How is AI Redefining Energy KPIs for Data Centers?
AI is reshaping the energy profile of digital infrastructure and exposing the limits of the metrics we have traditionally relied on.
Over the past few years, its adoption has driven an unprecedented increase in processing demand, pushing data centers toward higher-density environments, more variable workloads, and significantly more intensive computational cycles. As a result, energy consumption is at the center of how infrastructure is designed, operated, and evaluated.
This shift has placed the data center industry at the core of broader discussions around sustainability and environmental impact.
Historically, metrics such as Power Usage Effectiveness (PUE) and Water Usage Effectiveness (WUE) have been widely adopted as benchmarks for operational efficiency. And they remain relevant; but these metrics were designed to measure the efficiency of physical infrastructure — not the performance or impact of computation itself.
In the context of AI this raises the question: are we measuring what truly matters?
The Limits of Traditional KPIs
Metrics such as PUE have played an essential role in improving operational standards. They brought visibility to infrastructure efficiency and helped establish a common baseline. However, they were designed to measure the performance of the building — not the true impact of the operation.
In AI environments, this limitation becomes clear. Workloads are dynamic, GPU-intensive, and vary significantly in complexity, which means that measuring only how energy is distributed is no longer enough.
The industry is now moving toward a new logic — from building efficiency to computational efficiency – and this shift requires connecting energy consumption to what is being processed, through metrics such as:
– performance per watt
– carbon intensity per workload
– energy consumption across model lifecycles
At the same time, energy analysis must become more systemic. Now we are looking at absolute consumption, energy sources, and carbon intensity. In this context, higher consumption does not necessarily mean higher impact — what matters is how that energy is generated and used.
New KPIs for a New Infrastructure
As the data center ecosystem becomes more complex, the way we measure its impact must become more complex as well.
Scope 3 emissions are a good example. They encompass the entire value chain from equipment manufacturing to infrastructure construction and represent a significant portion of total emissions. It is not easy to measure, but our market is not simple.
This reinforces the need to evolve KPIs toward a lifecycle perspective, incorporating elements such as embodied carbon, supply chain emissions, and total asset impact. Data centers are like an iceberg: you can’t evaluate the operation based on what you see, because most of the operation is below the surface.
At the same time, regulatory pressure and technological innovation are accelerating this transition. New solutions like liquid cooling expand efficiency possibilities but also challenge existing metrics, which were not designed to capture this level of complexity.
Conclusion
AI is redefining the role of data centers — placing them at the intersection of digital infrastructure and energy systems.
This new context exposes the limits of traditional metrics and demands a broader approach, one that integrates computational performance, environmental impact, and lifecycle considerations.
More than new tools, this scenario requires maturity, transparency, and collaboration, which will allow industry to evolve in alignment with the challenges of the economy and society.
By Cora Nogueira, ESG Executive Manager of Elea Data Centers.
