Whether for the optimisation of energy consumption, autonomous management of workloads or predictive maintenance – artificial intelligence is finding its way into data centres. In this interview, Artur Faust, Energy & Sustainability Expert at EkkoSense and spokesperson at this year's Data Centre Expert Summit, talks about opportunities, current developments and the question of whether autonomous data centers will soon become a reality.
The integration of AI into data centre operations is bringing about significant changes for the industry – from load balancing to energy efficiency. What types of optimisations can AI models offer for data centre operations?
Current AI models for DC operations are all still based on machine learning and are pre-set to the topology and type of infrastructure. Instead of classic rule-based automation, they evaluate all data relating to system behaviour, statuses and consumption over the course of history using a ‘complex sensor data fusion’. The information generated from this ‘dynamic view’ opens up all kinds of optimisation approaches, such as the consistent improvement of energy performance or the immediate detection of anomalies before failures even occur. This provides site managers with a powerful set of tools that can offer everything from cost savings, process optimisation and decision-making to increasing availability.
Where do you currently see the greatest opportunities in the use of AI for data centres?
The in-depth analysis of infrastructure data and the ability of AI to simulate certain issues on the digital twin very closely to real-life behaviour creates new skills in the team and compensates for the shortage of specialists to a large extent. There is also no need to commission external players. As a result, the organisation gains decision-making speed, which is a major competitive advantage in today's world because it can react quickly and competently to the market.
How exactly does AI improve dynamic load management in hyperscaler and edge data centres?
This is an exciting topic that is hard to beat in terms of complexity. Because when AI workloads are used in data centres, you quickly have to deal with load dynamics that become a fundamental problem for conventional infrastructures. If you want to continue to ensure availability and efficiency, it is very likely that adjustments will have to be made to the design and control in order to meet the new requirements – but which ones? AI-supported simulations on the digital twin can provide significant insights into the existing system that hardly existed before. This makes it much easier to successfully adapt the design and control system, as you can run through the best possible scenario on the drawing board until all requirements and costs are viable.
How can data centres flexibly adapt and scale their capacities to demand with the help of AI?
Site managers can use the data generated by complex machine learning models to gain precise insights into current electrical and thermal capacities depending on current and historical load curves and distribution. This can even include analysing the thermal air flow in the server rooms and the volume flows for liquid cooling. AI tools, such as EkkoSense's ‘Cooling Advisor’ or the EkkoSim data center simulation, provide the necessary information framework to make informed decisions about necessary optimisations and adjustments to changing load conditions.
In which direction will AI-supported optimisation of data centres develop over the next few years? How far away are we from a fully autonomous data centre that optimises and repairs itself and perhaps even further develops its own architecture?
So far, there are very few manufacturers who are even addressing the issue of AI-supported DC operations management – the topic is complex and we are still a very long way from generative AI in this area, because we cannot afford and do not want so-called ‘hallucinations’ of AI in DC operations. While some providers focus on selective AI solutions – for example to control cooling systems – EkkoSense is one of the first companies to take a holistic approach to supporting data centres with AI. However, it will certainly be another decade or two before AI is ready to learn on its own and take over the operational management of a data centre. We are still pretty much at the beginning when it comes to AI.
