Carbon accounting for data centres

Customers, investors and regulators increasingly ask data centre operators a simple question: how much greenhouse gas does your facility emit, and what are you doing about it? Answering it credibly requires a carbon inventory built according to recognized rules, which define what is counted, how electricity emissions are calculated, and how clean energy purchases may be reflected. The rules are more subtle than they first appear, and the same facility can report very different numbers depending on the method. This lesson explains the Greenhouse Gas Protocol's scopes, the sources of emissions in a data centre, the two methods for Scope 2 electricity emissions, emission factors, Scope 3 including embodied carbon, and the principles that make carbon accounting credible.

The Greenhouse Gas Protocol and its scopes

The Greenhouse Gas Protocol, developed by the World Resources Institute and the World Business Council for Sustainable Development, is the most widely used standard for corporate emissions accounting, and it underlies most regulatory reporting schemes. It divides an organization's emissions into three scopes:

• Scope 1: direct emissions from sources the organization owns or controls, such as fuel burned in generators and boilers, and leaks of refrigerants and fire suppression gases.
• Scope 2: indirect emissions from the generation of purchased electricity, heat, steam and cooling consumed by the organization.
• Scope 3: all other indirect emissions in the value chain, upstream (such as purchased goods, capital goods including construction and IT hardware, transport, and the upstream emissions of fuels) and downstream (such as the use and end-of-life of products sold).

Emissions are expressed in tonnes of carbon dioxide equivalent (CO₂e), converting other gases using their global warming potentials (GWP). The Protocol is being revised, with updates to the Scope 2 guidance under discussion that could tighten the rules for clean energy claims, including moves toward hourly matching (Lesson 8).

Where data centre emissions come from

For most data centres, Scope 2 electricity emissions dominate the operational inventory. Scope 1 is usually small but not negligible: diesel burned during generator testing and outages, and refrigerant leaks, which can be surprisingly significant because many refrigerants have GWPs over a thousand. Scope 3 includes the embodied emissions of the building and the IT hardware, which become relatively more important as operational emissions fall.

For a colocation provider, the electricity used by customers' IT equipment is often counted in the provider's Scope 2 (because the provider buys the electricity), and in the customers' Scope 3 (or Scope 2, depending on contractual arrangements and how the customer applies the Protocol). Clear agreements on how emissions and clean energy attributes are allocated between provider and customers avoid double counting.

# Annual greenhouse gas inventory for a 20 MW (IT) facility, in t CO2e
facility_MWh = 20 * 8760 * 0.8 * 1.3                     # average 80 % load, PUE 1.3

# Scope 1: generator testing and refrigerant leakage
diesel_L = 60000                                          # monthly tests and annual load bank runs
scope1_diesel = diesel_L * 2.68 / 1000
leak_kg, gwp = 150, 1430                                  # R-134a, 100-year GWP (IPCC AR4 value, widely used)
scope1_refrig = leak_kg * gwp / 1000
scope1 = scope1_diesel + scope1_refrig

# Scope 2: location-based uses the grid average; market-based uses contracts and the residual mix
grid_g = 380                                              # grid average, g CO2e per kWh
residual_g = 450                                          # residual mix after others claim their clean energy
contracted_MWh = 0.70 * facility_MWh                      # matched by a qualifying renewable PPA
scope2_loc = facility_MWh * grid_g / 1000
scope2_mkt = (facility_MWh - contracted_MWh) * residual_g / 1000

# Scope 3 (partial): embodied emissions of servers, amortized over their life
servers, kg_per_server, life_years = 9000, 1300, 5
scope3_it = servers * kg_per_server / life_years / 1000

print(f"Facility energy: {facility_MWh:,.0f} MWh")
print(f"Scope 1: diesel {scope1_diesel:,.0f} t + refrigerant {scope1_refrig:,.0f} t = {scope1:,.0f} t")
print(f"Scope 2 location-based: {scope2_loc:,.0f} t")
print(f"Scope 2 market-based (70 % PPA, residual mix for the rest): {scope2_mkt:,.0f} t")
print(f"Scope 3, server embodied carbon per year: {scope3_it:,.0f} t")

Facility energy: 182,208 MWh
Scope 1: diesel 161 t + refrigerant 214 t = 375 t
Scope 2 location-based: 69,239 t
Scope 2 market-based (70 % PPA, residual mix for the rest): 24,598 t
Scope 3, server embodied carbon per year: 2,340 t

In this inventory, Scope 1 totals a few hundred tonnes, more than half of it from refrigerant leakage. Location-based Scope 2 is about 69,000 t, two hundred times larger. A renewable power purchase agreement covering 70 % of consumption reduces market-based Scope 2 to about 25,000 t. The embodied carbon of servers, spread over their five-year life, adds over 2,000 t a year, small compared with location-based Scope 2 on this grid but significant compared with market-based emissions or with Scope 2 on a clean grid.

Location-based and market-based Scope 2

The Protocol's Scope 2 Guidance requires organizations to report electricity emissions in two ways:

• Location-based: consumption multiplied by the average emission factor of the grid where the electricity is used. It reflects the physical emissions of the local grid and cannot be changed by purchasing decisions, only by using less electricity or by the grid becoming cleaner.
• Market-based: consumption multiplied by emission factors derived from the contractual instruments the organization holds, such as power purchase agreements (PPAs), utility green tariffs and energy attribute certificates (renewable energy certificates, RECs, in North America). Consumption not covered by instruments uses the residual mix factor: the grid average with the claimed clean energy removed, which is higher than the grid average.

Market-based accounting lets organizations reflect clean energy purchases, which has driven large investments in renewable energy. But it has been criticized because instruments can be bought from distant regions or different times than the consumption they cover, so that a facility running on a coal-heavy grid at night can claim zero emissions with certificates from solar farms elsewhere. Quality criteria in the Scope 2 Guidance and in schemes such as RE100 aim to limit this, and hourly matching (Lesson 8) addresses it more directly.

Emission factors

Location-based factors come from official sources for each grid region, and they differ enormously between regions, from a few grams per kilowatt-hour in hydro and nuclear regions to several hundred in fossil-dominated ones.

In Canada, grid emission factors are published by province in Environment and Climate Change Canada's National Inventory Report. They range from near zero in Quebec, Manitoba and British Columbia (largely hydroelectric), through low values in Ontario (nuclear and hydro, with some natural gas), to several hundred grams per kilowatt-hour in Alberta, Saskatchewan and Nova Scotia, where natural gas and coal remain significant.

Average factors describe the typical emissions of the grid; marginal factors describe the emissions of the generator that responds to a change in load, which is often a fossil plant even on a fairly clean grid. Marginal factors are useful for assessing the effect of new loads or load shifting (Lesson 8), but inventories use average factors.

Scope 3 and embodied carbon

Scope 3 is the hardest category to measure, because it relies on data from suppliers and on estimates. For data centres, the most important Scope 3 categories are usually capital goods (the embodied carbon of construction materials, mechanical and electrical equipment, and IT hardware) and purchased goods and services. Embodied carbon is estimated from life cycle assessments of products, environmental product declarations (EPDs) for construction materials, and supplier data, often supplemented by spend-based estimates. As data centres decarbonize their electricity, embodied carbon becomes a larger share of their footprint, which is the subject of Lesson 10.

Principles of credible accounting

The Protocol sets out five principles: relevance (the inventory reflects the organization's real emissions and serves decision-makers), completeness (all sources within the boundary are included, and exclusions disclosed), consistency (methods are consistent over time, so trends are meaningful), transparency (assumptions, methods and data sources are disclosed) and accuracy (systematic over- or under-statement is avoided and uncertainty is reduced). In practice, credible data centre reporting discloses both Scope 2 methods, the instruments used and their characteristics (location, vintage, type), the refrigerants and leakage, the boundary between provider and customers, and the basis for Scope 3 estimates. Third-party verification is increasingly expected and, under some regulations, required (Lesson 11).

Setting the organizational boundary

Before calculating anything, an organization must decide which operations its inventory covers. The Protocol offers three approaches: equity share (emissions in proportion to ownership), financial control and operational control. Data centre companies typically use operational control, including facilities they operate whether owned or leased. Leased facilities, joint ventures and sites operated by third parties need careful treatment so that emissions are neither omitted nor counted twice. The boundary, the base year against which progress is measured, and any recalculations of the base year after acquisitions or methodology changes must be documented and disclosed.

Data quality

The accuracy of an inventory depends on its activity data (metered electricity, fuel delivered, refrigerant purchased) and its emission factors. Metered data are best; estimates based on design capacity or spending carry much greater uncertainty. Refrigerant leaks are often estimated from the quantity of refrigerant added during maintenance, which requires good records from service contractors. A data quality assessment that rates each source and identifies improvements is a standard part of mature carbon accounting.

Carbon accounting is not an end in itself. Its purpose is to show where emissions arise, so that the measures in the rest of this course can be targeted where they matter most, and to track whether those measures are working.