Executive Summary

• Data centre capacity planning is a fickle art; it goes beyond planning for the current power demand and managing infrastructure so it can remain resilient and flexible for the changing power demands.
• The AI revolution has caused a phase of overprovisioning, which is proving costly, so now it’s time to re-evaluate, optimise and reduce OpEx.
• AI applications are an excellent tool for analysing data and creating reports based on infrastructure performance, therefore they should be implemented more into capacity planning.

Data centre capacity planning is going beyond planning for today’s power demand; it’s about predicting the unpredictable, high-stakes performance and planning for future demand that shifts the goalposts – and continuously doing so.

It’s all about forecasting and aligning IT infrastructure – power demand, cooling, space and network – to meet changing demands and avoid downtime; it’s a guessing game of precision and chaos, where operators walk a fine line between over- and underprovisioning, navigating a fibre optic tightrope over a canyon of uncertainty.

Capacity management involves monitoring,maanging and forecasting elements including power, cooling racks and physical floor space, network connectivity, data storage, tenancy availability and computing power consumption.

Getting it ‘right’ is a fine balance but is rather essential, for several reasons. Optimising the use of the facility and resources will conserve energy consumption and reduce operational costs while remaining on track to reach sustainability targets; proper performance optimisation will maintain high performance and prevent bottlenecks, so the risk of downtime reduces. This all leads to greater resilience and availability, resulting in environmental sustainability and scalability, enabling a smoother response to changing demands and ensuring flexibility.

A fickle art

Just like a conductor, weaving his baton to bring together an entire orchestra, one wrong move could result in off-key, mismatched timing – and the same could be said of capacity planning. Infrastructure and its systems are becoming more complex; there’s more risk of downtime if there are multiple points of failure, not to mention the locked grid capacity and intense high-density AI workload demand; capacity planning is such a fickle art.

Right now, thanks to the AI revolution, we’re in a state of overprovisioning. According to JLL, average global construction costs are hitting $11.3M per MW this year, so over building be even 5MW is a $56M mistake.

So we must tread carefully and capacity plan with reducing costs in mind, while still ensuring that we’re meeting fluctuating power demands.

AI and capacity management

AI adoption is cropping up across all data centres now in their capacity management, and it would be a missed opportunity to help you successfully forecast that could end up costing you thousands, if not millions.

Capacity management greatly benefits from the adoption of AI applications since it’s an excellent tool for analysing data and creating reports based on infrastructure performance data. As applications improve, so will the accuracy of the AI output and thus, the more successful capacity planning will be.