Data Centers Need to Move Beyond Power Usage Effectiveness

The traditional metric of Power Usage Effectiveness (PUE) has limitations in assessing data center efficiency, particularly in the AI era. PUE is well-suited for tracking changes in a facility’s efficiency over time but fails to account for factors like climate, resilience, and water use. To properly assess data center efficiency, a more comprehensive approach is needed, considering factors like flexibility, heat management, and reliability.

The share of operators running high-density racks has increased, driven by growing AI and GPU workloads. Cooling systems must be designed to adapt to fast-changing AI hardware, and ongoing engineering support is becoming increasingly important. Designing IT, power, and cooling as a coordinated system is crucial for overall facility performance.

Traditional data centers mainly use air cooling, but AI data centers often employ direct-to-chip liquid cooling, which requires a more complex thermal management system. Reducing cooling energy use can free up electricity for GPUs and AI computation, and industrywide advances in energy efficiency are necessary to mitigate the challenges of high temperatures, humidity, and water constraints.