Hotter summers are exposing a resilience gap in data centres built for a cooler climate. Future-weather modelling, airflow analysis and live performance data can protect uptime without driving unnecessary energy and water use. Colin Rees, co-lead of consultancy at IES, discusses further
The race to expand AI infrastructure is usually framed as a competition for chips, electricity and grid connections. Recent heat across the UK and Europe has exposed another constraint: the conditions in which data centres must operate. Servers run continuously, yet their cooling systems face the hardest test when outdoor temperatures rise and heat rejection methods become more difficult. Facilities designed around historical weather assumptions may face higher operating costs, reduced cooling headroom and greater operational risk over their lifetime.
Demand is also accelerating. The International Energy Agency (IEA) expects European data-centre electricity consumption to increase by more than 45 TWh between 2024 and 2030, a rise of over 70%. Nearly all electricity used by IT equipment ultimately becomes heat that must be removed. Higher-density AI workloads therefore create a much larger cooling duty alongside their power demand.
This growth is taking place on the world’s fastest-warming continent. The European State of the Climate 2025 report found that at least 95 per cent of Europe experienced above-average annual temperatures last year. In the UK, recent research by the Met Office concluded that a heatwave comparable with the summer of 1976 would already be around 3°C hotter in today’s climate. The weather assumptions behind long-lived digital infrastructure are changing during its operating life.
Yesterday’s weather is a poor design brief
Data centres have traditionally been assessed against historical weather files and a limited number of design conditions. That approach can miss sustained heat, warmer nights and combinations of events that place the greatest strain on cooling plant. A facility may perform adequately on an annual average while holding very little spare capacity during the hottest afternoon of the year. Without climate resilience testing, hidden capacity constraints can remain undiscovered until an extreme event exposes them, leaving operators with reduced resilience and greater exposure to equipment throttling, service disruption or emergency intervention.
Project teams should use future climate files from the earliest design stages and test a range of plausible conditions across the expected life of the facility. Hourly simulation can show how outdoor temperature, humidity, IT load, controls and equipment performance interact. Stress tests should also cover prolonged hot periods, maintenance outages and the loss of a fan, pump or chiller. Resilience depends on how the whole system behaves when several pressures arrive together.
Cooling resilience starts outside the server room
The cooling system extends beyond the racks and plant room. High ambient temperatures can reduce chiller capacity, while heat-rejection equipment can discharge warm air that is drawn back into its own intakes. Site geometry, nearby buildings, screening and local wind conditions all influence this effect. Recirculation raises inlet temperatures and forces cooling equipment to work harder, increasing energy use and operating costs. During extreme weather, it can reduce available cooling capacity and erode the plant’s safety margin, leaving less room to absorb a fault or a sudden increase in IT load.
Building performance simulation and computational fluid dynamics (CFD) answer different parts of the problem. Building simulation establishes how IT demand, cooling plant, controls, weather and energy use interact over time. CFD maps the movement of air and heat in much finer spatial detail, revealing hotspots, airflow short circuits and the effect of equipment layout. Used together, they allow teams to assess the server room, cooling plant and surrounding site as one connected thermal system.
Dense AI racks shrink the margin for error
Liquid cooling can remove heat from high-density racks more effectively, although the heat still has to pass through pumps, heat exchangers and facility-side rejection equipment before it reaches the outdoor environment. Each link needs enough capacity under peak conditions. Adding higher-density equipment to an existing data centre can expose bottlenecks in pipework, electrical infrastructure or cooling plant that were invisible at the original design load.
Physics-based models let operators test upgrades before committing capital or interrupting live operations. They can compare plant configurations and future workloads, then identify the intervention that provides the required resilience with the lowest additional energy use. That evidence helps avoid undersizing, which leaves the facility exposed at peak conditions, and oversizing, which ties up capital in cooling capacity that may rarely be used. Investment can then be directed towards the components and operating scenarios that govern performance.
Annual efficiency can hide the dangerous hour
Power Usage Effectiveness (PUE) remains an important measure, yet an annual figure can mask the critical few hours that determine whether cooling capacity is genuinely resilient. Operators need to see the distribution across the year, the peak value during extreme weather and the performance of the facility during credible fault conditions. Water Usage Effectiveness should be assessed alongside it, particularly in regions where hotter, drier summers may constrain water availability.
Every cooling strategy carries a different balance of electricity, water, capital cost and operational risk. Those trade-offs vary with local climate and the design of the facility. Modelling them against the same weather and workload scenarios creates a defensible basis for investment and prevents a gain in one metric from concealing pressure elsewhere.
Make the design model part of operations
The model should remain useful after handover. A performance digital twin can connect the design calculations with live information from meters, sensors, the building management system and local weather data. Operators can compare predicted and actual behaviour, investigate emerging deviations and test control changes before applying them to the facility.
This can expose performance drift such as a failing sensor, poor setpoint, clogged filter or equipment operating outside its intended sequence. Earlier diagnosis gives maintenance teams a chance to act while the facility still has headroom. Future expansion decisions can then use measured operating behaviour alongside the original design assumptions.
Build climate resilience into AI infrastructure
Europe can expand AI capacity while protecting uptime and controlling its environmental impact. Developers, operators and planning authorities will need credible evidence that new facilities can withstand their likely climate, including the worst hours as well as the average year.
Future-weather simulation, detailed airflow analysis and performance digital twins provide that evidence. They help teams locate risk early, target cooling investment and keep checking assumptions once the data centre is live. As AI places more heat into increasingly dense facilities, climate-ready cooling will become one of the foundations of Europe’s digital resilience.
The post Europe’s Artificial Intelligence cooling problem appeared first on Planning, Building & Construction Today.
