly transforming data centre operations and demands, changing and placing unprecedented pressure on cooling infrastructure. The diversity of workloads, ranging from traditional storage and enterprise applications to exceptionally high-density AI/GPU compute environments, requires a flexible, scalable approach.

Historically, air-cooled systems have served most data centres well. Air- and water-cooled chillers, combined with new ranges of Computer Room Air Handlers (CRAH) and Fan Wall Units (FWU) have offered robust coverage for facilities with low to moderate server rack densities and remain highly relevant for legacy applications and lower intensity operations also in the future.

However, the surge in AI workloads has resulted in server rack densities which can be tens of times higher than previous generations, which is an obvious challenge for cooling solutions and where traditional air cooling reaches its limits. For this type of serve rack densities, liquid cooling solutions are often required. The most common method today is direct-to-chip (DTC) cooling, where cooling liquid is looped directly via the cold plates of the server racks to remove the majority of the heat. But even in these applications, air-cooling solutions are still needed to reject the remaining residual heat.

Adopting liquid server rack cooling means that chilled water loops can be run at higher temperatures. This significantly increases the potential for free cooling — using ambient air, rather than higher-energy intensive compressor-based cooling. In temperate climates like the UK’s, free cooling can be used most of the time during the year and meet big part of the annual cooling delivering significant energy and cost savings.

There are now integrated systems available in the market combining intelligently managed dry coolers and chillers. In this case, mechanical and free cooling part is optimized to minimize the annual energy consumption and data centre PUE (Power Usage Effectiveness) factor. Compared to conventional free cooling chillers, the integrated Chiller/Dry Cooler approach can enable a further double digit reduction in energy consumption without increasing the unit footprint.

Planning for tomorrow

For data centre operators, planning for tomorrow is as essential as meeting today’s requirements. Retrofitting facilities for AI means adopting flexible, modular cooling systems that support both existing and future workloads. Key considerations include:

• Planning for expansion: Build in extra capacity and physical space for future increases in rack densities and cooling loads. The ability to scale quickly is crucial, particularly for hyperscale and co-location sites where requirements may change unexpectedly.
• Embracing mixed technologies: Get ready to use both air and liquid-cooled systems for current and anticipated future workloads. Trane’s comprehensive portfolio allows future proofing your data centres’ cooling needs.
• Maximising energy savings and efficiency: Global power demand for data centres will grow exponentially in the coming years, and power availability will be one of the bottlenecks for scaling. Big part of the data centre power is used for cooling which means that the future cooling systems needs to be even more energy efficient than today.
• Proactive heat reuse planning: Traditionally data centre excess heat has been extracted back to the ambient. This excess heat is a valuable energy source which can be repurposed and used in many heating applications increasing the total efficiency ratio of a data centre significantly. When constructing new facilities, choosing a location near heat end-users or established infrastructure should be a top priority to make heat recovery viable and economical. For retrofits, it becomes viable to capture and repurpose heat if there are means to physically transport it to the heat end user.
• Collaboration and circularity: Partnering with flexible solution providers is critical, whether updating existing systems or building from scratch. A holistic approach, where cooling, free cooling, and heat recovery solutions are all managed under one roof, ensures operators can create a complete circular load system and stop wasting energy.

In summary, operators should avoid rigid, over-specified system designs focused only on present requirements, as well as neglecting heat recovery, especially in regions where infrastructure links to heat end-users can be built or improved at reasonable cost.

Underestimating future demand or ignoring regulatory shifts risks expensive retrofits or loss of competitive edge.