How AI is shaping the future of data centre cooling

Data centres are the invisible architects of our digital existence, underpinning everything from cloud computing to complex AI applications. The UK government has announced its ambition to make the UK a world leader in Artificial Intelligence, which will likely mean – you’ve guessed it – more data centres.

This immense capability, however, comes with a significant environmental footprint. The International Energy Agency estimates that data centres currently consume approximately 1% of the world’s total energy. In the UK alone, data centres use roughly 2.5% of the country’s electricity, according to the National Energy System Operator (NESO). And the cooling systems can consume 30-40% of a data centre’s energy, making them an essential area for improvement.

As the AI era leads to sustained global demand for computing power and cooling needs, identifying and implementing more efficient, sustainable cooling solutions is crucial. Fortunately, AI is both the growth driver and part of the solution for energy management and optimisation.

AI offers a transformative approach to this problem by enabling intelligent, adaptable and automated cooling systems that improve performance while reducing energy waste. These systems use machine learning algorithms to continuously study thousands of data points across the data centre environment, creating dynamic thermal models that predict and prevent hotspots before they occur.

AI-driven cooling management brings predictive capabilities. Rather than simply reacting to temperature changes, these systems anticipate cooling needs based on historical patterns, scheduled workloads, and even time-of-day usage trends. For instance, an AI system might detect that certain AI training jobs consistently spike GPU temperatures on specific server racks and automatically adjust cooling resources before the job begins rather than after temperatures rise. These systems can dynamically adjust cooling strategies based on real-time data, including server workloads, external weather conditions, humidity and temperature.

Beyond immediate energy savings, AI-driven cooling systems can also provide valuable insights for long-term infrastructure planning. By understanding cooling efficiency patterns over time, data centre operators can make informed decisions about equipment placement, server consolidation, and future cooling investments.

AI-driven thermal management is most effective for optimising efficiency when it is paired with a blend of cooling solutions. In this context, AI helps to manage and orchestrate a mix of air- and liquid-based cooling solutions to optimise operations and better manage energy use.

Air cooling has been the backbone of data centre thermal management for many years. Using fans, heat sinks and strategic airflow, these systems dissipate heat from critical components. Approximately 99% of data centres still use some form of air cooling, and advances such as hot/cold aisle configurations and intelligent fan control have significantly improved efficiency.

Direct Liquid Cooling (DLC) utilises small heat exchangers, known as cold plates, to deliver liquid directly to heat-generating components, thereby handling thermal loads with lower energy spikes. Its superior thermal conductivity can efficiently dissipate heat, particularly with higher-density server configurations, and it also reduces noise by minimising fan use. Dawn, the UK’s fastest AI supercomputer, co-designed with Dell, Intel and the University of Cambridge, for example, is equipped with advanced liquid cooling and GPU efficiencies. When tested, Dawn achieved a PUE of 1.14 – a testament to how AI-driven cooling can cut energy consumption while boosting computational performance.

While it is possible to put cold plates on every heat-generating component in a server, it can be expensive, complex and impair the system’s serviceability. Hybrid-DLC offers a solution by combining direct-to-chip liquid cooling for high-priority components (like CPUs or GPUs) with air cooling to manage ambient temperature and cool other components like memory.  These racks of dense compute can still generate intense air heat loads up to 80 kW per rack, meaning how one deals with the hot air remains critical to help manage the full environment.

The growing energy demands of data centres, particularly with the rise of AI, have led to an urgent need for cooling innovation. While traditional air cooling has served its purpose, the future undeniably lies in advanced solutions, such as direct liquid and, more powerfully, hybrid cooling systems. By embracing AI-driven optimisation and cutting-edge thermal management, businesses can build a digital future that is not only powerful and efficient but also responsible.