A Defining Moment for Digital Infrastructure
Europe is entering a new era of digital infrastructure.
Artificial intelligence is accelerating demand for compute capacity at an unprecedented rate. Governments are seeking to attract investment while meeting ambitious environmental objectives. Operators are under increasing pressure to demonstrate sustainability performance, not only to regulators but also to investors, customers and local communities.
Against this backdrop, Spain's proposed Royal Decree on sustainable Data Centres has generated significant discussion across the industry.
Much of the debate has centred on renewable energy requirements. Equally important, however, are the proposed operational performance thresholds:
On the surface, these appear to be straightforward efficiency targets. In reality, they raise a much broader question:
As AI transforms the way Data Centres are designed and operated, are traditional sustainability metrics still capturing the full picture?
The Industry Is Changing Faster Than Many Realise
For years, the industry has relied on Power Usage Effectiveness (PUE) as its primary measure of efficiency.
The metric remains valuable.
It helped establish transparency across the sector. It encouraged operators to reduce energy overheads. It provided a common language for benchmarking performance.
However, the Data Centre industry of 2026 looks very different from the industry that first adopted PUE.
AI workloads are driving higher rack densities, greater adoption of liquid cooling and fundamentally new approaches to heat rejection. Facilities are operating at higher coolant temperatures, reducing reliance on mechanical refrigeration and changing the relationship between climate and efficiency.
That makes this an ideal moment to ask not only how we measure efficiency, but what we should be measuring.
AI Is Changing the Efficiency Equation
The common perception is that AI will inevitably increase the environmental impact of digital infrastructure.
The reality is more nuanced.
While AI undoubtedly increases demand for computing power and infrastructure, it is also accelerating the adoption of technologies that can reduce facility overheads and improve operational efficiency. This distinction is important. Lower PUE does not necessarily mean lower overall energy consumption, but it does provide insight into how effectively supporting infrastructure is operating.
To better understand this shift, RED modelled a highly efficient next-generation Data Centre design across five Spanish locations:
- Madrid
- Zaragoza
- Barcelona
- Málaga
- Santander
To explore these questions, RED used its in-house modelling platform, BRIZA, to simulate operational performance across representative Spanish climates and Data Centre typologies.
The study assessed both AI-focused and traditional Data Centre architectures under full-load and partial-load conditions, providing insight into the interactions between climate, cooling strategy and utilisation.
Modelling Scenarios
AI Data Centres
- Scenario 1: High-efficiency AI Factory, 100% IT Load, FWS 40-50°C D2C, 28-38°C Air (93% Liquid / 7% Air)
- Scenario 1a: High-efficiency AI Factory, 50% IT Load, FWS 40-50°C D2C, 28-38°C Air (93% Liquid / 7% Air)
Traditional Data Centres
- Scenario 2: High-efficiency traditional Data Centre, 100% IT Load, 28-38°C CHW temperatures with fanwall cooling and air-cooled chillers
- Scenario 2a: High-efficiency traditional Data Centre, 50% IT Load, 28-38°C CHW temperatures with fanwall cooling and air-cooled chillers
The study examined both AI-focused facilities and conventional Data Centre designs while assuming compliance with the proposed WUE requirements by eliminating evaporative cooling.
The modelling focused on cooling strategies without evaporative cooling; however, compliance with a WUE limit of 0.1 does not necessarily preclude limited adiabatic operation. In practice, carefully controlled adiabatic trimming may still provide opportunities to reduce peak plant requirements while maintaining very low annual water consumption.
The results revealed something significant.
The modelling indicated that highly optimised AI facilities could achieve annual PUE values between 1.107 and 1.122 across the locations assessed.
In contrast, conventional facilities were modelled to achieve annual PUE values ranging from 1.195 to 1.231, despite remaining highly efficient by today's standards.
The modelling indicates that a PUE of 1.15 can be achieved by highly optimised next-generation facilities.
However, the results also highlight that the pathway to achieving that target can vary significantly depending on cooling strategy, utilisation profile and overall facility architecture.
The analysis also provides visibility beyond annualised PUE values, allowing operational performance to be assessed across seasonal conditions and different load profiles.
Insight 1: Technology Is Overtaking Geography
Historically, climate has been considered one of the most important factors influencing Data Centre efficiency.
That assumption is becoming less true.
As facilities adopt liquid cooling and operate with reduced compressor dependency, cooling energy represents a smaller proportion of total facility consumption. The result is a reduced sensitivity to outdoor conditions.
RED's modelling demonstrated only modest variation between the coolest and warmest locations studied.
For next-generation AI facilities, technology choices increasingly have a greater impact on efficiency than geography alone.
However, air cooled racks are still being deployed for cloud and policymakers need to consider how future sustainability standards are impacted by technology and climate.
Insight 2: AI Is Redefining Efficient Design
Perhaps the most important finding is that AI facilities are not simply larger versions of conventional Data Centres.
They are fundamentally different.
- Modern AI infrastructure increasingly benefits from:
- Higher facility water temperatures
- Greater use of liquid cooling
- Reduced dependence on mechanical refrigeration
- More efficient heat rejection strategies
These characteristics enable facility infrastructure to operate more efficiently and help explain why AI-focused facilities can achieve significantly lower PUE values than traditional Data Centre architectures.
In some AI deployments, higher facility efficiencies can also be realised through greater operational flexibility, allowing power that would otherwise be consumed by cooling systems to be utilised by IT equipment. This benefit is not necessarily available across all facility types and highlights how performance outcomes can be strongly influenced by workload characteristics and overall architecture.
This challenges a common assumption within the sustainability debate.
AI is increasing demand for power, but it is also accelerating the adoption of technologies that can improve the efficiency of supporting infrastructure.
Insight 3: Utilisation Matters More Than Many Think
The modelling also highlighted the importance of utilisation when interpreting PUE performance.
PUE is ultimately a ratio. When facilities operate below their design load, much of the supporting infrastructure remains active:
- UPS systems
- Transformers
- Pumps
- Fans
- Controls
- Lighting
- Cooling equipment
As IT load decreases, these fixed overheads represent a larger proportion of total consumption, causing measured PUE to increase.
The result is straightforward:
The same building can achieve materially different PUE values depending on how much of the facility is occupied.
This becomes particularly relevant when considering that the proposed Spanish thresholds are defined as operational performance requirements rather than purely design-stage targets.
The modelling highlighted that utilisation can have a material impact on measured PUE, in some cases exceeding the variation observed between the locations assessed. AI facilities and highly variable load profiles can further influence measured PUE performance.
This matters because occupancy is often influenced by market conditions, leasing cycles and business growth rather than engineering quality.
If a facility can sit on either side of a compliance threshold simply because it is still filling capacity, then the metric may be reflecting operational circumstance as much as engineering performance.
Beyond PUE: What Should Future Regulation Measure?
None of this suggests that PUE has lost its value.
Far from it.
PUE remains one of the most effective benchmarking tools the industry has ever adopted.
The challenge is that Data Centres are becoming increasingly sophisticated, and sustainability is no longer defined by a single outcome.
Water use also cannot be considered in isolation. Metrics such as WUE do not distinguish between potable, grey or alternative water sources, nor do they directly reflect regional water stress or upstream resource consumption associated with electricity generation. In some circumstances, limited water use may reduce overall energy demand, carbon emissions and grid impact compared with entirely dry cooling approaches.
The modelling also highlights that fixed performance thresholds may naturally favour certain technical solutions over others. While this can help accelerate adoption of efficient technologies, it also raises important questions about maintaining design flexibility and encouraging multiple pathways towards sustainable outcomes.
Future frameworks may benefit from considering a broader range of factors, including:
- Facility occupancy and ramp-up profiles
- Cooling technology selection
- Water sourcing and regional water stress
- Workload characteristics
- Carbon intensity of electricity supply
- Energy reuse opportunities
- Operational resilience requirements
The objective remains unchanged: reducing environmental impact.
The question is whether a single operational metric can fully represent the increasingly diverse ecosystem of modern digital infrastructure.
Spain Has an Opportunity to Lead
Spain is rapidly becoming one of Europe's most important digital infrastructure markets.
Its growing renewable energy capacity, strategic location and expanding ecosystem of technology investment position the country to play a leading role in the development of sustainable AI infrastructure.
The current debate therefore extends beyond a single regulatory proposal.
It offers an opportunity to shape how the next generation of sustainability standards are defined.
Standards should be ambitious.
They should encourage innovation.
They should drive transparency and accountability.
Most importantly, they should ensure that performance metrics continue to reflect the realities of modern infrastructure design.
The Bigger Question
RED's modelling indicates that a PUE of 1.15 can be achieved by highly optimised next-generation facilities.
The more important question is what that achievement represents.
As AI reshapes Data Centre design, sustainability discussions must evolve from a focus on individual metrics toward a broader understanding of environmental performance.
Success will not be determined solely by the targets we set.
It will be determined by whether those targets accurately reward innovation, measure genuine sustainability outcomes and support the development of the infrastructure required for the digital economy.
That conversation is only just beginning.
RED Engineering Design submitted a consultation response on the proposed Spanish Royal Decree focused on the PUE and WUE requirements. The modelling referred to in this article was undertaken to support that submission and is offered as independent engineering analysis based on representative next-generation Data Centre designs.
