Every data centre story begins with fire.

 

From the first power plant to the latest AI accelerator, all computation is ultimately the controlled movement of energy—and the controlled removal of heat. Yet as AI workloads scale and liquid-cooled architectures become mainstream, many operators are discovering that traditional efficiency metrics no longer explain why compute stalls, even in facilities that are fully powered and permitted.

 

This session begins with The Journey of Fire, a narrative framework that traces how raw energy is transformed, constrained, and often lost on its path to usable compute. From grid delivery to silicon behaviour, Paul Quigley explores where today’s data centres quietly hit thermal and operational limits long before they hit electrical ones.

 

Building on that foundation, the talk introduces Power Compute Effectiveness (PCE), a framework originated by Yunshui Chen and developed in collaboration with operating teams to better reflect real-world AI performance. Rather than measuring how efficiently facilities operate, PCE focuses on how effectively delivered power becomes sustained compute output under real conditions.