Scope 3 Emissions Accounting for Data Center Supply Chains

Scope 3 emissions accounting for data center supply chains requires tracing physical infrastructure alongside purchases. Supply-chain emissions make up a substantial share of an operator’s total footprint. To measure emissions well, operators need to fill gaps in what suppliers disclose, show how they make estimates, and link embodied-carbon calculations to the records used to design and build facilities.

Why Scope 3 dominates the data center carbon story

A data center operator can cut its power usage effectiveness, or PUE, buy renewable energy, and still have a growing carbon footprint. Microsoft’s 2025 Environmental Sustainability Report showed that total greenhouse gas emissions were up roughly a quarter from its 2020 baseline, as operational emissions fell and supply-chain emissions rose. Scope 3 represented the vast majority of its emissions.

Google also reduced operational emissions year-over-year while supply-chain emissions grew substantially, driven by AI infrastructure buildout and Asia-Pacific supply chains dependent on fossil-heavy grids. These gains in operations remain important, but they cannot offset upstream and embodied emissions at the pace the buildout is generating them. The accounting boundary has to extend beyond the facility. The GHG Protocol groups these upstream sources into categories, with purchased goods and services, capital goods, and fuel- and energy-related activities dominating the data center supply-chain footprint.

Where the emissions actually come from — IT hardware, construction materials, and energy upstream

IT equipment manufacturing is the largest embodied-carbon source in AI data centers. The footprint starts well before equipment reaches the site. Semiconductor fabrication uses a lot of energy and high-purity process gases, which include potent greenhouse gases such as CF₄ and SF₆. Manufacturing-phase emissions typically exceed use-phase emissions for individual chips, so a substantial part of an AI build’s carbon cost is incurred before a single server is powered on.

Steel, concrete, and aluminum used in construction are the next large source of upstream emissions, and the rush to add new AI data center capacity has sharply increased how much of them is used. Category 3 then captures fuel- and energy-related activities outside Scope 2, including extraction, processing, and transmission losses associated with the electricity a facility consumes. This reaches upstream of the electricity bill. It includes grid-mix carbon that market-based Scope 2 accounting may obscure.

The balance between these sources depends on the facility and its grid. For facilities on average-carbon grids, Scope 2 operational emissions make up most of the impact over a facility’s lifetime. When a facility uses very low-carbon electricity, embodied Scope 3 can make up roughly 40% of its total lifetime greenhouse gas emissions. Cleaner power changes the balance while leaving the manufacturing history intact. These lifetime comparisons are distinct from an operator’s reported inventory, so they do not contradict the large Scope 3 shares reported during infrastructure expansion.

Why supplier disclosure gaps make the largest categories the hardest to measure

The largest categories are also the hardest to see. Operators depend on OEMs, semiconductor manufacturers, and construction suppliers for the primary emissions data needed to measure IT hardware and capital goods. Procurement records establish what was purchased, but they do not by themselves establish the emissions released to make it. That information depends on what suppliers choose, or are required, to disclose. An operator can control the purchase without controlling the evidence behind its carbon estimate.

Supplier-specific emissions data is the most accurate approach, but access is uneven. Hyperscalers can use procurement leverage to obtain information that is not yet standard across the industry. Smaller buyers need to account for the same emissions but may lack that leverage. The gap can also pass downstream. Organizations relying on cloud providers face a similar dependence on the information those providers make available, rather than an independent view of the underlying supply chain.

Equipment specifications in manufacturer PDFs are of little use if someone has to retype them by hand for engineering models or sustainability calculations. Specification extraction should feed models and validation directly, while preserving the distinction between equipment parameters and actual emissions disclosures. When suppliers do not provide primary data for the biggest categories, operators must turn to estimates, and the choice of estimate introduces another source of uncertainty.

How methodology choices compound uncertainty when primary data is absent

Spend-based accounting multiplies what an operator spends on purchases by an industry-average emissions factor. It is simple to implement and widely used for purchased goods and capital goods. Its weakness is the level of averaging. Factors derived from national sector-wide averages do not capture the material intensity of specific equipment types, so the result is not an equipment-level measurement. A precise-looking total can therefore conceal a broad assumption about what was bought.

The size of that error is itself unclear. The notes list no published work that compares equipment-level life-cycle assessment results with corporate greenhouse gas accounting approaches in a systematic way. That leaves the gap between the two unquantified. Spend-based data can fill an inventory, but it cannot establish how closely that inventory reflects the hardware in a particular facility. Without a visible method behind the figure, readers cannot distinguish a supplier measurement from a sector average.

Activity-based accounting moves closer to the physical source. It multiplies quantities such as tonnes of steel or kilowatt-hours of electricity at a supplier site by emissions factors. This is more accurate than spend-based accounting when the necessary data is available, but it still depends on suppliers sharing operational information they may be unable or unwilling to provide. Supplier-specific emissions figures offer the strongest basis. Their limited availability, however, means an operator’s inventory may combine primary data, physical estimates, and spending proxies.

That mixture is not automatically a failure. Presenting unlike evidence as though it had the same strength, then rebuilding the calculation manually each reporting cycle, creates problems. Better physical data can support more useful comparisons than spending alone, while a traceable method shows where uncertainty remains. The GHG Protocol revision proposes making the quality of those inputs visible in the reported inventory.

What the GHG Protocol revision requires operators to change before 2027

The GHG Protocol’s Scope 3 Standard Revisions Phase 1 Progress Update, published in March 2026, describes proposed changes. Its text remains subject to revision and approval. That distinction is important for planning before 2027. Operators face a clear direction toward more traceable, disaggregated, primary-data-anchored accounting, but the update should not be treated as a settled set of requirements already in force.

Under the proposed data-quality disaggregation, companies would have to split each category’s emissions by the type of data used, including primary supplier data, activity-based secondary data, and spend-based estimates. The method would become part of the disclosure. An inventory built around a single total would therefore need records that explain how its components were calculated. A result could no longer stand apart from the quality of its inputs.

The draft would also allow corporate-level supplier allocation only for homogeneous suppliers. Corporate-level allocation means taking a supplier’s company-wide emissions figure and assigning a share to the buyer. That can mislead when the supplier runs both a high-intensity semiconductor fabrication plant and a lower-intensity logistics operation. The average misrepresents both activities. For operators buying from diversified manufacturers, facility- or product-level data would be needed instead of an allocation from a blended corporate total.

A proposed Primary Data Share metric would make the proportion of the inventory based on supplier-specific data publicly visible, with annual improvement targets. Data quality would become a performance indicator. Meanwhile, the EU CSRD Omnibus revision narrowed reporting scope and delayed timetables, California SB 253 moved Scope 3 disclosures to 2027, and the SEC climate rule was dropped. Those changes do not establish that value-chain accounting is no longer required wherever other obligations apply. Building records that can support changing disclosure rules allows operators to adapt across reporting timetables.

How the embodied-carbon complexity of physical infrastructure sits outside most current accounting programs

Construction-phase embodied carbon combines the problems described above in a single capital event. Steel, concrete, aluminum, and mechanical and electrical systems have already generated emissions before the facility opens. Operators buy energy year after year, while construction is a one-time event. Most Scope 3 programs are not yet equipped to handle that combination of scale, uneven supplier disclosure, and measurement uncertainty. Reports made each year must account for choices made on the separate schedule used to deliver a project.

Two suppliers can provide structurally identical steel with footprints that differ by an order of magnitude, depending on whether production uses an electric-arc or blast furnace and which grid supplies it. The physical performance can match while emissions diverge. This makes a generic steel factor a poor substitute for production-specific evidence. It also explains why the proposed restriction on corporate-level allocation reaches directly into construction procurement.

For diversified material suppliers, facility- or product-level Environmental Product Declarations, or EPDs, would provide the evidence needed. Availability varies sharply by material and supplier geography. Accelerated AI construction timelines make the gap harder to close because sourcing decisions can happen faster than EPD data can be gathered. Retroactive collection is often impossible. By the time a sustainability team assembles its report, the project may already have passed the point when the relevant information was easiest to request and connect.

The useful records belong in the project itself. Equipment parameters, material quantities, and supplier declarations exist, or should exist, during design and procurement, when teams specify systems and select materials. Capturing those records as structured, connected project data makes embodied-carbon accounting a use of the project record and avoids a separate reconstruction. It preserves the link between what was specified, what was sourced, and the evidence available to calculate its footprint.

What a credible Scope 3 measurement program for a data center supply chain actually requires

A credible program needs defined sources, structured collection, traceable methods, and ownership beyond the sustainability team’s inbox. Software alone cannot provide those foundations. The first commitment is supplier engagement focused on the OEMs, semiconductor manufacturers, and major construction suppliers responsible for the largest share of exposure. Contractual disclosure mandates, as Microsoft has implemented, provide a durable mechanism for major OEM relationships, while industry-average fallbacks remain necessary for the long tail.

Methodology transparency belongs in the data architecture from the start. Each category and reporting period needs to retain the distinction between primary data, activity-based secondary data, and spend-based estimates. Otherwise, a future disaggregated disclosure will require reconstructing the inventory. Primary Data Share gives operators a target they can work toward before any mandate takes effect. It lets a program track whether its evidence is getting stronger independently of whether total emissions rise or fall.

Capital projects require the same discipline at specification and procurement. Equipment specifications, material quantities, and EPDs need to be connected to the project records from which embodied-carbon calculations draw. When equipment parameters defining power loads, cooling requirements, and cable routing also anchor embodied-carbon inputs, the report no longer depends on a separate manual search for the same equipment information. Connected records and supplier evidence are preconditions for defensible capital-goods accounting.

IT-driven data architecture makes that evidence reusable across reporting cycles. A better spreadsheet does not solve missing disclosures, hidden estimation methods, or disconnected project records. Operators that build the infrastructure now will be positioned to report against the revised standard with a measurable Primary Data Share as requirements take shape. Those that do not risk reconstructing both their inventory and their methodology under audit pressure.