Free cooling has a temperature ceiling: what rising temperatures mean for chiller performance

How cooling architecture affects free-cooling availability, chiller use and data centre resilience during extreme heat.


Free cooling helps data centres reduce the energy required to manage heat. But its effectiveness depends partly on conditions outside the building. As ambient temperatures rise, the hours available for free cooling may shrink, increasing reliance on chiller and other equipment to handle the thermal load.

For operators of AI and high-performance computing (HPC) infrastructure, the question is not whether free cooling works. It is how the cooling system performs when outdoor conditions move beyond its most favourable operating range.

Why free cooling has limits

Free cooling uses suitable outdoor conditions to reject heat with little or no compressor-based refrigeration. Depending on the design, it can involve outside air, heat exchangers, cooling towers or dry coolers. As the temperature difference between the coolant and the external environment narrows, transferring the required amount of heat becomes more difficult. Additional cooling may be needed to maintain the same conditions for IT equipment.

There is no universal free-cooling temperature ceiling. The available hours depend on coolant temperature, heat-exchanger performance, humidity where relevant, and system load. Liquid cooling does not automatically remove these constraints: heat still has to leave the facility.

How extreme heat affects chiller performance

In hybrid installations, rising ambient temperatures can reduce free-cooling capacity and increase reliance on mechanical refrigeration. Air-cooled chillers may operate at higher condensing temperatures. Depending on the equipment and operating point, efficiency can fall and available cooling capacity can decrease. Compressors and fans may consume more electricity even when the IT workload remains unchanged.

This can raise daily Power Usage Effectiveness (PUE), increase operating costs and reduce the thermal margin available for peak workloads. If cooling becomes insufficient, operators may need to consider derating: limiting equipment performance or workloads to remain within safe thermal limits. These outcomes are not inevitable. Equipment selection, control strategies, redundancy and capacity margins all matter. What operators need to understand is how the cooling architecture behaves during prolonged heat, not merely its efficiency on an average day.

Europe’s 2026 heatwaves: an infrastructure warning

On 12 August 2026, the Copernicus Climate Change Service reported that western Europe had experienced its warmest June–July period on record. The regional mean reached 21.62°C, 2.79°C above the 1991–2020 average. Météo-France also recorded 44.3°C at Pissos during the June 2026 heatwave.

These figures do not demonstrate failures at particular data centres. They illustrate the outdoor conditions against which long-lived cooling infrastructure must be assessed. Consecutive hot days and warm nights are especially relevant because they can reduce the periods previously available for free cooling.

54%

of global data centre is located in markets facing elevated chronic heat or drought stress

Resilience is not just a local concern

First Street’s report of 18 June 2026 analysed climate exposure across 97 global data centre markets. It found that approximately 54% of global data centre capacity is located in markets facing elevated chronic heat or drought stress. The figure is not a measure of heat exposure alone. Nevertheless, it shows why cooling resilience must account for geography and changing environmental conditions. Operators need to assess how many free-cooling hours remain available, the demands placed on backup systems and the cooling headroom needed as rack power densities increase.

How direct to chip cooling changes the equation

Cooling resilience depends on two linked tasks: collecting heat from the processors and rejecting it outside the building. Improving the first does not automatically solve the second. In Quattro’s patented two-phase direct-to-chip cooling technology circulates a dielectric fluid through cold plates. As the fluid absorbs heat, part of it changes from liquid to vapour, helping transfer thermal energy away from the processors. The distinction becomes particularly relevant at the heat-rejection stage. The system can accept facility (secondary) water at inlet temperatures of up to 45°C, which allows a configuration with dry coolers and without chillers or evaporative cooling towers.

Because the facility water can run warmer, dry coolers have a larger temperature difference to the outdoor air to work with. Compressor-free cooling may therefore remain possible at higher outdoor temperatures and for more hours of the year, which may reduce exposure to the efficiency penalties of chiller operation during extreme heat. The 45°C figure is the temperature of the water entering the system, not of the outdoor air. Actual free-cooling hours and delivered capacity depend on local climate, system design and operating load. Dry coolers also require electrical power, notably for their fans.

Design for the difficult hours

Chillers remain appropriate in many data centre architectures. The central issue is understanding when they are needed, how their performance changes with the weather, and whether sufficient cooling remains available for the IT load. As computing density increases, resilience cannot be assessed through nominal capacity alone.

Free cooling has a temperature ceiling. The question for operators is how many hours a year their cooling architecture stays within it, and what remains available when it doesn’t. An architecture that accepts warmer facility water gives dry coolers more room to work. This does not eliminate climate-related risk, but it can change how much of the year a facility is exposed to it.

Editorial sources: