Ai Datacentre Physics Set to Make On-Premise Datacentres Obsolete
The era of on-premise datacentres is rapidly coming to an end, thanks to the insatiable appetite of GPUs in AI processing. With power and cooling requirements spiralling out of control, enterprises are facing a daunting reality.
Direct-to-Chip Liquid Cooling Becomes a Necessity
GPUs are the unsung heroes of AI processing, capable of delivering unprecedented computing muscle. However, this comes at a significant cost. A single GPU can consume up to 350W of power, while the entire datacentre can guzzle 100MW or more. To put this in perspective, powering a single datacentre would require as many as 100,000 GPUs. This is where direct-to-chip liquid cooling comes in – a technology that’s about to become mandatory for any serious AI processing operation.
The Rise of AI Factories and Hybrid Datacentres
The power draw and cooling requirements of GPUs are so extreme that traditional air-cooled datacentres can no longer keep up. This has given rise to a new era of AI factories, where companies are building custom-built datacentres that can handle the intense heat generated by GPUs. While this may sound like a futuristic concept, it’s already a reality. Companies like Google and Amazon are investing heavily in these AI factories, which will likely become the norm for serious AI processing.
Enterprises Will Need a Hybrid Approach
While AI factories may be the future of AI processing, on-premise datacentres won’t disappear overnight. Companies will still need to rely on them for certain workloads, such as legacy applications and non-AI-related tasks. This is where the hybrid datacentre comes in – a model that combines the benefits of on-premise and cloud-based infrastructure. Enterprises will need to adopt this hybrid approach to stay competitive in the AI era. What this means is that companies will need to invest in both AI factories and traditional datacentres, creating a complex and costly infrastructure.
In summary, the rise of AI datacentre physics is set to revolutionise the way companies approach datacentre management. While on-premise datacentres may be on their way out, a hybrid approach will likely become the norm for enterprises. As the AI factory revolution gains momentum, companies will need to adapt quickly to stay ahead of the curve.
