Datacentres

A datacentre combines IT equipment, internal networking, technical infrastructure, cooling, electricity supply and geography. An assessment may cover the whole site or an explicitly allocated part, using equipment, component or system models.

The intended boundary includes embodied inventories and operational resources. Current component simulators do not establish a complete building or site LCA; omitted infrastructure must remain explicit.

Characteristics

Characteristic	Unit	Treatment	Purpose
Model	Approach	Equipment, components or system factor according to evidence	Calculation structure
PUE	Ratio	Site-period value, otherwise disclosed assumption	Total site energy relative to IT energy
REF	%	Unknown if undocumented	Renewable-energy information; not an automatic impact discount
Electricity location	Country / grid region	Actual site location	Electricity-factor selection
Period	Day, month, year	Observed activity period	Consistent energy and factors
Site configuration	Inventory	Documented archetype if unavailable	Servers, storage, network and support equipment

Modelling principle

A datacentre includes computing, storage, networking, cooling, security and power infrastructure.¹

Available evidence	Model representation
Reliable physical inventory	Equipment approach: aggregate known equipment.
Known component quantities, unknown devices	Component approach: aggregate documented unit resources.
Compatible aggregate factor only	System approach: factor matching the assessed boundary.

Embodied impacts

Calculation

An equipment approach sums non-overlapping inventories following Equipment modelling.

System factors describe a defined boundary, often with specific geography, duration and functional units. Some cover storage more readily than computation; none should be extended to another function without justification.

A component approach combines CPU, GPU, memory, storage, network and structural contributions. Building, cooling equipment and electrical infrastructure require separate inventories when not already included. PUE does not account for their manufacturing.²

When data are missing

Calculation detail depends on the available inventory. Component quantities support a partial calculation even without server model references. They do not reveal the remaining equipment or site energy efficiency.

Situation	Fictional example
Starting information	The provider reports 8 CPUs and 512 GB RAM. Storage, networking and PUE are undocumented.
Selected reference	Compatible factors are available for these CPUs and memory. No reference supports completing the other contributions.
Calculation inputs	The calculation covers known components. Storage, networking and structures remain unassessed; PUE remains unknown. The result is a subtotal, not a complete site footprint.

A reliable equipment inventory takes precedence. If component reconstruction is too uncertain, a compatible aggregate service factor can be used for the assessed boundary and unit. It replaces the calculation of that same boundary rather than being added to it. Unknown PUE, WUE and REF are not replaced with zero.³

End of life

Coverage depends on each equipment or system factor. This iteration has no separate complete site end-of-life model; that gap remains explicit where relevant.

Operational impacts

PUE boundary

Power Usage Effectiveness is:

$$PUE=\frac{E_{datacentre}}{E_{IT}}$$

Both energies refer to the same site boundary and period. IT includes all relevant computing, storage and networking equipment, not servers alone.

When the input excludes site overhead:

$$E_{site}=E_{IT}\times PUE$$

A total site-meter reading or a factor already including overhead requires no further PUE multiplier.

Total data centre electricity

Example with PUE of 1.21,200 kWh

1,000

200

IT equipmentSite auxiliaries

1,000 kWh IT × 1.2 = 1,200 kWh in total

PUE applies to IT electricity. It does not multiply manufacturing impacts.

Component electricity

$$E_{IT}=E_{components}+E_{network}$$

For a simplified component model:

$$\begin{aligned} E_{components}&=D_{usage}(P_{cpu}+P_{gpu}+P_{ram}+P_{storage})\\ P_{cpu}&=N_{vcpu}\left(P_{cpu,min}+\tfrac12(P_{cpu,max}-P_{cpu,min})\right)\\ P_{gpu}&=N_{gpu}\left(P_{gpu,min}+\tfrac12(P_{gpu,max}-P_{gpu,min})\right)\\ P_{ram}&=N_{ram}\times0.000392\\ P_{storage}&=\begin{cases}1.2\times10^{-3}V_{storage}&\text{SSD}\\0.65\times10^{-3}V_{storage}&\text{HDD}\end{cases} \end{aligned}$$

Power in kW multiplied by duration in hours gives kWh. $N_{ram}$ is decimal GB and $V_{storage}$ decimal TB. CPU coefficients must represent the physical share per vCPU. The midpoint CPU/GPU equations assume 50% load and are not measured profiles. Historical average power coefficients do not represent every device.

Missing power supplies, fans and boards complete partial estimates; a complete AC measurement already includes them. Electricity conversion relies on compatible electricity impact factors.

Site parameters and renewable-energy claims

PUE, electricity geography, load, resource sharing and supplier reporting periods affect interpretation. WUE describes a separate water quantity. REF must be interpreted within its documented supply boundary.

The reference result uses the site's average consumption electricity mix. REF does not automatically reduce it. The historical interpolation below is only a separate sensitivity scenario, requiring complementary supply shares and compatible factors:⁴

$$I_{operational_{i,p}}=E_{elec_p}\left(REF\times RIF_{i,p}+(1-REF)\times IF_{i,p}\right)$$

$E_{elec_p}$ is kWh in location $p$; factors are in $U_i/\text{kWh}$. REF is a fraction from zero to one: 80% means 0.8. Unknown remains unknown.

A national mix already contains renewables. Interpolating it with a renewable-only mix does not establish a verified physical supply mix. This sensitivity is not an SCI calculation. Purchase certificates do not justify zero factors or automatic reductions across all indicators.

Direct water and electricity-related water

For site WUE in litres per IT kWh:

$$V_{direct}\,[L]=WUE\,[L/kWh_{IT}]\times E_{IT}\,[kWh]$$

Direct water uses IT energy without another PUE multiplier. Electricity-related water depends on total site energy. Aggregation into WU requires characterisation compatible with the method and geography.

Calculating data centre water use

1,000 kWh IT

× WUE : 0.5 L/kWh IT↓

500 LDirect site water

× PUE : 1.2↓

1,200 kWhSite electricity

↓× Electricity water factor↓Indirect water

Teaching example. Litres are added only for compatible flows; a scarcity indicator requires separate characterisation.

Allocation and annualisation

Shared site impacts can be allocated by reserved capacity, consumed resources or a suitable functional quantity. These choices require explicit treatment of idle capacity and supporting resources.

Annualisation uses equipment lifetimes and the assessed period. See Allocation and sharing.

Assumptions and limitations

Working assumptions

A representative observed PUE is preferable to an undocumented generic value. Component reconstruction is useful only with defensible inputs. A datacentre is broader than a single server or application.

Known limitations

PUE, WUE and REF may be disclosed at different geographic and temporal scales. Review supplier statements against their actual boundaries. A detailed component model cannot compensate for missing building or technical-infrastructure inventories.

Improvements

Expand site data and archetypes and reconcile equipment, component and system approaches with consistent boundaries.

Interactive example

This exploratory simulator models a physical server configuration and electricity overhead through PUE and the usage-country mix. It does not calculate a whole datacentre inventory or automatically allocate a VM share.

Example inputs

Assessment period (years)Site PUEResource share reserved for the service (%)Hosting countryFranceGermanyUnited StatesUnited KingdomChinaIndiaPolandSwedenNorwayJapanBrazilAustraliaTaiwanSouth KoreaIreland

Equipment typeServersSwitchesCount

Equipment factors

Manufacture (kg CO₂e/device)Lifetime (years)Energy (kWh/device/year)

Equipment typeServersSwitchesCount

Equipment factors

Manufacture (kg CO₂e/device)Lifetime (years)Energy (kWh/device/year)

Adjust assumptions

Direct water consumption per IT kWh (L/kWh IT)

Calculated subtotal2,669 kg CO₂e

2,669 kg CO₂e / year

Not assessed: Direct water not provided ; Building, cooling and backup: manufacture excluded.

Result and activity breakdown

Contribution	kg CO₂e
Allocated manufacture	580
Use-phase electricity · 65,440 kWh	2,089

• Equipment energy : 36,350 kWh

Prefilled activity and hosting values are example assumptions. The result covers climate impact; unassessed contributions remain identified.

Simulator factors and conventions

Electricity: D4B 2025 national averages. Use country does not change manufacture. Network factors remain unknown until supplied; they must cover the full path, manufacture and operation included. In this illustration, the distance from the D4B per-GB·km model is incorporated upstream into the per-GB route factor.

Annual energy per device, all power states included. For hosting, enter IT energy before PUE. Prefilled inventory profiles are assumptions from this page’s worked example.

Applications

Hosting infrastructure, cloud, and generative AI.

Footnotes

1. Versioned data/cloud/*, data/factor/*, data/energy/* and identified site inventories; publications including Green Cloud Computing and the OVHcloud environmental-impact tracker methodology. Digital4Better data catalogue, snapshot 77cae4a. References and scope: Sources and corpus. ↩

2. The Green Grid PUE definition. The Green Grid PUE definition. ↩

3. The Green Grid WUE paper. The Green Grid WUE paper. ↩

4. SCI specification for electricity and allocation boundaries. SCI specification. Online documentation, unpinned version. ↩