Advanced cooling technologies are changing data centre design decisions

Cooling strategy is becoming one of the earliest design decisions in high-density data centre projects, as rack densities rise and operators look for systems that can support different workloads within the same facility.

The issue is no longer limited to choosing between air and liquid cooling. Designers need to decide when a cooling approach should be fixed, how much flexibility can be retained during early design, and what operational requirements will be introduced once more advanced systems are built into the facility.

Rack-level heat density is increasing as data centres support more high-performance computing, including GPU-based infrastructure and other accelerated workloads. Traditional enterprise and cloud environments have commonly operated at around 5–15kW per rack, while higher-density requirements are now regularly moving to 30kW and above, with some reaching 50–100kW per rack and beyond.

Once that level of power and heat is concentrated into a smaller footprint, cooling begins to influence the wider building design. Data hall layout, plant space, technical corridors, structural loading, distribution routes, commissioning and long-term asset flexibility all become affected by the cooling route selected.

Air-based systems remain important for lower-density environments. Below around 15–20kW per rack, fan wall units, computer room air handlers (CRAHs) and hot/cold aisle containment continue to support many facilities effectively, with dry air or evaporative heat rejection used depending on the site and climate.

There is no fixed threshold at which air cooling stops being viable. Local climate, energy restrictions, site constraints, CapEx and OpEx priorities all influence the decision, as does the level of future flexibility the operator needs to preserve. In many data centre applications, however, air-based cooling begins to become less attractive once power densities move beyond roughly 15–25kW per rack.

Across a typical row of 24–40 racks, that can equate to around 600kW. At this level, the air volumes required become harder to manage without compromising spatial efficiency, while air velocity through racks, pressure losses, fan energy, noise and uneven air distribution all become more difficult to control.

Air cooling can often still be engineered at these densities, but the question is whether it remains the best use of space, power and capital once the wider building constraints are taken into account.

For high-density environments, direct-to-chip liquid cooling is currently the most widely adopted liquid-based approach. By removing heat closer to the source, it reduces reliance on moving large volumes of air through the data hall, which is why cooling distribution units and technology water loops are now being considered much earlier on projects where high-density zones are expected.

Immersion cooling offers strong heat transfer performance, although it is not yet being adopted at the same scale in conventional commercial data centre projects. Its use often requires changes to operational procedures, maintenance practices and supply chains, making direct-to-chip a more practical route for many operators at present.

Hybrid cooling strategies are becoming more common as operators try to retain flexibility across the data hall. These allow lower-density air-cooled loads and higher-density liquid-cooled loads to sit within the same facility, which matters while customer requirements, hardware roadmaps and cooling technologies are still developing.

A hybrid hall cannot be treated as a conventional hall with liquid cooling added in selected areas. High-density zones need to be deliberately planned, with cooling and power infrastructure concentrated in the right locations. Cooling distribution units and technology water loops serve the liquid-cooled zones, while air-based systems support lower-density rows.

This zoning is determined by hall geometry, airflow distribution, resilience requirements and the ability to maintain operating temperatures during failure conditions. Computational fluid dynamics modelling now needs to inform the design earlier, helping teams test normal and failure scenarios, including outages affecting CRAHs or fan wall units.

During due diligence and early RIBA Stage 2, the priority is usually to keep options open. Design teams will assess a range of cooling approaches across a broad vendor base, often allowing for the most onerous equipment footprints and plant replacement requirements so the scheme remains vendor-agnostic for as long as practical.

That flexibility has to narrow before detailed design. By the end of RIBA Stage 2, the cooling approach needs to be fixed if the project is to progress with confidence. The chosen system determines plant allocations, technical corridors, primary distribution routes, structural design loadings and permitting requirements.

Leaving the decision too late can appear to preserve options, but it often increases the risk of redesign once other systems are fixed. In high-density facilities, cooling is too closely linked to the physical and operational model of the building to be treated as a late-stage package decision.

Water use also influences the choice of system. From a thermal perspective, water is an effective medium for heat transfer and can help reduce mechanical cooling loads and improve power usage effectiveness. That does not mean water-based heat rejection is always the right answer.

Developers increasingly have to balance power usage effectiveness against water usage effectiveness. Water availability, abstraction limits, discharge constraints, local climate, access to potable or non-potable water and the potential for rainwater harvesting all affect whether water-based heat rejection is viable. In water-scarce locations, every cubic metre of water use may need to be justified.

Operational and public health factors also need to be considered. Adiabatic and evaporative systems can support efficiency, but they introduce maintenance requirements, including water treatment and legionella control. In some locations, modern air-cooled chillers and dry air coolers, particularly those using free cooling and magnetic bearing compressors, may offer a more appropriate balance between efficiency, resilience and water use.

Where some water availability exists, hybrid air and evaporative systems can provide a useful compromise. These systems can operate in dry mode for much of the year and use water during peak ambient conditions, balancing energy and water performance rather than optimising one metric at the expense of the other.

Liquid cooling brings further design, installation and operational considerations. Leakage remains an obvious concern, particularly where distribution systems or components are not installed and maintained correctly. Installation quality becomes more important, and contractors and operations teams need the right training to manage coolant systems, cooling distribution units and distribution networks.

Compared with conventional air-cooled environments, there is also limited operational field experience. As more high-density data centres enter service, operating teams are likely to influence future system design more strongly, particularly around access, maintenance, isolation, monitoring and coolant management.

Standardisation remains another constraint. Direct-to-chip cooling is becoming more established, but vendor approaches still vary, creating integration challenges and increasing the risk of lock-in if flexibility is not protected during design and procurement.

Cooling strategy now affects technical performance, programme, water use, operational readiness and long-term asset value. The decision is no longer based only on which technology removes heat most efficiently. It is about selecting an approach that allows the facility to support higher densities without reducing resilience, limiting future adaptation or introducing operational issues that have not been fully accounted for.

As rack densities increase, advanced cooling technologies will form a larger part of the design response. Projects that retain flexibility during early design, while fixing the cooling strategy before detailed design, will be better placed to protect programme, resilience and long-term asset value.