Uptime Institute PUE (Power Usage Effectiveness)
Power Usage Effectiveness is the single number every data centre operator quotes and almost everyone misreads. It tells you how much electricity a facility burns to deliver one watt to a server — and nothing whatsoever about how clean that electricity is.
A data centre can run a world-class PUE and still pour carbon into the atmosphere; PUE measures the overhead, not the emissions.
PUE is the ratio of total data-centre energy to IT-equipment energy, standardised by ISO/IEC 30134-2:2026. A PUE of 1.0 is the theoretical ideal; modern hyperscale sites reach ~1.1, while the global average sits near 1.5. PUE measures efficiency, not carbon.
Executive Summary
Power Usage Effectiveness (PUE) is the most widely cited efficiency metric in the data-centre industry. Coined by The Green Grid in 2007 and standardised internationally as ISO/IEC 30134-2, it answers one question: for every unit of energy delivered to IT equipment, how much total energy does the facility consume? The arithmetic is deceptively simple — total facility energy divided by IT energy — but the metric carries a decade and a half of accumulated convention, edge cases, and misuse.
PUE does three things well. It gives operators a single, trackable number for infrastructure overhead. It exposes where energy is lost — cooling, power conversion, lighting, and distribution. And it has become the regulatory anchor for data-centre energy reporting, most consequentially under the EU Energy Efficiency Directive. What PUE does not do is measure carbon. A facility on a coal-heavy grid with a PUE of 1.1 emits far more per unit of compute than a facility on a hydro grid with a PUE of 1.6. The gap between efficiency and emissions is the single most important thing a sustainability team needs to understand about PUE, and it is where this reference spends its weight.
(1) PUE = total facility energy ÷ IT equipment energy; the theoretical floor is 1.0 and it has no upper bound. (2) It is standardised by ISO/IEC 30134-2:2026, which superseded the 2016 first edition in January 2026. (3) Measurement categories PUE1–PUE3 differ by temporal resolution, not by whether the number is “better.” (4) PUE measures energy efficiency, never carbon intensity — converting PUE to emissions requires a grid emission factor. (5) Under the EU EED, data centres with ≥500 kW IT load must report PUE annually to a public European database.
Provider directory
Now it has to hold up all year.
See who does this work. Every listing names the standards it works to, and paid placements are labelled. Including Pangolin Associates and Small World Consulting.
Browse 6 carbon accounting & inventory providers →Do this work? A listing is US$390 a year. Get listed →
What PUE Is
PUE is a dimensionless ratio expressing how efficiently a data centre uses energy. It compares the energy drawn by the entire facility — servers, storage, networking, cooling, power conversion, lighting, and every supporting system — against the energy drawn by the IT equipment alone. The closer the ratio is to 1.0, the smaller the non-IT overhead, and the more of every purchased kilowatt-hour reaches the workload that justifies the building’s existence.
A PUE of 1.0 is a theoretical ideal: every watt entering the facility reaches IT equipment, with zero loss to cooling or conversion. It is unreachable in practice because cooling, uninterruptible power supplies, transformers, and distribution all consume energy. A PUE of 2.0 means the facility draws two watts for every watt of compute — half the energy is overhead. The metric is unbounded upward; a poorly run legacy facility can exceed 2.5 or 3.0.
ISO/IEC 30134-2 is deliberate about terminology: it uses “effectiveness” rather than “efficiency” to preserve continuity with the metric’s original market name, while noting that “efficiency” is the more technically accurate description of what the ratio captures. PUE is one of a family of data-centre key performance indicators in the ISO/IEC 30134 series — Part 1 sets the general KPI framework, Part 2 defines PUE, and later parts define the renewable energy factor, cooling efficiency, and related measures covered under sister metrics below.
The PUE Formula and Its Components
PUE is defined as the total energy consumed by the data centre divided by the energy consumed by its IT equipment, measured over the same period. The standard expresses it two equivalent ways:
PUE = Total Facility Energy ÷ IT Equipment Energy = 1 + (Non-IT Energy ÷ IT Equipment Energy). Because non-IT energy is always positive, PUE is always greater than 1.0. The “1 +” form makes the overhead explicit: a PUE of 1.4 means non-IT systems consume 40% of the IT load on top of the IT load itself.
The two terms in the ratio are precisely scoped by the standard:
| Term | What it includes | What it excludes |
|---|---|---|
| Total Facility Energy | All energy entering the data-centre boundary: IT equipment, cooling and chillers, UPS and power-conversion losses, lighting, building management, security, and any other supporting system within the boundary. | Energy serving co-located non-data-centre functions (offices, retail) where the boundary excludes them — a focus area of the 2026 mixed-use-building guidance. |
| IT Equipment Energy | Energy drawn by servers, storage, network switches, and the directly supporting IT hardware (KVM, monitors, workstations in the white space). | Cooling, power distribution losses, lighting, and every other facility system — these are the overhead the metric is designed to expose. |
Two engineering notes matter. First, PUE is properly an energy ratio (kWh over a period), not an instantaneous power ratio (kW at a moment) — instantaneous “PUE” readings swing with load and weather and are not comparable between sites. Second, the measurement period must be the same for numerator and denominator; the standard’s preferred reporting period is a full year, which averages out seasonal cooling swings.
Measurement Boundaries — IT Load vs Facility Load
The single largest source of non-comparable PUE numbers is inconsistent boundary definition. Two facilities can report 1.3 and 1.5 not because one is more efficient, but because they drew the IT-load boundary in different places. ISO/IEC 30134-2 exists largely to close this gap.
Where the IT-equipment boundary sits
The decisive question is the metering point at which “IT equipment energy” is captured. Measuring further downstream — closer to the server power feed — captures power-conversion and distribution losses as overhead, producing a higher (more honest) PUE. Measuring further upstream — at the UPS output — folds some of those losses invisibly into the IT term, flattering the number. The measurement categories below formalise this.
The on-site generation and unaccounted-energy traps
Two boundary edge cases are common findings in assurance. On-site generation — rooftop solar, fuel cells, behind-the-meter plant — must be accounted consistently, because energy generated and consumed inside the boundary changes what “total facility energy” means; the 2026 edition adds explicit guidance here. Unaccounted energy — loads that are real but unmetered — must be estimated and disclosed rather than silently dropped, which would understate the total and depress PUE.
A PUE quoted without its measurement category and boundary definition is uninterpretable. “Our PUE is 1.2” is a marketing claim until the operator states where IT energy is metered, what period the figure covers, and how on-site generation and unaccounted energy were treated. Always demand the category alongside the number.
The PUE Measurement Categories (PUE0–PUE3)
ISO/IEC 30134-2 defines measurement categories that distinguish how rigorously a PUE figure was captured. The categories are about measurement quality — temporal resolution and metering rigour — not about whether the resulting number is good or bad. A facility with a genuine PUE of 1.6 measured at Category 3 is more credible than one claiming 1.3 at an estimate level.
| Category | Resolution | Measurement interval | Typical use |
|---|---|---|---|
| PUE0 | Estimate | Single demand reading or empirical estimate; no continuous metering | Initial assessment, early-design sizing, facilities without instrumentation |
| PUE1 — Basic | Basic | Energy totals over a period, commonly monthly | The common operational baseline; IT energy typically read at the UPS output |
| PUE2 — Intermediate | Intermediate | Energy totals at finer cadence, commonly daily | Sites with PDU-level metering and tighter instrumentation |
| PUE3 — Advanced | Advanced | Continuous measurement, typically every 15 minutes or finer | Highly instrumented facilities with DCIM; the highest-credibility figure |
When comparing facilities or reporting to a regulator, hold the category constant. A PUE3 annual figure is the gold standard for comparability because it captures the full seasonal cooling swing at high resolution. A PUE0 estimate is acceptable for early design but should never be presented as an operational result.
A note on convention: some industry write-ups invert the category labels, mapping higher numbers to coarser utility-meter readings. The authoritative ISO/IEC 30134-2 framing — used here — orders the categories by increasing measurement resolution, with PUE3 the most granular and rigorous. Always cite the standard’s definition rather than a vendor’s.
Governance: The Green Grid, Uptime Institute, ISO/IEC 30134-2
PUE has three governance layers, which is why the metric is sometimes attributed to The Green Grid, sometimes to Uptime Institute, and sometimes to ISO. All three are correct for different parts of its life.
The Green Grid (origin). PUE was introduced by The Green Grid, a non-profit industry consortium, in a 2007 white paper, alongside its reciprocal Data Center Infrastructure Efficiency (DCiE). The Green Grid authored the foundational definitions and the early measurement-category framework that ISO later formalised.
Uptime Institute (stewardship). The Green Grid’s assets and ongoing stewardship sit within Uptime Institute, the organisation best known for its data-centre Tier classification. Uptime publishes the long-running annual global data-centre survey that supplies the industry’s headline PUE trend, and maintains the practitioner guidance around the metric.
ISO/IEC JTC 1/SC 39 (standardisation). The international standard ISO/IEC 30134-2 was developed by the joint ISO/IEC subcommittee responsible for sustainability for and by information technology. It turned the Green Grid’s industry convention into a formal international standard with defined measurement points, calculation rules, and reporting requirements — the version regulators and auditors now reference. The first edition published in 2016; the second edition, ISO/IEC 30134-2:2026, replaced it in January 2026.
This separation — industry originator, steward, and formal standards body — mirrors the structure seen across the disclosure ecosystem, where a metric’s creator and its formal standard are distinct. For the energy-management system context that often sits alongside PUE governance, see the ISO 50001 Energy Management reference, and for the broader data-centre design and facility standards, see ISO/IEC 22237 / EN 50600 data-centre design.
What Changed in the 2026 Edition
The second edition of ISO/IEC 30134-2 published on 16 January 2026 and cancelled and replaced the 2016 first edition, which was withdrawn the same day. The revision is technical rather than conceptual — the formula and the category framework are unchanged — but it tightens several of the boundary edge cases that produced non-comparable numbers under the first edition.
| Area | 2026 edition change |
|---|---|
| Mixed-use buildings | New guidance for facilities where the data centre shares a building with non-data-centre functions, clarifying how to draw the boundary so shared loads are not misattributed. |
| On-site generation | Explicit rules for handling behind-the-meter generation (solar, fuel cells, on-site plant) so self-generated energy is accounted consistently in the total-facility term. |
| Unaccounted energy | Defined treatment for loads that are real but unmetered, requiring estimation and disclosure rather than omission. |
| Measurement and documentation | Refined measurement requirements and energy-documentation expectations to improve comparability across facilities and over time. |
Any guide, audit template, or supplier brief that still cites ISO/IEC 30134-2:2016 as the operative version is out of date as of January 2026. The 2016 edition is withdrawn. Confirm that PUE figures presented for comparability are computed under the 2026 edition’s boundary rules, particularly for mixed-use sites and facilities with on-site generation.
Interpreting PUE — Benchmarks and the “Good PUE” Myth
There is no single threshold that separates a “good” PUE from a “bad” one, because the achievable floor depends heavily on climate, facility age, design, and IT load profile. A site in a cool, dry climate can use free-air cooling for much of the year and approach 1.1; the same design in a hot, humid climate cannot. Benchmarking PUE without controlling for climate and facility type is one of the most common analytical errors.
The broad industry picture, drawn from the long-running Uptime Institute global survey and EU reporting data, looks like this:
The industry average has improved markedly since the metric’s introduction, when typical figures sat around 2.0, but the rate of improvement has plateaued near 1.5 in recent years — the large efficiency gains from hot/cold-aisle containment, free cooling, and higher operating temperatures have mostly been captured, and further gains are incremental. The headline trend over time:
| Point | PUE |
|---|---|
| 2011 | 1.98 PUE |
| 2014 | 1.70 PUE |
| 2017 | 1.58 PUE |
| 2020 | 1.59 PUE |
| 2023 | 1.55 PUE |
| 2025 | 1.56 PUE |
The three archetypes below show why a single benchmark misleads. Each can be well-run for its context.
Hyperscale (cool climate)
Purpose-built, free-air cooling, high operating temperatures, optimised power chain. PUE near 1.1. The overhead is squeezed almost to the floor — but the carbon depends entirely on the grid or procurement behind it.
Colocation (mixed climate)
Multi-tenant, varied IT density, contractual cooling commitments. PUE typically 1.4–1.6. The operator controls facility load; tenants control IT load — splitting accountability for the ratio.
Enterprise edge (warm climate)
Older room, retrofit cooling, low and variable utilisation. PUE 1.8–2.5. Often the largest improvement opportunity, but constrained by building fabric and a small IT base that makes overhead proportionally heavy.
Sister Metrics — WUE, REF, ERF, CER, DCiE
PUE captures energy overhead and nothing else. A complete data-centre sustainability picture needs companion metrics for water, renewable share, heat reuse, and cooling efficiency — several of them defined in the same ISO/IEC 30134 series. The EU EED reporting scheme requires PUE, WUE, ERF, and REF together precisely because no single one is sufficient.
| Metric | Full name | What it measures | Standard / source |
|---|---|---|---|
| PUE | Power Usage Effectiveness | Total facility energy ÷ IT energy — overhead efficiency | ISO/IEC 30134-2 |
| DCiE | Data Center Infrastructure Efficiency | The reciprocal of PUE, expressed as a percentage (1 ÷ PUE) | The Green Grid (2007) |
| WUE | Water Usage Effectiveness | Litres of water consumed per kWh of IT energy | ISO/IEC 30134-9 |
| REF | Renewable Energy Factor | Share of facility energy from renewable sources | ISO/IEC 30134-3 |
| ERF | Energy Reuse Factor | Share of energy reused outside the data centre (e.g. waste-heat capture) | ISO/IEC 30134-6 |
| CER | Cooling Efficiency Ratio | Cooling output relative to cooling energy input | ISO/IEC 30134-7 |
PUE tells you how much energy the facility wastes as overhead. REF tells you how much of that energy is renewable. Carbon accounting needs both, plus a grid emission factor — which is exactly the bridge built in the next section. A facility can hold REF and PUE constant and still change its emissions if the underlying grid factor moves.
Known Limitations and Gaming Vectors
PUE is a useful metric used badly more often than it is used well. The limitations are structural, not incidental, and a sustainability team should know each one before relying on a quoted figure.
PUE rewards a smaller denominator but is blind to whether the IT load itself is doing useful work. A facility running idle, over-provisioned servers at high power can post a flattering PUE while wasting most of its compute. PUE captures infrastructure overhead, not workload productivity — for which separate “useful work per energy” metrics exist.
Because fixed cooling and power overhead is spread across whatever IT load exists, raising IT power — even with inefficient hardware — can lower PUE. A site can “improve” its PUE by running more servers harder, which is the opposite of an efficiency gain. PUE must always be read alongside utilisation.
Metering IT load at the UPS output rather than the rack inflates the IT term and depresses PUE; quoting a favourable winter month rather than an annual figure does the same. A PUE reported without its category, boundary, and period is not a comparable number.
The same design yields different PUE in different climates. Ranking facilities on raw PUE penalises operators in hot climates for physics they cannot change, and rewards location choice over engineering. Benchmark within climate bands, not across them.
PUE → Scope 2 Emissions: The Accounting Bridge
This is the section most PUE references omit, and the reason a sustainability team reads about PUE at all. PUE is an energy-efficiency multiplier; it becomes a carbon figure only when combined with the IT load and a grid emission factor. The bridge has three steps.
Step 1: Total facility energy = IT energy × PUE. Step 2: Scope 2 emissions (location-based) = total facility energy × grid emission factor. Step 3: Scope 2 emissions (market-based) = total facility energy × contractual instrument factor (supplier-specific rate, EAC-backed rate, or residual mix). PUE enters only at Step 1 — it scales the energy, not the carbon intensity.
The consequence is structural: improving PUE reduces total facility energy and therefore reduces emissions proportionally, but it cannot change the carbon intensity of each kilowatt-hour. Two levers act on a data centre’s Scope 2 footprint — the PUE (how much energy) and the emission factor (how dirty each unit is) — and they are independent. A facility that cuts PUE from 1.6 to 1.3 reduces its energy and emissions by roughly 19% with no change in procurement; a facility that signs a renewable PPA can drive market-based emissions toward zero with no change in PUE.
The grid emission factor that completes the bridge comes from the same sources GreenCalculus uses across its Scope 2 tooling: sub-national factors such as EPA eGRID subregions in the United States, national factors from UK DEFRA, and the consolidated set in the grid emission factors 2026 dataset. The accounting rules that govern which factor applies — and the mandatory dual location-based and market-based reporting — are set by the GHG Protocol Scope 2 Guidance, and the underlying concepts are covered in the Scope 2 emissions and electricity emission factor glossary entries.
Turn a PUE figure and an IT load into a location-based and market-based Scope 2 number in the same view.
Why a Low PUE Is Not a Low-Carbon Data Centre
The headline confusion in data-centre sustainability is treating PUE as a carbon metric. It is not, and the gap is large enough to invert rankings. A facility with PUE 1.1 on a grid at 0.50 kg CO₂e/kWh emits more per unit of compute than a facility with PUE 1.6 on a grid at 0.05 kg CO₂e/kWh — the efficient site loses on carbon by an order of magnitude because the grid factor dominates.
This is why the carbon picture needs the market-based layer. A data centre can hold its physical grid exposure (the location-based number) while driving its contractual emissions toward zero through renewable procurement — power purchase agreements, energy attribute certificates, or supplier-specific renewable rates. PUE is silent on all of this. The renewable-procurement decision sits entirely in the emission-factor term of the bridge, governed by the RE100 Technical Criteria and operationalised through energy attribute certificates. The distinction between the two carbon measures is set out in Scope 2 location-based and Scope 2 market-based.
Cutting PUE shrinks the energy bill and the emissions proportionally but cannot decarbonise a single kilowatt-hour. Procuring clean energy decarbonises every kilowatt-hour but does nothing for efficiency. A credible net-zero data-centre strategy pulls both: a low PUE to minimise the energy, and verified renewable procurement to minimise the carbon per unit. Reporting PUE as a climate achievement, without the emission factor, is the error to avoid.
PUE in Regulation — EU EED, Germany, CSRD
PUE moved from a voluntary industry benchmark to a regulated reporting metric over 2023–2026, most consequentially in the European Union. A sustainability team operating EU data centres must treat PUE as a compliance datapoint, not just an operational KPI.
EU Energy Efficiency Directive (2023/1791)
Under Article 12 of the recast Energy Efficiency Directive and its supplementary Commission Delegated Regulation (EU) 2024/1364, data centres with an installed IT power demand of 500 kW or more must report their energy performance annually to a European database. The reporting set includes PUE alongside WUE, ERF, REF, floor area, installed power, energy consumption, temperature set points, waste-heat utilisation, and renewable energy use. The Delegated Regulation, in force since June 2024, specifies the full data set; the first report covered the 2023 calendar year, with subsequent reports due annually by 15 May. Defence and civil-protection facilities are exempt.
A persistent misconception is that the EED mandates a PUE of 1.2 across the EU. It does not. The Directive mandates reporting and leaves performance thresholds to national discretion. The binding PUE limits people are thinking of come from member-state law, principally Germany.
Germany — Energieeffizienzgesetz (EnEfG)
Germany’s Energy Efficiency Act is the strictest data-centre performance law in the EU. New facilities commissioned from July 2026 must reach a PUE of 1.2 within two years of operation. Existing facilities follow a graduated schedule — PUE ≤1.5 by July 2027, then ≤1.3 by July 2030. The Act also ramps waste-heat reuse (ERF) targets and renewable-energy requirements, and from January 2026 obliges larger facilities to operate an energy-management system certified to ISO 50001 or EMAS, with annual reporting to BAFA.
CSRD and broader disclosure
Data-centre operators in scope of the Corporate Sustainability Reporting Directive report energy and emissions under ESRS E1, where PUE underpins the energy-consumption figures and the Scope 2 line that flows from them. The directive’s value-chain reach also pulls smaller operators into data collection as suppliers to in-scope customers. Where a data centre’s electricity falls under an emissions-trading or carbon-pricing regime, the energy total that PUE helps establish feeds into instruments such as the EU Emissions Trading System.
Common Misinterpretations
PUE measures energy overhead, not carbon. Emissions depend on the grid emission factor or procurement, which PUE does not capture. A 1.1 PUE on a fossil grid can out-emit a 1.6 PUE on a clean grid.
The EU EED mandates PUE reporting for facilities ≥500 kW, not a PUE threshold. The 1.2 limit is German national law (EnEfG) for new builds from July 2026, not an EU-wide cap.
Only if measured at the same category, boundary, and period in comparable climates. Raw cross-site PUE comparison without those controls is meaningless and often misleading.
The category describes measurement rigour, not efficiency. A PUE3 figure of 1.6 is more credible than a PUE0 estimate of 1.3. Category and value are independent.
Because overhead spreads across IT load, increasing IT power — even inefficiently — can lower PUE. PUE must be read with utilisation, or the number can move the wrong way for the right reasons.
Common Measurement and Reporting Errors
- Reporting instantaneous power PUE as annual PUE. A kW snapshot is not a kWh-over-a-year figure; the two are not comparable and the standard’s preferred reporting period is annual.
- Metering IT load at the UPS output and calling it Category 3. The metering point and the temporal category are separate questions; misstating either inflates credibility the figure has not earned.
- Dropping unaccounted energy. Omitting real-but-unmetered loads understates total facility energy and depresses PUE — the 2026 edition requires estimation and disclosure instead.
- Mishandling on-site generation. Behind-the-meter solar or fuel cells must be accounted consistently in the total-facility term; inconsistent treatment makes the ratio non-comparable.
- Quoting a favourable season. A winter free-cooling month flatters PUE; comparability requires a full annual period that captures the cooling swing.
- Treating PUE as the carbon metric in disclosure. Under ESRS E1 and similar regimes, PUE supports the energy figure but the Scope 2 line requires a grid or contractual emission factor — applying PUE alone is a category error.
- Comparing across climates without banding. Cross-climate PUE league tables penalise physics, not engineering.
Worked Examples
Three worked examples connect the PUE ratio to energy and then to emissions. All figures are illustrative and hardcoded to show the calculation chain — they are not live grid factors for any specific market.
| IT equipment energy (annual) | 8,000,000 kWh |
| Measured PUE (Category 3, annual) | 1.45 |
| Total facility energy = 8,000,000 × 1.45 | 11,600,000 kWh |
| Non-IT overhead = total − IT | 3,600,000 kWh |
| Total facility energy (from Example A) | 11,600,000 kWh |
| Illustrative grid factor (location-based) | 0.207 kg CO₂e/kWh |
| Location-based Scope 2 = 11,600,000 × 0.207 | 2,401 t CO₂e |
| Lever | Action | Resulting Scope 2 |
|---|---|---|
| Baseline | PUE 1.45, grid factor 0.207 | 2,401 t CO₂e |
| Efficiency lever | Cut PUE to 1.25 (energy → 10,000,000 kWh) | 2,070 t CO₂e (−14%) |
| Procurement lever | Keep PUE 1.45, cover energy with compliant EACs (market-based) | ~0 t CO₂e market-based |
Example C is the whole argument in one table: the efficiency lever and the procurement lever are independent, and the location-based figure is reported in parallel with the market-based figure regardless of which lever is pulled. To run a scenario like this against current factors and proper dual reporting, use the Scope 2 electricity calculator.
Implementation Checklist
For a data-centre operator standing up PUE measurement that survives both EU EED reporting and corporate carbon disclosure, the practical workflow runs as follows.
- Define the boundary. Draw the data-centre boundary explicitly, applying the 2026 mixed-use-building guidance where the facility shares a building.
- Place the meters. Decide and document where IT-equipment energy is captured; meter as close to the IT load as the instrumentation allows.
- Set the category. Target Category 3 (continuous) for regulatory and disclosure use; document the temporal resolution honestly.
- Use an annual period. Report PUE over a full year to capture the seasonal cooling swing; do not quote a favourable month.
- Handle on-site generation and unaccounted energy. Account behind-the-meter generation consistently and estimate-and-disclose unmetered loads per the 2026 edition.
- Compute the sister metrics. Capture WUE, ERF, and REF alongside PUE — the EU EED requires the set together.
- Bridge to emissions. Multiply total facility energy by the appropriate grid factor for location-based Scope 2, and by contractual instruments for market-based Scope 2.
- Report to the right destinations. The European database for EED; ESRS E1 for CSRD-in-scope entities; national registers such as BAFA where applicable.
Test your PUE-to-Scope-2 chain, then check overall target alignment.
Frequently Asked Questions
PUE is the ratio of the total energy a data centre consumes to the energy consumed by its IT equipment, over the same period. It is standardised by ISO/IEC 30134-2 and expresses infrastructure overhead: a PUE of 1.0 is the theoretical ideal (no non-IT overhead), while a PUE of 1.5 means the facility draws 1.5 units of energy for every unit reaching IT equipment. PUE measures energy efficiency, not carbon.
There is no universal threshold because the achievable PUE depends on climate, facility age, and design. Best-in-class hyperscale facilities reach about 1.1; modern well-run facilities sit around 1.3–1.4; the global industry average is near 1.5; and legacy enterprise rooms often exceed 2.0. A figure should always be judged within its climate band and against its measurement category, not against an absolute target.
PUE = Total Facility Energy ÷ IT Equipment Energy, which is equivalent to 1 + (Non-IT Energy ÷ IT Equipment Energy). Both terms are measured in energy units (kWh) over the same period, with a full year as the preferred reporting period. Total facility energy includes cooling, power-conversion losses, lighting, and all supporting systems; IT equipment energy includes only servers, storage, and networking hardware.
They are measurement categories defined by temporal resolution. PUE0 is an estimate with no continuous metering; PUE1 (Basic) uses energy totals at a basic cadence, commonly monthly; PUE2 (Intermediate) uses finer, commonly daily, totals; and PUE3 (Advanced) uses continuous measurement, typically every 15 minutes. The category describes how rigorously the figure was measured, not whether the facility is efficient.
No. PUE measures energy overhead, not carbon intensity. Emissions depend on the grid emission factor or renewable procurement, which PUE does not capture. A facility with a PUE of 1.1 on a fossil-heavy grid can emit more per unit of compute than a facility with a PUE of 1.6 on a clean grid. To convert PUE to emissions, multiply total facility energy by a grid emission factor (location-based) or by contractual instruments (market-based) under the GHG Protocol Scope 2 Guidance.
PUE is standardised by ISO/IEC 30134-2, developed by ISO/IEC JTC 1/SC 39. The metric originated with The Green Grid in 2007 and is stewarded by Uptime Institute, but the formal international standard is ISO/IEC 30134-2. The second edition, ISO/IEC 30134-2:2026, published in January 2026 and replaced the 2016 first edition, adding guidance for mixed-use buildings, on-site generation, and unaccounted energy.
In the EU, yes, for reporting. Under the Energy Efficiency Directive (2023/1791) Article 12 and Delegated Regulation (EU) 2024/1364, data centres with 500 kW or more of installed IT power must report PUE — along with WUE, ERF, and REF — annually to a European database, due by 15 May each year. The EED does not set a PUE limit; binding thresholds come from national law, such as Germany’s EnEfG, which requires a PUE of 1.2 for new facilities commissioned from July 2026.
In three steps. First, total facility energy = IT energy × PUE. Second, location-based Scope 2 = total facility energy × grid emission factor (for example, an EPA eGRID subregion or DEFRA national factor). Third, market-based Scope 2 = total facility energy × contractual instrument factor (supplier-specific rate, EAC-backed rate, or residual mix). PUE only scales the energy at the first step; the emission factor determines the carbon intensity.
WUE (Water Usage Effectiveness), REF (Renewable Energy Factor), ERF (Energy Reuse Factor), and CER (Cooling Efficiency Ratio) — several defined in the same ISO/IEC 30134 series — plus DCiE, the reciprocal of PUE expressed as a percentage. The EU EED requires PUE, WUE, ERF, and REF together because no single metric captures energy overhead, water, renewable share, and heat reuse at once.
The large efficiency gains — hot/cold-aisle containment, free-air cooling, higher operating temperatures, and improved power chains — have mostly been captured across the modern fleet, so the industry average has plateaued near 1.5 in recent years after falling from roughly 2.0 a decade earlier. Further gains are incremental, and the headline number is also weighed down by the long tail of older enterprise facilities that are hard to retrofit.
Sources and References
Every numerical claim and definitional statement on this page reconciles to the primary sources below. Where a definitive standard exists, it is cited directly; benchmark ranges are drawn from the named industry and regulatory datasets.
Primary standard
- ISO/IEC 30134-2:2026, Information technology — Data centres key performance indicators — Part 2: Power usage effectiveness (PUE) (2nd edition, published 16 January 2026; cancels and replaces the 2016 first edition).
- ISO/IEC 30134-1, Overview and general requirements; ISO/IEC 30134-3 (REF); ISO/IEC 30134-6 (ERF); ISO/IEC 30134-7 (CER); ISO/IEC 30134-9 (WUE). Developed by ISO/IEC JTC 1/SC 39.
- The Green Grid, The Green Grid Data Center Power Efficiency Metrics: PUE and DCiE (White Paper #6, 2007) and subsequent measurement-protocol white papers.
Regulation
- Directive (EU) 2023/1791 (Energy Efficiency Directive recast), Article 12 and Annex VII.
- Commission Delegated Regulation (EU) 2024/1364 (data-centre reporting scheme; in force June 2024).
- Germany, Energieeffizienzgesetz (EnEfG) — data-centre PUE, ERF, renewable-energy, and ISO 50001/EMAS provisions.
Benchmark data
- Uptime Institute, annual Global Data Center Survey (industry-average PUE trend).
- European Commission / EU EED reporting database (average PUE by data-centre type).
Related GreenCalculus reference pages
- ISO/IEC 22237 / EN 50600 data-centre design
- ISO 50001 Energy Management
- GHG Protocol Scope 2 Guidance
- EPA eGRID
- UK DEFRA Emission Factors
- RE100 Technical Criteria
- CSRD / ESRS E1
- EU Emissions Trading System
- Scope 2 emissions
- Scope 2 (location-based)
- Scope 2 (market-based)
- Electricity emission factor
- Energy Attribute Certificates
- Grid emission factors 2026
- Scope 2 electricity methodology
- Scope 2 Electricity Calculator
- SBTi Readiness Checklist
What changed in this revision
Published 20 June 2026. Initial publication. Documents PUE as governed by ISO/IEC 30134-2:2026 (second edition, January 2026): the formula and its two terms, the measurement boundary edge cases, the PUE0–PUE3 measurement categories, the Green Grid / Uptime Institute / ISO/IEC JTC 1/SC 39 governance lineage, the sister metrics (WUE, REF, ERF, CER, DCiE), the known limitations and gaming vectors, the PUE→Scope 2 emissions bridge, the renewable-procurement distinction, and the regulatory regime (EU EED 2023/1791, Delegated Regulation (EU) 2024/1364, Germany’s EnEfG, CSRD/ESRS E1).