Data center sustainability reporting and metrics field guide

What data center sustainability reporting is

Data center sustainability reporting is the measured disclosure of how much energy, water, and carbon a facility uses to do its computing, stated against a boundary you have defined and a framework you have named. It is not a single PUE number on a slide. A credible report carries a set of metrics, each covering something the others miss, and it is honest about how the renewable energy was counted.

The industry got into this position by treating PUE as the whole story for the better part of a decade. PUE measures energy overhead and nothing else. It says nothing about whether the power was clean, whether the cooling drank water, or whether the heat was thrown away or reused. A plant can post a world-class PUE on a coal grid while evaporating millions of liters a day, and the energy scorecard alone will never show it.

So the working definition of a real program is four things: measure the right metric set, draw and disclose the boundary, report carbon both ways across all three scopes, and avoid dressing up an annual average as clean power. The PUE guide covers the energy metric in depth and the WUE guide covers water. This guide is the level above both, the reporting program that ties the metrics together and stands up when a regulator or an auditor asks how you got the number.

Why data centers are under the spotlight now

The pressure is new because the load is. The International Energy Agency projects data center electricity demand roughly doubling from about 415 TWh in 2024 to around 945 TWh by 2030, near 3 percent of global electricity, with demand from AI-optimized servers more than quadrupling over that span. In the United States, data centers are projected to account for nearly half of electricity demand growth to 2030. A sector that was a rounding error on the grid a decade ago is now a line item utilities and regulators plan around.

Water moved the same direction and later than energy. Amazon disclosed about 2.5 billion gallons of water across its data centers in a single year, and Google and Microsoft have reported figures in the same league. Those disclosures put the water question in front of communities and water authorities in drought-affected regions, and a string of projects have been delayed or blocked over it. The WUE guide covers that fight in detail.

And the reporting is going from voluntary to mandatory. The EU Energy Efficiency Directive now requires annual reporting from larger data centers in the bloc, several jurisdictions are following, and large tenants are writing sustainability disclosure into their contracts. The credibility stakes are real: a plant that cannot produce a measured, defensible set of numbers is a plant that cannot answer its regulator, its customers, or the planning board for its next site. Treat sustainability reporting as a permit input, not a marketing exercise.

The metric set: no single number tells the story

There is no one number for data center sustainability, and any report that offers one is hiding something. The credible approach is a family of metrics that share the same denominator, the IT energy, so they read together. Each one was created to expose a cost that the others let through.

PUE measures energy overhead, the power spent on cooling and losses per watt of compute. WUE measures water, the liters used per kilowatt-hour of IT energy, which exists precisely because the cheapest way to cut PUE is often to spend water. CUE measures carbon, the kilograms of CO2 per kilowatt-hour, which folds in how dirty the grid is. ERF measures heat reuse, the share of energy sent off-site for productive use instead of thrown away. The Green Grid originated the family and the ISO/IEC 30134 series carries several of them.

Read alone, each metric flatters a different bad decision. A great PUE can sit on a heavy water draw. A clean CUE on paper can rest on annual REC accounting that hides a dirty grid hour by hour. The point of the set is that you cannot win all of them by gaming one, so report them together against the same boundary. The PUE and WUE guides cover the first two in depth; the rest of this guide covers carbon, reuse, the boundary, and the honesty that ties the report together.

PUE: the energy metric and its limits

Power usage effectiveness is total facility energy divided by the energy reaching the IT equipment, a ratio at or above 1.0 where lower is better. It is the most recognized data center metric and the one a report cannot skip, because it is the headline efficiency figure in most disclosures, energy codes, and the EU reporting rule. The PUE guide covers the math, the measurement categories, and how to meter it.

What matters for the sustainability report is what PUE does not cover. It measures overhead and only overhead. It says nothing about carbon, because a low PUE on a dirty grid still emits heavily. It says nothing about water, because the evaporative cooling that buys a low PUE can drive WUE several times the average. And it says nothing about whether the IT itself is doing useful work, because idle servers burning power sit in the denominator and make the ratio look good.

So PUE belongs in the report as one axis, annualized and stated with its measurement category, never as the whole answer. The single most common error in data center sustainability is reporting a flattering PUE and going quiet on everything else. Pair it with at least WUE on any site that uses water and CUE on any site you are accounting carbon for, and the boundary the PUE was measured against has to match the others.

WUE: water and the trade against PUE

Water usage effectiveness is the water a facility uses divided by the IT energy, in liters per kilowatt-hour, lower being better. The Green Grid published it in 2011 after PUE had already become the energy scorecard, because the cheapest path to a low PUE is often evaporative cooling, which trades electricity for water. WUE puts that water back on the board. The WUE guide covers the cooling-water mechanics, the site-versus-source split, and the levers that cut the draw.

For the report, the key fact is that PUE and WUE pull against each other through the cooling design. Lean hard on the cooling tower and the PUE drops while the WUE climbs. Go air-cooled to zero the on-site water and the PUE rises, worst on the hottest days. There is no design that wins both axes at once, so a report that shows only the flattering metric is hiding the trade, not avoiding it.

State the basis every time: site WUE counts on-site cooling water, source WUE adds the water used to generate the electricity, and the two are not interchangeable. An air-cooled plant on a thirsty grid has moved its water problem upstream, where source WUE catches it and site WUE does not. Report WUE alongside PUE on any site that uses water for cooling, on a stated site-or-source basis and over a full year.

What is CUE?

Carbon usage effectiveness is the carbon emitted per unit of IT energy, in kilograms of CO2 equivalent per kilowatt-hour, where the ideal is 0.0. It is the metric that turns energy into emissions by folding in how dirty the power is. A common way to compute it is CUE equals the carbon emission factor of the grid times the PUE, so a plant on clean power scores well even at a mediocre PUE, and a plant on a dirty grid scores poorly even at a good one.

That relationship is the whole reason CUE matters for sustainability rather than just efficiency. PUE tells you how much total energy you spend per unit of compute. CUE tells you what that energy did to the atmosphere, which depends on the grid carbon intensity at the meter. Two identical halls, one in a hydro-rich region and one on a coal grid, can post the same PUE and wildly different CUE.

The honesty problem is which carbon emission factor you used. As The Green Grid defines it, CUE is an operational metric: it covers the carbon of the energy consumed and excludes the embodied carbon of the building and the IT hardware. And the factor itself can be a location-based grid average or a market-based figure adjusted for purchased renewables, which can differ by a wide margin. State the factor, its source and year, and whether it is market-based or location-based, or the CUE is a number a reviewer cannot trust.

ERF: crediting the heat you reuse

Energy reuse factor is the share of energy a data center sends off-site for productive use instead of rejecting it, on a scale of 0.0 to 1.0, where 0.0 means nothing is reused and 1.0 would mean all the energy brought in leaves again as useful heat. It is one of the four metrics the EU reporting rule asks for, alongside PUE, WUE, and the renewable energy factor, so it has moved from a nice-to-have to a disclosed number for larger EU sites.

The heat a data center makes is low-grade, warm rather than hot, which has historically held back reuse. Warm-water liquid cooling changes that, because it delivers heat at a temperature a district heating loop, a greenhouse, or an industrial process can actually use without a large heat pump. Where there is a heat customer next door, the energy that would have gone up an evaporative tower goes into productive use, and the ERF rises.

The related metric ERE, energy reuse effectiveness, expresses the same idea as a modified PUE that subtracts the reused energy, and it can fall below 1.0 because the reuse is credited out. The WUE guide covers how reuse also trims the water a plant has to evaporate. The honest caveat is that reuse needs a matched, nearby heat customer, which most sites do not have, so a low ERF is often a siting fact rather than a failure. Report it for what it is and do not credit heat that has no buyer.

Define the reporting boundary or the numbers lie

Every metric in the report is a ratio or a total, and every one of them changes with where you draw the boundary. The boundary is the line around what counts as inside the facility, what counts as IT, and what counts as someone else's responsibility. Draw it loosely and the numbers are not wrong by a little, they are not comparable to anyone else's and not defensible to an auditor.

The first cut is IT versus facility. PUE, WUE, and CUE all divide by IT energy, so the meter point that defines IT energy sets the denominator for all of them. Read it at the UPS output and the ratio looks better than reading it at the rack, because more of the downstream loss counts as IT instead of overhead. The PUE guide covers the Green Grid measurement categories that pin this down.

The second cut is the colocation split, and it is where reporting gets political. In a colo, the operator controls the cooling and the power chain while the tenant controls the IT and buys its own power for it. Who reports the IT energy, who reports the purchased-electricity carbon, and who owns the water all depend on the contract and the metering. A landlord-controlled report and a tenant-controlled report of the same hall can both be honest and still not add up, because they drew the boundary in different places. Write the boundary down, in the report, so the next reader knows exactly what is in and what is out.

Carbon scopes 1, 2, and 3 under the GHG Protocol

Carbon accounting splits emissions into three scopes under the GHG Protocol, and a credible report covers all three rather than the convenient one. Scope 1 is direct emissions from sources the operator owns or controls, which for a data center is mainly the on-site fuel: the backup generators when they run and test, and any gas-fired equipment. It is usually the smallest of the three for a running facility, but it is not zero, and generator testing burns real diesel.

Scope 2 is the indirect emissions from the purchased electricity that runs the plant, and for a data center it is almost always the dominant operational source. This is where the grid carbon intensity and the entire renewable-procurement question live, and it is the scope the market-versus-location debate is about. Get scope 2 right and you have the bulk of the operational carbon; get it wrong and the headline number is wrong.

Scope 3 is everything else in the value chain: the embodied carbon in the building and the construction, the manufacturing and shipping of the servers, network gear, and storage, and the hardware refresh cycle that replaces them every few years. It is the hardest to measure and the most often left out, and for a data center it can be a large share of the lifetime total. The SBTi ICT guidance points operators at the scope 3 categories that matter most for the sector, purchased goods and services and use of sold products. Report all three scopes, and say which scope 3 categories you included and which you could not yet quantify.

What is the difference between market-based and location-based carbon?

Location-based carbon accounting values your electricity at the average emissions intensity of the grid you are physically plugged into, regardless of what you bought. Market-based accounting values it at the emissions of the specific power you contracted for, through renewable energy certificates or power purchase agreements, with any uncovered load priced at the residual grid mix. The GHG Protocol asks companies to report both, which is called dual reporting, and the gap between the two numbers is where the honesty lives.

Here is why both matter and why this is the part to read twice. The market-based number is the one operators quote, because a PPA or a stack of RECs can drive it toward zero on paper. The location-based number is the one that reflects the actual carbon coming out of the wires that ran your servers. A plant can report a market-based figure near zero while its location-based figure shows a heavy draw on a fossil grid, and both can be technically correct under the protocol.

The credible report shows both and does not let the market-based number stand alone. The market-based figure tells you what you procured. The location-based figure tells you the real grid impact your load caused, hour by hour, in the place it actually happened. When the two diverge a lot, that divergence is the report's most useful disclosure, not something to bury. A report that quotes only the market-based number, with no location-based companion, is making a claim the grid does not support, and an auditor or a sharp tenant will ask for the other figure first.

RECs, PPAs, and the renewable claim

Renewable energy gets onto the books through certificates and contracts, and the credibility of a renewable claim depends entirely on which kind and how it was matched. An unbundled REC is a certificate bought separately from the power, attesting that a megawatt-hour of renewable energy was generated somewhere, sometime, on some grid. A power purchase agreement is a contract for the output of a specific project, which is a stronger link between your money and new clean generation. They are not the same quality of claim even though both can zero out a market-based number.

The weak end of the spectrum is buying unbundled certificates to match annual consumption with no relationship to where or when your load actually ran. The renewable energy factor, REF, that the EU rule asks for is a useful disclosure here, but a high REF built on annual certificate matching does not mean the electrons feeding the hall were clean at the hour they were drawn. That is the credibility question the next two sections turn on.

The stronger end is a PPA that adds new renewable capacity on or near the grid that serves the site, ideally one that can be matched to consumption on a fine time resolution. The honest framing for a report is to disclose the instrument, the additionality, and the matching basis, not just the percentage. A renewable claim with no detail behind it is a claim a reviewer cannot weigh, and increasingly one a regulator will discount.

What is 24/7 carbon-free energy?

24/7 carbon-free energy, often shortened to CFE, means matching electricity consumption with carbon-free generation on every hour of every day, in the same grid region, rather than averaging it out over a year. It is the granular alternative to annual matching, and the difference between the two is the heart of the credibility debate in data center carbon reporting. Google put the term on the map with a 2020 white paper and reports its progress against it; as of 2024 it cited a global average around 66 percent CFE, with about 9 of 20 grid regions at or above 80 percent hourly matching.

The reason hourly matching is harder, and more honest, is the clock. Solar generates at midday and a data center runs all night. Match annual totals with certificates and you can claim 100 percent renewable while drawing fossil power for the hours the sun is down. One worked example often cited: a load in Ireland matched to 100 percent renewable energy on an annual basis is only about 85 percent carbon-free when measured hour by hour. The annual figure and the hourly figure are different numbers, and only the hourly one tracks what the grid actually delivered.

Be honest about where this debate sits. 24/7 CFE is widely seen as the more credible standard, and bodies like the NewClimate Institute argue for hourly matching over looser annual approaches in the GHG Protocol and SBTi rules. It is also genuinely hard and expensive to reach, the data and tooling are still maturing, and the standards have not fully settled. A report does not have to be at 24/7 to be credible, but it does have to be clear about whether its renewable claim is annual or hourly, because presenting an annual average as fully clean power is the claim that draws the greenwashing charge.

The greenwashing pitfalls to avoid

Greenwashing in data center reporting is rarely an outright lie. It is the selective truth: a real number presented in a way that implies something the data does not support. The most common version is the annual-average renewable claim dressed as clean power. Matching a year of consumption to certificates and calling the site carbon-free hides every fossil hour the load actually ran, and the hourly reality can be far from the annual headline.

The second version is the market-based-only carbon claim. Report the market-based number, which a PPA or RECs can push toward zero, and stay silent on the location-based number that shows the grid impact, and you have told a true fact that paints a false picture. The GHG Protocol's dual-reporting expectation exists to close exactly this gap.

The third is leaning on offsets to reach a net-zero claim without cutting the actual emissions, which is the carbon equivalent of the certificate game. Buying offsets for emissions you could have reduced, and reporting the netted figure as if the carbon never happened, is the claim regulators and standards are tightening against. The defense against all three is the same: report measured data, report it both ways for carbon, disclose the matching basis for renewables, and let the numbers carry the message instead of the framing. Measured, not claimed, is the line that separates a report from a brochure.

The EU Energy Efficiency Directive reporting

The EU Energy Efficiency Directive, through Article 12, imposes an annual reporting obligation on data centers in the bloc with an installed IT power demand at or above 500 kW. Operators report energy performance and a set of sustainability indicators into a European database, or to a national platform where a member state requires it. Member states may lower the 500 kW threshold but not raise it, so the figure that applies to a given site depends on the country it sits in. Confirm the current threshold and platform for the jurisdiction before you rely on it.

The indicators the rule centers on are PUE, WUE, the energy reuse factor, and the renewable energy factor, which maps directly onto the metric set the rest of this guide describes. The reporting runs on the calendar year, with the deadline for the preceding year having moved to mid-May from 2025 onward after an earlier first cycle. Those dates and the exact data fields have been revised since the directive was published, so treat any specific date as something to verify against the current Commission guidance rather than a fixed fact.

This is a fast-moving area and the safest posture is to hedge hard. The Commission has signaled a dedicated data center energy efficiency package that is expected to tighten thresholds, monitoring, and the link between permitting and efficiency, water, and circularity. The EN 50600 series sits alongside the directive on the standards side. What is stable enough to design around is the direction: mandatory, database-backed disclosure of the metric set against a defined boundary for larger EU sites. The exact numbers, deadlines, and fields are what you confirm against the adopted rule each cycle.

Other jurisdictions and the rising reporting bar

The EU is the furthest along, but it is not alone, and a multi-site operator now faces a patchwork rather than one rule. In the United States there is no single federal data center reporting mandate, but state and local requirements are appearing, often tied to permitting, water connections, or utility interconnection rather than to a clean energy code. Some jurisdictions have paused new data center water connections outright, which functions as a reporting and sourcing requirement by another name.

Broader corporate disclosure rules also reach data centers even when they do not name them. Mandatory climate disclosure regimes, including the EU corporate sustainability reporting requirements and various national and sub-national rules, pull a data center operator's scope 1, 2, and 3 emissions into a filing that carries legal weight. A company that owns its halls reports their footprint; a company that leases reports its share, and the colocation boundary decides which.

The honest summary is that the requirement depends entirely on the jurisdiction, the size of the site, and whether the operator owns or leases. There is no global standard you can report against and be done. Map the rules that apply to each site and each legal entity, confirm them against the current adopted text rather than a summary, and design the metering so the same measured data can feed whichever filing a given jurisdiction demands.

Measured, not estimated: the data behind the report

A sustainability metric is only as good as the data under it, and the dividing line between a defensible report and a guess is whether the numbers were measured or estimated. PUE, WUE, and CUE all rest on metered energy and water over a defined period. Estimate the IT energy from nameplate ratings, estimate the water from pump curves, estimate the carbon from a national average when a regional factor exists, and the report inherits every one of those errors stacked together.

The metering hierarchy is the foundation. You need energy meters at the facility boundary and at the IT measurement point that match the category you are claiming, water meters on the cooling make-up and any other process water, and fuel records for the generators that feed scope 1. The numbers have to reconcile down the chain, so the sub-meters sum to the main within a sensible tolerance, or the boundary is leaking and you cannot see where. A DCIM or power monitoring platform is where this lives day to day, and the PUE guide covers the meter accuracy the ratios depend on.

Annualize it. A metric quoted off the best month of the year is the figure a vendor wants you to see, not the one that matches the resource draw or the bill. The standards expect a continuous measurement period for this reason. When a number has to be estimated because the metering is not there yet, say so in the report and mark it as an estimate. An honestly labeled estimate is defensible. An estimate presented as a measurement is the thing assurance is designed to catch.

Embodied carbon and the scope 3 chunk nobody likes

Embodied carbon is the emissions baked into the building and the equipment before the plant ever runs a workload: the concrete and steel of the construction, the manufacturing of the servers, switches, and storage, and the energy that went into making and shipping all of it. It sits in scope 3, it is the hardest part of the footprint to measure, and it is the part most reports leave out, which is exactly why it deserves attention.

For a data center the IT hardware is the part that bites, because of the refresh cycle. Servers turn over every few years, and each refresh brings a fresh wave of manufacturing emissions that no operational efficiency improvement touches. A hall that drives its PUE and CUE down on the operational side can still carry a large and recurring embodied footprint from the gear it keeps replacing. As the grid decarbonizes and operational carbon falls, the embodied share of the lifetime total rises, so the piece everyone ignored becomes the piece that dominates.

The honest position is that embodied carbon accounting is still maturing and the data is patchier than operational metering. Estimates lean on supplier figures and life-cycle databases that vary in quality. A credible report includes what it can quantify, names the categories it could not, and resists the temptation to drop scope 3 entirely just because it is hard. The SBTi ICT guidance treats purchased goods and services as a scope 3 category that data center operators have to engage with, not skip, given how much capital goes into the hardware.

Water reporting: withdrawal, consumption, and the watershed

Water reporting carries a distinction that energy reporting does not, and getting it wrong overstates or understates the footprint by a lot. Withdrawal is the total water taken from a source. Consumption is the portion that does not return to that source, mostly lost to evaporation in the cooling tower. For an evaporative data center the consumed fraction is the one that matters to the local watershed, because that is the water the basin does not get back.

The source and the watershed context turn a volume into a risk. The same daily draw is a non-issue on a water-rich grid and a permit-stopping problem in a stressed basin, so a credible water disclosure reports not just liters but the source, whether it is potable or reclaimed, and the stress level of the watershed it sits in. GRI 303 is the common framework for water and effluents disclosure, and it pushes reporting toward this watershed-aware picture rather than a bare withdrawal total. The WUE guide covers the cooling-water mechanics and the site-versus-source split in depth.

Be honest about the gaps. Scope 3 water, the water embedded in the electricity and the supply chain, is rarely reported because the data is thin, even where scope 3 carbon is disclosed. And source water, the water used to generate the grid electricity the plant consumes, often exceeds the on-site draw on a thermoelectric grid. A water report that gives only on-site withdrawal, with no consumption figure, no source, and no watershed context, is a number a regulator in a dry region will not accept.

The reporting frameworks and what each one does

Several frameworks govern how the numbers get reported, and they do different jobs rather than competing for the same one. The GHG Protocol is the accounting standard underneath almost everything: it defines the scopes and the market-based and location-based methods for scope 2. If the report counts carbon, it is counting it the GHG Protocol way whether or not it says so.

On top of that sit the disclosure and target frameworks. GRI provides sustainability reporting standards, including GRI 303 for water, used widely for stakeholder disclosure. CDP runs a questionnaire-based disclosure system on climate, water, and forests that many large customers and investors ask suppliers to complete. The Science Based Targets initiative, a collaboration of CDP, the UN Global Compact, the World Resources Institute, and WWF, validates emissions-reduction targets against a net-zero pathway and publishes sector guidance, including ICT guidelines that cover data center, fixed, and mobile network operators across all three scopes.

Do not over-claim what any of these gives you. They are accounting and disclosure structures, not a guarantee of a clean footprint, and they are revised on their own cycles, so confirm the current version before citing a method. A report that names its framework, applies it consistently, and reports the metric set against a stated boundary is doing the job. A report that drops a logo on the cover without applying the method underneath is doing the opposite.

Third-party assurance and the credibility of the report

Assurance is an independent check on whether the reported numbers are right, and it is what separates a report you can stand behind from one you hope nobody audits. It comes in two strengths. Limited assurance, the more common and cheaper level, gives a negative conclusion: the assurer found nothing that suggests the numbers are materially wrong. Reasonable assurance is the higher bar, closer to a financial audit, giving a positive opinion that the numbers are fairly stated. The two are not the same level of confidence, so a report should say which it carries.

What an assurer checks is exactly the discipline the rest of this guide describes: that the boundary is defined and consistent, that the metrics rest on metered data with a clear trail back to the instruments, that the carbon is reported across scopes with the methods named, and that estimates are labeled as estimates. A report built on loose boundaries and unlabeled estimates does not survive even limited assurance.

Be honest about where assurance is and is not required. Mandatory climate disclosure regimes increasingly require assurance on carbon, often starting at limited and ratcheting toward reasonable over time. Water assurance lags: CDP, for instance, does not require water data to be assured to earn a top score, though it encourages it. So the assurance picture depends on the framework and the jurisdiction. Design the data trail to survive reasonable assurance even where only limited is required today, because the bar is moving up, not down.

Targets, net-zero pledges, and the AI-growth tension

A target is a promise about the future, and the gap between the pledge and the measured trajectory is one of the most scrutinized parts of a sustainability report. The common targets in this sector are a net-zero or carbon-neutral date, a 24/7 carbon-free energy percentage, and a PUE or WUE goal for new builds. The SBTi Corporate Net-Zero Standard is the structure most credible targets are validated against, and it has been revised, so confirm the current version before quoting a company against it.

The honest distinction is between a target with a measured path and a pledge with a press release. A net-zero date backed by year-on-year reductions in measured scope 1, 2, and 3 emissions is a target. A net-zero claim that leans on offsets and annual certificate matching to net the figure to zero is a pledge wearing a target's clothes. The difference shows up in whether the location-based emissions are actually falling, not just the market-based headline.

And the AI buildout has put real tension on every one of these targets. Several operators that had been reducing emissions have seen their absolute footprint climb as they build out AI capacity, even while their per-unit metrics improve. A falling CUE on a rising total energy base can still mean rising total carbon. The credible report does not hide that tension behind an efficiency metric. It shows the absolute numbers next to the intensity numbers, and it is clear about whether the target is on track against the measured data or slipping behind the growth.

What to document

A sustainability number with no method behind it is a number an assurer rejects and a successor cannot reproduce. The record is what lets a reviewer trace each metric back to the meters, see which boundary was drawn, and understand how the renewable energy was counted. It is also what turns a once-a-year scramble into a repeatable filing.

Capture the boundary and the colo split, the measurement period, the metered IT and facility energy with their meter points, the water withdrawal and consumption with the source and watershed, the carbon by scope reported both market-based and location-based with the emission factors and their sources, the renewable instruments and their matching basis, and which figures are measured versus estimated. A field tool such as FieldOS is where the metered readings, the metric calculations, the boundary definition, and the assumptions can live together with a trail, so the same data feeds the EU database, a CDP questionnaire, or an assurance review without being rebuilt each time. The table frames the core metrics and the note each one carries.

Common mistakes

• Reporting a single PUE as the whole sustainability story and going quiet on water, carbon, and reuse.
• Quoting a market-based carbon figure alone, with no location-based companion that shows the real grid impact.
• Presenting an annual-average renewable match as fully clean power, ignoring the fossil hours the load actually ran.
• Leaving the reporting boundary undefined, so the colo split and the IT meter point are open to interpretation.
• Using estimated data where metering exists, or presenting an estimate as a measurement.
• Dropping scope 3 and embodied carbon because it is hard, when the hardware refresh makes it a large share.
• Leaning on offsets to reach a net-zero claim instead of cutting the measured emissions.
• Reporting a falling intensity metric while the absolute footprint climbs with AI growth, and not showing both.

Field checklist

Standards and references

The GHG Protocol is the carbon accounting standard underneath the report. It defines scope 1, 2, and 3, and the market-based and location-based methods for scope 2, with the dual-reporting expectation that both be disclosed. The Green Grid originated the metric family, PUE, WUE, CUE, and the energy reuse metrics, and the ISO/IEC 30134 series formalizes several of them, with PUE in ISO/IEC 30134-2 and the European EN 50600 series carrying the same metrics. CUE is defined as an operational metric, carbon emission factor times PUE, excluding embodied carbon. Confirm the current part and edition before citing a method on a report.

On the regulatory side, the EU Energy Efficiency Directive Article 12 mandates annual reporting for EU data centers at or above 500 kW installed IT power, into a European database, on PUE, WUE, ERF, and the renewable energy factor, with member states able to lower the threshold. The thresholds, deadlines, and fields are being revised and a further Commission package is expected, so verify them against the current adopted rule and the jurisdiction rather than a summary. Other jurisdictions impose their own disclosure and permitting requirements, which control where they apply.

For disclosure and targets, GRI provides the sustainability reporting standards including GRI 303 for water, CDP runs the climate and water disclosure questionnaires, and the Science Based Targets initiative validates targets and publishes ICT-sector guidance covering data centers across all three scopes. Assurance, limited or reasonable, is required by some frameworks and jurisdictions and encouraged by others. All of these revise on their own cycles, so hedge every metric definition, every reporting rule, and every renewable claim to the framework, the jurisdiction, and the boundary it was measured against. Measure the full metric set rather than just PUE, report carbon honestly in both market-based and location-based terms across the scopes, and define the boundary so the report cannot be read two ways.

Units, terms, and conversions

The metric set carries different units, and the same concept shows up under several names across an energy model, a carbon filing, and a water permit. The terms below travel across the whole reporting scope.

PUE is a dimensionless ratio. WUE is liters of water per kilowatt-hour of IT energy, sometimes reported in gallons, so confirm the unit before comparing. CUE is kilograms of CO2 equivalent per kilowatt-hour of IT energy. ERF and REF are fractions from 0.0 to 1.0, sometimes given as percentages. Carbon totals are in tonnes of CO2 equivalent, split by scope. Energy is in kilowatt-hours or megawatt-hours over the measurement period.

PUE

Power usage effectiveness, total facility energy over IT energy; the energy metric

WUE

Water usage effectiveness, water used per unit of IT energy, in liters per kWh

CUE

Carbon usage effectiveness, carbon per unit of IT energy, in kg CO2e per kWh; carbon factor times PUE

ERF

Energy reuse factor, the share of energy sent off-site for reuse, 0.0 to 1.0

Scope 1 / 2 / 3

Direct emissions, purchased-electricity emissions, and value-chain emissions under the GHG Protocol

Market-based vs location-based

Carbon valued at contracted power versus the average emissions of the physical grid; report both

REC / PPA

Renewable energy certificate (an attribute) and power purchase agreement (a contract for specific output)

24/7 CFE

Carbon-free energy matched to consumption every hour, not averaged annually

Reporting boundary

The defined line around what counts as facility, IT, and another party's responsibility

Embodied carbon

Emissions in construction and hardware manufacture, in scope 3, separate from operational energy

Related tools

Calculators and readiness checks for this work

Compare your options

FAQ

People also ask

• What is WUE?
• What is PUE?
• Cast-in vs post-installed anchors: which should I use?
• What is the difference between cold and hot commissioning of a BESS?
• What is a CDU in a data center?