Summary
- The EU is preparing a data centre energy efficiency package covering reported data, ratings, and minimum performance standards.
- Common metrics such as PUE, WUE, ERF, and REF can be accurate while still giving an incomplete picture.
- Planning, procurement, and financing scrutiny will increasingly depend on operational data, local context, and evidence that claims can be audited.
Data centre sustainability is measured through a growing set of numbers: power usage effectiveness, water usage effectiveness, energy reuse factor, renewable energy factor, carbon intensity, design performance, operating performance, and heat reuse. Each can be useful. Each can also leave out enough context to give a cleaner impression than the facility, grid, or local environment deserves.
A site can report a strong PUE while adding a major absolute load to a constrained electricity network. A low WUE can say little about indirect water consumption embedded in power generation. A heat reuse proposal can look technically plausible while lacking a heat customer, a network route, or a temperature level that works without additional energy input. Renewable energy claims can vary widely depending on whether they rely on certificates, annual matching, local procurement, additional generation, or hourly matching.
The European Commission is trying to make that evidence more comparable. Under the Energy Efficiency Directive, data centres above 500kW of power demand are subject to mandatory public reporting requirements, and Delegated Regulation 2024/1364 established harmonised reporting elements and the first phase of a Union rating scheme. The Commission’s Data Centre Energy Efficiency Package is expected to assess reported data, introduce a European rating scheme, and launch work on minimum performance standards.
That work sits within a wider policy effort to connect digital infrastructure to energy-system planning. On 3 June, the Commission published its Strategic Roadmap for Digitalisation and AI in Energy as part of the European technological sovereignty package, linking data, AI, energy infrastructure, and sustainability.
PUE remains the most familiar data centre efficiency metric. By comparing total facility energy with IT energy, it shows how much overhead is consumed by cooling, electrical losses, lighting, controls, and other non-IT loads. Within one facility, tracked consistently over time, it can show whether plant is being operated well. Across similar sites in similar climates and utilisation ranges, it can support useful comparison.
Used on its own, PUE can also over-compress the problem. It does not show total electricity demand, local grid constraint, carbon intensity, or peak-period impact. It does not show whether IT load is being used efficiently, or whether a facility with excellent overhead performance is still adding hundreds of megawatts to a system already struggling to connect other demand. A very efficient large load remains a large load.
WUE brings a different lens, particularly where evaporative cooling, adiabatic systems, humidification, or local water stress are relevant. Direct water consumption cannot be treated as a secondary detail in regions facing drought risk or competing industrial, agricultural, and domestic demand. Yet WUE also needs context. A facility using less direct water may consume more electricity, and the electricity system may carry its own water footprint elsewhere. A facility using more water may lower electrical demand enough to reduce wider system pressure. Without water source, basin stress, seasonality, and cooling mode, the metric is only a partial view.
Heat reuse exposes the gap between technical potential and usable infrastructure. Data centres produce large amounts of low-grade heat. Capturing it is the simpler part of the problem. Moving it to a customer at the right temperature, through a viable network, under a contract that justifies the capital cost, is harder.
Germany has already pushed heat recovery into regulatory form. Under its Energy Efficiency Act, new data centres above the relevant connected-load threshold coming online from July 2026 must achieve an energy reuse factor of 10%, rising to 15% in 2027 and 20% in 2028, according to Cundall’s analysis of the law.
Those targets make the commercial and engineering questions more visible. Low-grade heat often needs heat pumps before it can serve a district heating network. The data centre may be best located for fibre, land, and power, but poorly located for heat offtake. A heat network may not exist, or it may not have enough demand at the right times. A credible energy reuse claim therefore needs an offtake customer, a network plan, a temperature strategy, and a commercial model, not only an engineering diagram.
Renewable energy factor also needs careful treatment. A facility can buy renewable electricity through different mechanisms, and those mechanisms are not interchangeable. A long-term power purchase agreement with additional generation has a different system value from certificate-backed annual matching. Hourly matching has a different meaning from annual matching. Local generation connected to the same constrained system has a different effect from renewable procurement elsewhere in the market.
Carbon intensity can draw those questions together, but methodology decides how useful the figure becomes. Annual average carbon can smooth away the stressed periods that power systems most need to manage. Hourly and locational carbon data gives a more accurate picture of when and where electricity is being consumed, particularly for large users with some ability to shift workloads, charge storage, or participate in flexibility markets. The more granular the claim, the more demanding the evidence becomes.
Design figures need the same caution. Planning documents and investment cases often rely on expected PUE, projected heat reuse, future renewable procurement, and modelled operating conditions. Those assumptions may be reasonable, but they are not the same as measured operating data from a loaded facility across different seasons. The difference between design intent and operational performance is familiar across infrastructure, and data centres are not exempt from it.
Location changes the meaning of every metric. In west London and Slough, sustainability cannot be separated from grid constraint, land scarcity, heat-network opportunity, backup generation, and public scrutiny. In Ireland, national power-system pressure and hyperscale concentration put absolute load under greater political and regulatory attention. In Germany, heat reuse and efficiency requirements are more explicit. In the Nordics, cleaner power and cooler ambient conditions can improve some indicators, while in parts of Iberia, water stress can make cooling choices more difficult even where renewable generation is abundant.
A rating scheme can improve consistency, and minimum performance standards can set a floor. Neither will erase the need for local interpretation. The same PUE can mean different things in different power systems. The same WUE can carry different consequences in different river basins. The same heat reuse percentage can be commercially robust in one city and fragile in another.
A stronger sustainability evidence file would start with operational PUE, direct water use, annual and peak electricity demand, carbon intensity methodology, backup generation assumptions, refrigerant choices, heat reuse agreements, and measured performance once the facility is loaded. It would separate design claims from operating data and show how expansion phases affect the numbers. It would explain whether renewable energy is additional, local, hourly matched, certificate-based, or some combination of those mechanisms. It would also describe what happens during heatwaves, drought periods, grid constraints, generator testing, and maintenance.
Efficiency metrics have helped the sector measure improvement, and they remain necessary. The next phase will place them under heavier planning, procurement, financing, and regulatory scrutiny. The numbers that carry weight will be the ones connected to operating conditions, local infrastructure, and evidence beyond the data hall wall.
