How AI Data Centres Improve Operations

The development of AI is closely related to the development of data centres. That’s because AI services require increasing numbers of powerful AI data centres to run and because AI also enables those selfsame data centres to work more efficiently and sustainably. Does AI need data centres? Sure. But it’s fair to say data centres also need AI. This guide explores the development of the data centre industry and how AI is helping make data centre operations more efficient and sustainable – even as they support the growing power and sophistication of AI itself. Read on to find out more.

AI is both a driver of data centre development and a means of making data centres more efficient. At the end of 2024, there were more than 1,300 hyperscale data centres (massive computing facilities that handle enormous amounts of information) worldwide. That’s double the number of hyperscale facilities in 2019. Generative AI is a major driver of hyperscale data centre growth, with facilities built by tech giants such as AWS, Google, and Microsoft. Synergy Research Group forecasts that total hyperscale data centre capacity will double in the next four years, with some 130 – 140 hyperscale centres added each year.

It’s not just hyperscale data centres that are booming. Co-location data centres (where businesses rent space to house IT infrastructure) and edge data centres (smaller, localised data centres situated closer to customers than traditional centralised facilities) are also expanding to support generative AI, complementing the efforts of the hyperscalers.

The rise of data centres sucks up energy (and water). Goldman Sachs expects total global data centre power demand to rise from about 55GW in 2023 to 84GW by 2027. About a fifth of this electricity demand comes from AI.

Why do AI data centres use so much energy? Key factors are that data centres run 24-7. The IT equipment they contain – servers, processors, and AI chips, along with storage devices and networking equipment – sucks up power. Then there’s the need for cooling systems such as air con to keep servers running, and power conversion and distribution, where electricity losses occur as energy is converted to meet the specific needs of IT hardware.

Add in the need for back-up and redundant power and the fact that AI training is more intensive than traditional computing, and it’s easy to see why data centres are huge consumers of power. Data centres also consume a lot of water, because water is one of the most efficient and cost-effective ways of keeping systems cool, especially at hyperscale facilities. A single facility may use millions of litres of water a day for cooling.

All of this means engineers are looking for ways to make data centres more sustainable, and – somewhat ironically given the demands it puts on data centres – AI has an important role to play in doing this.

Artificial intelligence can be used to make data centre operations more efficient, helping to cut energy use. For example, rather than uniformly cooling a system at a certain rate throughout the day, AI enables cooling to be dynamic and responsive to environmental conditions, saving energy. Using AI and networks of sensors, it’s possible to monitor temperature, humidity, and airflow to target cooling much more precisely.

AI can also make data centres more efficient and sustainable by scheduling workload for times when electricity is cheaper or is being generated from renewable sources such as wind and solar. Alternatively, loads can be spread so that some servers are not sitting idle while other systems strain to meet demand. It’s also possible to improve efficiency at a software level by optimising how AI models that suck up resources are stored and loaded, or by using data compression techniques to cut the computational demand posed by AI.

Meanwhile, data centres are increasing the efficiency of their back-up power supplies such as uninterruptible power systems using AI, helping cut losses of energy when these systems are idle. AI is also helping data centres to interact with the grid more efficiently to not only use more clean power but also shift demand outside the peak.

One of the well-known examples of using AI for data centres is smart cooling from Google, which teamed up with its AI research unit DeepMind to improve cooling efficiency at the company’s hyperscale data centres. AI took over cooling fans, pumps, chillers, airflow systems, and heat exchangers while receiving inputs from hundreds of sensors that monitor temperature, loads on servers, humidity, pressure, and so on. The result was a 40% reduction in the energy required for cooling. This type of application of AI overcomes some of the issues with data-centre cooling – such as using broad setpoints to maintain temperature, which often means cooling overcompensates for environmental conditions.

That’s not all. Facebook owner Meta has developed an AI-powered predictive thermal control system to dynamically manage airflow and cooling at its data centres. The system works by using a digital twin of Meta’s data halls, which simulates factors such as server heat, rack and room temperatures, and airflow. Using machine learning, this simulation is used to predict thermal conditions based on weather patterns, heating, ventilation, air conditioning configurations, and loads on servers. The AI adjusts fan speeds, airflow routes, and cooling setpoints dynamically in response to the predictions of the model. The approach is said to have reduced the energy required for cooling by up to 15%. One of the side-effects has been allowing more densely packed racks, which cuts the physical footprint required in a data hall, also boosting efficiency. If you want to learn more about innovation and AI, read our guide, and find out more about which jobs are likely to be impacted by automation.

Finally, Microsoft is also using AI to optimise cooling. The tech giant’s sustainable data centre design uses AI to coordinate cooling, airflow, and energy source selection.

Since artificial intelligence can be used to predict failure of hardware before it becomes a critical problem, it is also helping companies to maintain and replace systems before they fail or use up more energy than they should. Faulty equipment such as fans or power supplies can drain energy, but predictive maintenance enables engineers to intervene to minimise the impact on the data centre’s power use. A typical predictive maintenance regime at a data centre analyses data from sensors to identify patterns and foresee when components will fail. This enables timely repairs to be carried out, avoiding unplanned downtime and wastage of energy and other resources. For example, sensors may be installed in servers, HVAC, or power systems. These sensors detect everything from overheating of a server, to signs of early wear on fan systems, or blockages in filtration equipment. AI analyses normal working conditions and patterns to enable it to recognise anomalies in performance. A predictive maintenance system will then alert engineers to the need to maintain or replace equipment before problems worsen or pieces of kit fail altogether.

Companies adopting data centre predictive maintenance regimes include Microsoft, which uses AI to predict battery health degradation, air filter clogging, and cooling system wear, while Google uses predictive maintenance techniques to detect overheating graphics processing units, clogged fans, and server rack airflow problems.

AI data centres are power and water-hungry beasts – and overall, that beast is getting bigger. But AI holds out the promise of mitigating the impact of data centres on the environment and making them cheaper to run. Traditionally, data centre management has been static and reactive rather than dynamic and proactive. AI is helping to change that by dynamically adjusting key parts of operations such as cooling systems to make them more efficient, or using more power from renewable energy sources. Data centres consume vast amounts of electricity, so making them less reliant on fossil fuel sources of power is vital to increasing sustainability. The most efficient data centres work to reduce power per computation, which helps cut the demand the centre makes on the electricity grid. This helps to reduce the carbon footprint of the data centre. And data centres that use water efficiently – through dynamic cooling systems, for example – are also helping to benefit the environment.

While this is all good in terms of sustainability, hyperscalers and other operators of data centres can also drastically reduce cost by making improvements to efficiency using technologies such as AI. Energy usage and cooling represent more than 50% of the cost of running a data centre. Using AI to improve both benefits the operators and sees cost reductions passed on to data centre customers. AI is also helping ensure data centres endure less downtime through techniques such as predictive analytics and maintenance – benefiting operators and customers alike, as well as operators of critical national infrastructure that rely on data centres.

Managing data centres more efficiently also makes it more likely that the grid copes with the continued expansion of computational power. This has important implications in terms of making the benefits of data centres available to as many people as possible, and not just the preserve of developed countries.

Data centres of the future will be more sustainable and efficient than today’s designs. That means making greater use of renewable energy sources such as wind power and solar and adopting intelligent cooling systems, many of which will be powered by AI and machine learning. Data centre operators will also need to carefully consider their water usage and what measures they can take to reduce it. There’s a danger that even as we move toward a net-zero world, the growing demands of data centres for resources have a detrimental impact on the environment.

But AI is already pointing to how data centre operations can be made more efficient with some notable success stories. In the future, the aim will be to create data centres that support the growth of technologies such as AI – while minimising the impact of data centres on the planet.

Learn about the future of manufacturing with RS. Want to learn more about workplace automation? We’ve got you covered.