Data centres rapidly integrating liquid cooling as AI causes temperatures to rise.
The rapid adoption of AI technology is driving demand for the redesign of data centres. This is impacting cooling systems and fuelling the debate about immersion versus air cooling. Could a hybrid option be the answer?
Use of AI has increased dramatically throughout 2023 and 2024. A McKinsey & Company survey found that 65% of survey respondents now report that their organisation regularly uses generative AI. With that trend set to continue, data centre owners must move quickly to ensure their facilities can meet these changing demands.
AI is causing some significant challenges for data centres. Chief amongst these are problems with power availability, grid congestion, site selection and planning regulations. Innovations such as battery energy storage systems (BESS) and the potential of small modular reactors will have an impact in the coming years.
" One of the most pressing issues facing data centre owners is the question of cooling.
Finding cost effective, efficient and more sustainable ways to address the additional heat generated by AI chip servers is a top priority for owners of both new and existing centres.
Liquid cooling for AI chips
With high computational, data processing, and communication requirements, AI is leading to an increase in data centre rack density. In turn, this requires highly efficient cooling systems to mitigate the extra heat generated. Increasingly the solution has been for data centres to adopt liquid cooling methods. With improved power utilisation efficiency compared to air cooling systems, liquid cooling is seen as a more cost effective and sustainable choice.
Liquid cooling takes various forms, more common currently is immersion cooling but direct liquid cooling techniques are gaining traction. Bringing liquid cooling directly to heat-generating components, allows for more efficient heat transfer making it a good solution for AI chips which generate more heat than traditional ones.
Framing the debate as liquid versus air cooling is too simplistic.
" As data centres seek to retrofit existing facilities or adjust in-progress designs to allow for increased AI server storage the answer needs to be more nuanced.
Designing for a flexible future
Exploring future resilience ahead of the first plans being drawn is the ideal scenario. But the rate of technology development and evolution of end-user demand, means we are increasingly asked to redesign existing data centre plans.
" With data centres still needing to house traditional server racks alongside AI we have developed a hybrid cooling approach.
This approach harnesses the improved performance of direct liquid cooling but continues to use air for non-server related cooling tasks.
This is particularly helpful in a retrofit project. Moving from low density server racks to the high density needed for AI chips results in unused space in need of traditional air-cooling.
Helping clients to understand and predict end user requirements is key to achieving the correct blend of cooling mechanisms. Whether it’s a retrofit of a built data centre or an adjustment of plans, our feasibility studies harness the extensive experience available within Deerns’ global network.
An industry defining design standards
As with the adoption of any new technology it can feel like a bit of a race for different methods or products to establish themselves as the industry standard – just think video vs Betamax (or Blu-ray vs HD DVD for more recent generations). Until there are accepted design standards in place we expect to see a variety of technologies and methods adopted by the tech giants.
To support clients with technology selection, Deerns brings over 20 years’ experience and involvement in over 550+ global data centre projects ranging from small to hyperscale. Giving us insights that result in tailored solutions that meet our clients’ needs both now and into the future.
Want to make sure your data centre is fit for the future? Get in touch to discuss your project.
Related thoughts
