In a thought-provoking article, Paul Quigley, President of Airsys USA, highlights the importance of thermal effectiveness in data centers for unlocking meaningful AI capacity, even when they appear efficient on paper. The traditional focus on efficiency metrics like Power Usage Effectiveness (PUE) has been effective in improving infrastructure design and operations, but the rise of high-density AI workloads has exposed the limitations of these metrics.
AI workloads have challenged the industry’s reliance on efficiency metrics by emphasizing the need for effectiveness in delivering usable compute. While traditional metrics measure how cleanly energy is delivered, they do not fully capture what that energy ultimately produces, especially in transient AI workloads where only a small fraction of processed information becomes durable intelligence.
The shift from efficiency to effectiveness in data centers is becoming more apparent, with concepts like Power Compute Effectiveness (PCE) gaining traction. PCE goes beyond traditional metrics by focusing on how much sustained compute emerges once power reaches IT equipment, highlighting the importance of managing heat effectively at the source to maximize usable output. Return on Invested Power (ROIP) further emphasizes the economic impact of effectiveness, showing that facilities with higher PCE generate more value from each unit of power consumed.
As operators face constraints in expanding their data centers, particularly in managing thermal challenges within existing footprints, liquid cooling emerges as a solution that enhances effectiveness by bringing cooling closer to the heat source and optimizing the transport of heat. By improving PCE and ROIP, liquid cooling enables facilities to unlock previously unreachable capacity within powered and permitted sites, emphasizing the importance of effectiveness in driving progress in an AI-driven world.
