Cooling strategies that deliver PUE gains through evaporative or adiabatic systems can consume 26 million litres of water annually per megawatt of IT load

Decision Lens

The core tension is structural: PUE has been the industry’s default sustainability shorthand for over a decade, yet global average PUE moved only from 1.58 to 1.56 between 2018 and 2024 — a near-stagnant result despite significant investment. More critically, optimizing PUE in isolation can actively worsen carbon intensity and water consumption. As AI build-out drives electricity demand higher and disclosure requirements tighten, operators who report on PUE alone face a credibility gap with the exact audiences — regulators, institutional investors, and large enterprise customers — whose confidence underpins both license to operate and capital access.

90-Second Brief

Today, global data center electricity demand is set to double between 2022 and 2030, with AI facilities consuming four to five times more electricity than traditional infrastructure. Europe’s data center consumption alone is projected to grow from 70TWh in 2024 to 115TWh by 2030. Against this backdrop, reporting frameworks built around a single efficiency ratio are no longer defensible. Three metrics, PUE, carbon usage effectiveness (CUE), and water usage effectiveness (WUE), need to be tracked together to give stakeholders an accurate picture.

What’s Actually Happening

The industry built its reporting culture around PUE because it is calculable, comparable, and relatively easy to improve at the margin. That logic held when data centers were homogenous, density was moderate, and regulators were largely disengaged. None of those conditions apply today.

AI infrastructure changes the physics. High-density GPU clusters carry substantial embodied carbon from energy-intensive manufacturing and shorter hardware refresh cycles. A facility running at a respectable PUE can simultaneously carry high CUE if it sits on a carbon-intensive grid — one megawatt-hour consumed in Poland carries roughly 600g CO₂ versus approximately 10g in Sweden. PUE captures none of that variance.

Water compounds the problem in the opposite direction. Cooling strategies that deliver PUE gains through evaporative or adiabatic systems can consume 26 million litres of water annually per megawatt of IT load. In water-stressed regions — parts of Spain and Italy already face material drought risk — that trade-off is landing on regulatory radar and community opposition agendas simultaneously. The European Commission’s 2022 Climate Neutral Data Center Pact explicitly includes water metrics, signalling that WUE is transitioning from voluntary disclosure to compliance territory.

The result is a measurement architecture where improving one number can worsen two others. Operators who have not connected all three metrics into a unified reporting framework are making infrastructure decisions with incomplete information.

Why It Matters for Global Heads of Data Center Energy?

For energy leaders managing multi-gigawatt portfolios, the PUE gap creates three distinct operational risks.

First, PPA and site selection decisions made on energy efficiency grounds alone may not survive sustainability due diligence from investors or enterprise customers running their own Scope 2 and 24/7 CFE assessments. A facility with strong PUE on a high-carbon grid can undermine your organisation’s reported carbon intensity regardless of procurement strategy.

Second, cooling infrastructure choices made today carry long-duration consequences. Committing to evaporative cooling to hit PUE targets in a region approaching water stress is a stranded-asset scenario in slow motion. Water availability is becoming a site-selection variable with the same strategic weight as grid capacity, and decisions made now will be difficult to reverse at scale.

Third, the regulatory trajectory is compressing the timeline. Disclosure requirements are tightening across multiple jurisdictions simultaneously. Operators who build CUE and WUE tracking into their monitoring and reporting infrastructure now will be positioned to respond to mandatory disclosure frameworks rather than scrambling to retrofit measurement capability under deadline pressure. Those who do not will face the same credibility problem with regulators that they already face with sophisticated investors.

The Forward View

The operational implication is a shift in how sustainability performance is structured internally. Energy procurement strategy will need closer integration with carbon intensity data by grid region, making real-time or near-real-time CUE tracking a functional requirement rather than a reporting nice-to-have. Workload flexibility — shifting compute to periods of higher renewable availability — becomes a tool for managing CUE, not just cost.

On the water side, the pipeline of zero-water and low-water cooling technologies is advancing, with warm-water liquid cooling architectures gaining traction at high rack densities. In European markets facing water regulation, the capital allocation question is not whether to transition but when and at what pace.

Longer term, sustainability metric frameworks are likely to factor into utility and regulatory negotiations. ISOs and state PUCs are beginning to treat large load additions with greater scrutiny. Operators who can demonstrate credible, multi-dimensional sustainability performance will have a stronger position in interconnection and permitting discussions than those presenting PUE figures alone.

What We’re Uncertain About?

  • How quickly CUE and WUE will become mandatory disclosures. The European Commission’s Climate Neutral Data Center Pact includes water metrics, but enforcement timelines and penalty structures across member states are not yet uniform. What would resolve this: clarity from the European Commission on mandatory reporting thresholds and timelines, and equivalent signals from FERC or EPA in the US context.

  • Whether grid carbon intensity data will become granular and real-time enough to operationalize CUE at scale. Locational marginal emissions vary significantly by hour and node. Operators managing multi-region portfolios need market-grade carbon intensity signals to act on CUE, not just report on it. What would resolve this: ISO/RTO provision of nodal carbon intensity data comparable in quality to LMP data.

  • The long-run cost trajectory of low-water cooling at AI-relevant rack densities. Liquid cooling reduces water consumption and improves thermal performance at high density, but capital cost premiums and maintenance complexity at scale remain variables. What would resolve this: published operational data from hyperscalers running large liquid-cooled AI deployments beyond initial pilots.

One Question to Bring to Your Team

When you evaluated your last major site selection or cooling infrastructure decision, did CUE and WUE carry explicit weighting alongside PUE — and if not, which of those gaps is most likely to generate a regulatory or investor exposure in the next 24 months?

Sources

  • Datacenterdynamics — Beyond PUE: Rethinking how data center sustainability is measured (Link)

Tags