ENERGY & POWER
AI has fundamentally changed energy behaviour within data centres. It is now dynamic. Making sense of this volatility and designing systems capable of supporting it requires a level of precision, nuance and synchronisation in measurement that legacy approaches were never built to deliver.
Traditional benchmarking metrics, such as Power Usage Effectiveness( PUE) and Data Centre infrastructure Efficiency( DCiE), remain helpful for more general tasks. Providing a high-level view, they are used to understand net consumption figures and for sustainability benchmarking. However, what they cannot do is reveal how energy actually behaves within the system – offering no explanations for power quality, losses, nor identifying any transient events.
To support AI and modern high-performance computing( HPC) workloads, operators need granular visibility. These environments have introduced electrical characteristics such as sub-millisecond load changes, high crest factor currents and increased harmonics, while simultaneously demanding massive power densities and accelerating hardware refresh cycles. This places power architecture and supporting infrastructure under severe stress.
Under these conditions, measurement approaches built
on steady-state assumptions fall short. Averaging power over long intervals obscures the very events that now define system behaviour: transient behaviour, harmonic distortions, dynamic inefficiencies and rack-level underutilisation – all are completely masked despite posing potentially critical threats to system availability.
For systems designers, infrastructure manufacturers and validation engineers, the goal is clear: evolve your measurement tooling.
54 September 2026