DATA CENTRES

“ Data centres now need to accommodate increasingly dense IT loads making optimised power and cooling management even more critical ,” explains Alex Brew , Regional Director , Northern Europe at Vertiv . “ With rack densities predicted to exceed 100kW per rack , power and cooling infrastructure design and deployment have become significantly more complex .”

Adam Asquith , Technical Director at Black & White Engineering adds : “ With the projected growth rates and increases in chip TDP or rack density , new methods of cooling and power distribution will need to be adopted . This may include submersed cooling methods and power distribution consisting of conductors with higher ampacity .

Questions of place ( and space ) With AI being extremely computational , it requires increased power and cooling demands . This means training or inference models don ’ t need the same levels of uptime that traditional cloud computing requires , which could result in space and efficiency and cost related opportunities .

This is why location is a strategic consideration when it comes to data centre construction . Already , data centres are having to move AI workloads closer to the network edge in order to handle large data volumes .

“ Edge computing is experiencing a significant rise in popularity ,” Alex says . “ As the volume of data generated by connected devices continues to grow at scale , organisations are recognising the immense potential of edge capabilities to meet network demand . But as the edge of the network becomes more sophisticated , so too will the infrastructure needed to support it .”

Whilst the industry is eager to build and operate data centres that can support higher IT loads , they are also focusing on regions and climates that can provide greater benefits such as free cooling or renewables to decarbonise . “ Traditionally , data centres were situated on-prem or close to their main business location ,” Anna explains . “ This is still sometimes necessary for sensitive data or regulatory compliance yet it is no longer necessary or even advantageous in many cases . Most AI workloads do not require very low latency networks making them location agnostic and as a result there will be a shift away from metro

RAJESH SENNIK HEAD OF DATA CENTRE ADVISORY , KPMG UK

78 December 2024