PCE and ROIP: The Metrics That Will Decide The World Cup of AI

As teams prepare for the 2026 FIFA World Cup with unprecedented tactical data, the data center industry is undergoing its own transformation — shifting past legacy energy consumption toward high‑density cooling metrics that validate true compute capability.

As the 2026 FIFA World Cup kicks off in the United States, Canada, and Mexico, beneath all of the hype and excitement, data and metrics will play a critical role in deciding the winner.

National teams will arrive with entire analytics departments, GPS‑tracked training data, and real‑time physiological monitoring. FIFA also has its own Technical Studies Group — headed by FIFA chief of global football development Arsène Wenger — which will include an ‘Enhanced Football Intelligence’ service to enrich the coverage and analysis of every game.

“We are not only collecting more data than before but also trying to strike the right balance between technical expertise and data,” says Wenger.

In many ways, the early data center industry resembled football before this analytics revolution — a world where the dominant question was not “How well are we performing?” but “Are we fit enough to play?” Servers were mostly modest, workloads predictable, and the primary concern was whether the facility could keep the lights on and the equipment cool.

The arrival of Power Usage Effectiveness (PUE) in 2006 marked the first major attempt to quantify that early data center game. Much like the first generation of fitness trackers that counted calories and steps, PUE offered a simple, universal way to measure how cleanly a data center converts energy into usable IT power. It exposed inefficiencies, drove investment in better cooling, and helped the industry shed the equivalent of excess weight. For a time, it was exactly the metric the industry needed.

When Counting Calories — or Passes — No Longer Tells the Whole Story

But as with human physiology and elite sport, counting calories or passes only gets you so far. Two athletes can burn the same number of calories and achieve entirely different levels of performance. And two data centers with identical PUEs can deliver radically different amounts of compute, depending on how they are designed, cooled, and operated.

As the industry matured, it began to adopt a series of additional metrics — Carbon Usage Effectiveness (CUE), Water Usage Effectiveness (WUE), Thermal Usage Effectiveness (TUE), and Server Power Usage Effectiveness (SPUE) — each one an attempt to understand a different aspect of data center efficiency and sustainability.

• CUE resembled the early attempts in sports science to measure the environmental cost of training, the equivalent of tracking the carbon footprint of a World Cup campaign.
• WUE reflected the hydration demands of the system, much like the growing focus on electrolyte balance and water intake in endurance sports.
• TUE captured how effectively the body shed heat under strain, mirroring the rise of thermal‑stress monitoring in elite athletics.
• SPUE focused on the efficiency of the muscles themselves — the server‑level equivalent of measuring muscle activation or power output during a sprint.

These metrics were important, but didn’t tell the whole story. They told us how a facility consumed resources, not how well it performed. They are the footballing equivalent of counting kilometers run without understanding sprint intensity, recovery time, or pressing efficiency (regaining possession or forcing errors).

The Physical Reality of AI Infrastructure

AI has transformed data centers from steady‑state environments into high‑intensity performance engines. Racks that once drew 5 or 10 kilowatts now routinely exceed 80, 150, or even 300 kilowatts — entire clusters behaving like elite athletes that demand rapid bursts of power and generate extraordinary heat. While specialized hyper-scale clusters routinely push toward these extreme densities, the broader enterprise sector requires accessible, modular solutions to scale up safely. Managing these modern thermal strains across corporate infrastructure requires a fundamental shift toward advanced liquid architecture — a transition that Airsys is standardizing for the enterprise market through server-level spray liquid cooling platforms like LiquidRack™.

AI training resembles the most demanding form of sporting preparation — long intervals of sustained exertion punctuated by explosive peaks. Training runs for days or weeks, pushing the digital “muscles” to their limits. Inference, by contrast, is the equivalent of match day: short, sharp, latency‑sensitive bursts where performance must be immediate and consistent.

In this new environment, the shortcomings of PUE, CUE, WUE, TUE, and SPUE are more obvious. They tell us how efficiently energy, water, and cooling are consumed but not how well the facility performs. They describe the diet, the hydration, the temperature regulation, and the muscle efficiency but not the athletic output.

PCE: Shifting from Metabolism to Real-World Capability

This is the context in which Power Compute Effectiveness (PCE), developed by Airsys, has emerged. If PUE is the metabolic metric — the calories‑in, calories‑out view of the world — then PCE is the data center equivalent of VO₂ max, the measure of how much oxygen a footballer’s body can deliver. PCE captures how provisioned capacity is allocated at the design level. It reveals stranded capacity, cooling constraints, and design inefficiencies that PUE simply cannot see.

• PCE reveals how much of your provisioned power is allocatable to compute, surfacing idle or underutilized capacity.
• Formula:

P_{ITCompute, prov}

Total Power Capacity (kW provisioned)

Provisioned facility power minus cooling, electrical losses, and aux loads

• Direction: Higher is better
• Range: 0.50 (less effective) to 0.90+ (highly effective)

The industry including the Open Compute Project has recognized the value of PCE. Data center certification and advisory organization Uptime Institute states: “PCE does not measure real-time utilization, evaluate workload efficiency or compete with PUE. Instead, it answers a structural question: Given a permitted power envelope and designed redundancy, how much of that envelope is sustainably allocatable to IT?”

ROIP: The Financial ROI of Compute Power

Alongside PCE, Airsys has developed a second metric that has begun to reshape how operators think about power. Return on Invested Power (ROIP) reframes power not as an operating expense but as a capital asset — something that must generate a measurable return. In the AI era, where a single rack can cost more than a house and a cluster can rival the budget of a small town, ROIP provides a way to tie power, performance, and economic value together in a way that shows up clearly on balance sheets.

• High ROIP means power is producing revenue-generating compute, while low ROIP indicates megawatts are being consumed with little economic output.
• Formula:

Revenue or Compute Value ($)

Power Invested (kW)

(total active power used)

• Direction: Higher is better

If PCE is VO₂ max, ROIP is the equivalent of asking how much performance an athlete delivers for every hour of training invested. It captures the relationship between the intensity of AI training, the responsiveness of inference, and the economic value of the compute produced.

The Next Era of Infrastructure Performance

Together, PCE and ROIP represent a shift in mindset that mirrors the development of sports performance science. We have moved from counting calories to measuring endurance, strength, and output.

Airsys sits at the centre of this transformation. The company’s long history in thermal innovation — from its origins in the telecommunications boom to its new global headquarters in South Carolina and its European expansion in Hungary — has always been defined by an ability to anticipate what the industry will need next.

When it comes to AI and data center innovation, as well as the World Cup, metrics and data are vital but ultimately they exist to enable the real drivers: passion, performance and the desire to win.