The conversation around data center performance is changing. Investors, analysts, and several global operators have begun asking a question that PUE cannot answer, how much compute do we produce for every unit of power we consume. This shift is not theoretical, it is already influencing how facilities are evaluated and compared.
Investors are beginning to favor data center operators who can demonstrate not only energy efficiency but also compute productivity per megawatt. Capital is moving toward facilities that understand and quantify this relationship. Several Asian data center groups have already started benchmarking facilities in this way, particularly in high density and liquid cooled environments.
Industry organizations are paying attention to these developments. The Open Compute Project has expressed interest in reviewing a white paper on Power Compute Effectiveness, PCE, and Return on Invested Power, ROIP, to understand how these measures could inform future guidance and standards. These signals point in a consistent direction. PUE remains valuable, but it can no longer serve as the primary lens for evaluating performance in modern facilities.
PUE is simple and recognizable.
PUE = Total Facility Power ÷ IT Power
It shows how much supporting infrastructure is required to deliver power to the IT load. What it does not show is how effectively that power becomes meaningful compute.
As AI workloads accelerate, data centers need visibility into output as well as efficiency. This is the role of PCE.
PCE = Compute Output ÷ Total Power
PCE reframes performance around the work produced. It answers a question that is increasingly relevant, how much intelligence or computational value do we create for every unit of power consumed.
Alongside PCE is ROIP, the operational companion metric that reflects real time performance. ROIP provides a view of how effectively power is being converted into useful compute at any moment. While PCE shows long term capability, ROIP reflects the health of the system under live conditions and exposes the impact of cooling performance, density changes, and power constraints.
This shift in measurement mirrors what has taken place in other sectors. Manufacturing moved from uptime to throughput. Transportation moved from mileage to performance and reliability. Data centers, especially those supporting AI and accelerated computing, are now moving from efficiency to productivity.
Cooling has become a direct enabler of compute and not just a supporting subsystem. When cooling performance changes, compute output changes with it. This interdependence means that understanding the relationship between power, cooling capability, and computational output is essential for real world performance, not just engineering design.
PUE still matters. It reflects operational discipline, mechanical efficiency, and the overall metabolism of the facility. What it cannot reveal is how much useful work the data center is actually producing or how effectively it can scale under load. PCE and ROIP fill that gap. They provide a more accurate view of capability, consistency, and return on power, especially as the industry moves from traditional air cooled environments to liquid ready, high density architectures.
The next phase of data center optimization will not be defined by how little power we waste, but by how much value we create with the power we have. As demand increases and the grid becomes more constrained, organizations that understand their true compute per megawatt performance will have a strategic and economic advantage. The move from energy scarcity to energy stewardship begins with measuring what matters.
The industry has spent years improving efficiency. The AI era requires us to improve output.
# # #
About the Author
Paul Quigley is the former President and current Chief Strategic Partnership Officer of Airsys Cooling Technologies, and a global advocate for high density, energy efficient data center design. With more than three decades in HVAC and mission critical cooling, he focuses on practical solutions that connect energy stewardship with real world compute performance. Paul writes and speaks internationally about PCE, ROIP, and the future of data center health in the age of AI.
