The data center industry has spent too long chasing “good on paper” sustainability metrics, all the while absolute emissions continue to rise.

In many ways, traditional metrics such as power usage effectiveness (PUE) represent only the visible tip of the iceberg – a narrow surface indicator of efficiency, while the far larger mass of environmental impact sits below the waterline, unseen.

Predictable, relatively uniform classical workloads that could be steadily optimized over the past two decades have led to operator complacency. But today, the AI era has fundamentally changed that landscape, introducing volatile, highly dynamic demand patterns that place unprecedented pressure on infrastructure planning and energy systems alike.

In this new reality, instead of hiding behind optics or outdated efficiency benchmarks alone, the industry must shift its focus to real-world outcomes. Achieving net-zero will require a more transparent and accountable approach – one that ties every watt consumed, every token processed, and every deployment decision to tangible impact.

“The next thousand days are going to define the next 30 years of infrastructure – the pace of acceleration is reshaping not just how we operate but how and where demand shows up,” says Aparna Prabhakar, chief strategy and sustainability officer at Schneider Electric.

The intersection of strategy and sustainability

Where strategy and sustainability have historically been treated as parallel considerations in energy and power infrastructure, AI has brought the two together in a far more immediate and tangible way.

What is unfolding is a system-level transformation reshaping how infrastructure is designed, powered, and operated, with sustainability embedded from the outset.

“On the surface, strategy and sustainability can look like two separate roles – one focused on growth and the other on impact. But in reality, they are deeply connected. Today, growth is increasingly constrained by energy infrastructure, and sustainability is what unlocks it – whether that’s faster to deploy solutions like solar and storage, more efficient architectures that reduce demand, or approaches that accelerate regulatory approvals,” says Prabhakar.

The challenge is further complicated by the fact that AI is not a single, homogenous workload. Its demands are varied, distributed, and highly nuanced. Training workloads, for example, tend to be centralized, episodic, and defined by immense power consumption. Inference workloads, by contrast, are distributed, always-on, and highly latency-sensitive.

As a result, infrastructure operators are now balancing the needs of hyperscale campuses alongside a growing network of dispersed Edge and inference environments – all while navigating increasingly constrained grid capacity.

Working at the intersection of strategy and sustainability in the data center sector now means recognizing that sustainability is not a constraint on growth, but a key enabler of it. From unlocking faster access to power to reducing overall demand and accelerating deployment timelines, sustainable strategies are increasingly central to how infrastructure scales. As Prabhakar explains:

“I’m in the business of not just crafting strategy, but translating that strategy into execution across our hardware, software services, and entire stack, and making sure it holds up for the real conditions.”

Finding meaning in metrics

On the surface, users interact with AI-enabled applications like Amazon Alexa and OpenAI ChatGPT. Beneath that surface sits a far more complex web of impacts spanning carbon intensity, water consumption, grid strain, embodied emissions, and broader community effects – areas where metrics like PUE offer only limited visibility.

Many of the industry’s traditional facility-level metrics emerged in an era when operators were largely focused on managing heat and optimizing mechanical performance within relatively predictable computing environments. Metrics such as PUE became shorthand for operational efficiency, offering investors and operators a simplified way to benchmark facilities.

But those metrics reveal little about how effectively energy is being converted into useful computational work. Classical enterprise workloads, AI training clusters, and distributed AI inference environments all operate with strikingly different operating characteristics – meaning efficiency can no longer be captured through a single universal benchmark.

For example, in Schneider’s white paper, “Bending the energy curve: Decoupling digitalization trends from data center energy growth,” the company shares a methodology correlating annual information and compute technology (ICT) energy consumption with annual data production. This angle frames the conversation between infrastructure and ICT technology providers, demonstrating where infrastructure and energy converge. Although the methodology helps to evaluate operational efficiency, it says little about carbon footprint, local air pollution, water use, or other outcomes that matter to regulators, customers, and communities.

“You could have infrastructure and processes that are absolutely identical across facilities, yet deliver completely different outcomes in terms of carbon footprint, performance, and utilization,” says Prabhakar, adding:

“Looking beyond AI, the industry is already exploring quantum computing, adiabatic computing, and fundamentally new processing architectures. You can’t think in silos anymore – you have to think in terms of complex, system-wide interactions.”

The path to real-world value

From an environmental perspective, the atmosphere does not care how efficient a facility appears on paper – only about the total impact that facility has on the world around it.

That is why operators are being pushed to move beyond isolated, facility-level measurements and toward more workload-aware, system-level metrics. The real challenge is understanding how the full infrastructure stack performs – from chip to grid, cooling to compute, and ultimately how effectively energy is converted into meaningful computational output.

Today, the industry’s greatest constraint is not compute capacity, but energy infrastructure itself. The race is shifting from speed to deployment toward speed to power – the ability to secure, distribute, and optimize energy fast enough to meet accelerating AI demand.

That challenge brings carbon intensity, energy access, grid capacity, and deployment readiness into infrastructure planning far earlier than before. It also raises two critical questions: how to deploy the most efficient technologies possible, and how to unlock trapped or underutilized capacity already sitting within the grid.

According to Prabhakar, the first step is treating controls, power, cooling, and compute as a unified system rather than isolated domains. It also requires closer collaboration between suppliers, operators, technology partners, and infrastructure stakeholders to accelerate innovation and share operational insight across the ecosystem.

Drawing on broader industry discussions, Prabhakar notes that sustainability conversations are increasingly framed through the lens of operational and business risk, rather than climate messaging alone.

That shift has major implications for supply chains. Scope 2 and supplier Scope 3 emissions increasingly shape an organization’s overall sustainability profile, meaning infrastructure providers can no longer treat supplier emissions as “someone else’s problem.”

Responsibility moves across interconnected ecosystems – rippling from one organization to another. So, creating a more sustainable infrastructure future can’t depend solely on building entirely new data centers, but on helping established organizations modernize existing operations, unlock stranded efficiency, and reach their unrealized potential.

To address this, Schneider launched its Zero Carbon Project, aiming to help suppliers reduce their emissions intensity by 50 percent.

“About 70 percent of our suppliers had not even started carbon initiatives, so it was a hugely ambitious undertaking. We built capabilities, trained people, and spent countless hours helping suppliers reduce their carbon footprint – which ultimately reduces ours and strengthens the wider industry ecosystem.”

The company has also expanded its sustainability advisory efforts, partnering with major enterprise organizations to operationalize decarbonization strategies across infrastructure and energy systems.

“We are effectively saying to clients: we are drinking our own champagne – now let us help you drink yours.”

At its core, the transition is about helping organizations rethink sustainability as operational transformation – not just compliance. On the customer side, Prabhakar adds:

“We have been working with clients much earlier, with platforms like EcoStruxure, and working with AVEVA to create system-level visibility linking electrical behavior, cooling performance, and workload demand in real-time.”

Digital twin capabilities, including ETAP, are also helping operators model thermal and energy performance before facilities are even built – reducing design timelines and significantly accelerating speed to power.

The AI paradox

As infrastructure systems become increasingly integrated and intelligence moves deeper into operations, AI is emerging as both a driver of demand and a tool for managing it – creating a difficult paradox at the heart of the industry.

AI-enabled platforms are already being used to orchestrate workloads, optimize cooling systems, improve power distribution, and model infrastructure performance in real time. But as organizations deploy more AI to manage increasingly AI-driven environments, where does the balance ultimately sit?

“There is a lot of conversation around AI being bad for power but good for energy. In that sense, AI can be both the pain and the solution,” says Prabhakar.

“Yes, AI is driving a massive increase in energy demand – that is real. But the conversation should focus on how that demand materializes. The question is not whether AI itself is sustainable, but whether we can design infrastructure intelligently enough to make it sustainable. There is a delicate balance.”

For Prabhakar, the challenge is not “AI versus sustainability,” but whether infrastructure systems can absorb and manage AI-driven demand efficiently enough to keep pace with growth.

That means deploying AI as an operational layer as well as a workload – using it to design infrastructure more intelligently and improve energy distribution across increasingly complex environments.

Without that level of optimization, however, demand growth risks outpacing the infrastructure designed to support it.

The bottom line

The data centers – and increasingly the AI factories – of today are no longer uniform, predictable environments. They are shaped by volatile workloads, fluctuating power demands, and rapidly evolving infrastructure requirements. In that reality, traditional metrics such as PUE remain useful, but only as surface-level indicators.

The focus must therefore widen – beyond efficiency alone, toward long-term system impact and asking the key questions about how infrastructure interacts with energy grids, water resources, surrounding communities, and the accelerating demands of the future. As Prabhakar emphasizes:

“How we think about efficiency, sustainability, and risk management cannot be measured solely in dollars. The real question is whether our infrastructure choices strengthen or constrain our ability to scale – across energy, carbon, water, and community. These are fundamental to long-term performance.”

That shift requires a fundamental rethinking of infrastructure strategy. Designing for outcomes, not just for input. Coordinating the system, not optimizing in siloes. Accounting for life cycle impact, not day-one efficiency.

Much of sustainability’s true impact remains below the surface – hidden beneath the visible outputs of AI and digital services. The challenge for the industry is learning how to measure, manage, and respond to those deeper system effects before growth outpaces the infrastructure supporting it.

“For us as an industry, there is an urgency to act now rather than waiting for perfect codification. We need to execute today and build the frameworks as we learn. The important thing is that we move forward,” concludes Prabhakar.

What we see today – the applications, models, and outputs – is only the visible tip of the iceberg. The far larger impact of AI sits below the surface, embedded in the energy systems, resources, and communities that make it possible.

Ready to turn AI insights into performance at scale? Discover how Schneider Electric can help you design, build, and optimize AI-ready data center infrastructure that meets your performance, efficiency, and sustainability goals by visiting se.com.