More data means more cooling
17 November 2025
Artificial intelligence is turning up the heat on data centres, with plans underway to integrate the technology through every sector in the UK. Tim Mitchell explores how data centres can rise to the challenge with energy-efficient and future-proof cooling systems.
In January 2025, the government published its ‘AI Opportunities Action Plan’, a set of 50 recommendations – accepted in full by the government – designed to position the UK as a leader in AI, accelerating growth through all sectors and harnessing its potential for the benefit of the British people. From healthcare to manufacturing, AI is not only reshaping industries, it is also placing unprecedented pressure on the infrastructure that underpins it.
Take a ChatGPT query as an example. It uses 10 times more electricity than a simple Google search, according to the International Energy Agency (IEA). The latest models, such as GPT-5, include advanced capabilities which put even more pressure on data centre infrastructure. AI processing is power-hungry and resource-intensive, forcing data centres to rethink how they manage energy, efficiency and cooling to meet expected growth over the next decade and beyond.
Turning up the heat
AI workloads push servers to run at much higher power densities than conventional applications, driving up energy consumption and generating far more heat. A recent analysis by data scientist Alex de Vries-Gao, estimates that AI power consumption could reach 23 gigawatts (GW) by the end of 2025. Comparatively, this is huge – around twice as much as the Netherlands, which reaches 11-12 GW at peak electricity consumption, and three times more than London (8 GW).
Average rack densities are forecast to rise from around 36 kW per rack in 2023 to about 50 kW by 2027. Meanwhile, the GPUs (graphics processing units) used for AI training are becoming significantly more demanding, increasing from around 700 watts today to as much as 1,200 watts in the next generation.
Many existing cooling systems were never designed to handle such densities, and in many cases, they are already operating close to their limits. As a result, operators are being forced to adapt, retrofit or redesign much sooner than planned.
Cooling under pressure
Cooling makes up around 40% of data centre power consumption meaning inefficiencies have a direct impact on costs and carbon emissions. Air-based cooling has long been the industry standard, but its ability to handle rising thermal loads is diminishing. Air simply doesn’t have the thermal conductivity required to efficiently remove large volumes of heat from increasingly compact and power-dense server environments.
The sector is therefore moving rapidly towards liquid-based solutions, which can manage higher thermal loads far more effectively. Those who invest early in the right cooling technologies will be better positioned to handle future demand without compromising efficiency or reliability.
Why liquid cooling?
By bringing the cooling medium directly to the heat source, technologies such as direct-to-chip, cold plate or full immersion cooling can dissipate heat much more efficiently than air. These systems are particularly suited to high-density AI workloads where temperature control is critical to both performance and hardware longevity.
In data centres, liquid cooling also enables higher server density by reducing the need for bulky airflow pathways, allowing more computing power in the same footprint. Additionally, liquids have a higher thermal conductivity than air, meaning they can remove heat faster and maintain stable temperatures under peak loads, resulting in lower energy consumption.
For modern data centres handling AI, high performance computing or hyperscale workloads, liquid cooling is increasingly the preferred solution due to its unmatched efficiency, scalability and ability to prolong hardware life.
Challenging retrofit
Hybrid systems are also gaining traction, particularly in facilities where full conversion to liquid cooling may not be viable. Hybrid systems combine traditional chilled water or air-based systems with liquid cooling at hotspots, creating a flexible, scalable solution that allows operators to adapt as demand grows.
Alternatives like our ultra-efficient Quantum chillers from Engie are another option for data centre retrofitting. The use of Turbocor magnetic-bearing, oil-free compressors mean the units are quieter, more resilient and lighter than conventional chillers for ease of installation, while reducing energy consumption by 30 – 50%.
When coupled with free-cooling and energy storage technology, these solutions can offer unrivalled advantages in energy saving and resilience. They can also form part of a system to enable chillers to become a revenue stream via solar firming, demand side response, or simple peak load lopping.
Importantly, the move towards advanced cooling also opens the door to heat recovery and reuse. The by-product of cooling — waste heat — doesn’t have to be wasted. Modern systems can capture and repurpose this energy for district heating networks, providing warmth for nearby homes, hospitals, leisure centres or manufacturing sites. Not only does this contribute to wider decarbonisation efforts, but it also improves overall energy efficiency and helps to offset operational costs.
For existing sites, retrofitting can be complex, but it’s rapidly becoming a priority to prevent legacy data centres from becoming stranded assets. Working with experienced partners ensures that transitions are well-managed and downtime is minimised or avoided altogether. By adopting the latest cooling technologies and techniques, the data centre industry can build a secure, energy efficient and resilient network of digital infrastructure, ready to support the UK’s mission to become a leader in AI.
