5. Inputs and Evidence Sources

The methodology draws on the following categories of inputs.

5.1 Cloud Usage Data

To produce a metric for a unit of cloud usage, the methodology requires the usage attributes that identify it, being the cloud provider, the service or SKU, the region, the usage period, and the quantity or duration of use. These attributes are read from the client's cloud usage data, most commonly a cloud usage report. No cost or spend information is required.

5.2 Hardware Power Consumption Data

Server power consumption constants are aligned with methodologies used by bodies such as the US Environmental Protection Agency, the EU Ecodesign Regulation, and ISO/IEC 21836:2020, including benchmarks from the Standard Performance Evaluation Corporation. Power consumption data for memory components is adapted from memory module manufacturer documentation. Storage power data is adapted from drive manufacturer datasheets, with the Berkeley National Laboratory's Data Center Energy Usage Report held as context. Networking power data is derived from Intel network interface card specifications and a published study on the electricity intensity of fibre optic cables.

5.3 Cloud Provider Published Data

Power Usage Effectiveness values are sourced from each cloud provider's own published sustainability and efficiency data, being the AWS Cloud Sustainability portal, Google Cloud's data centre efficiency reports, and Microsoft Azure's documented measurements of data centre water and energy use.

5.4 Grid Carbon Intensity Data

Greenpixie uses an internal tool called GPX Grid to produce hourly, region-specific Grid Carbon Intensity factors. GPX Grid sources data from a tiered hierarchy of national grid operator APIs, modelled national datasets, and national inventory figures, with international averages from sources such as the IPCC to account for upstream and lifecycle emission factors of the technologies and fuel types. Methodology and data selection are validated in collaboration with academic partners including Professor Alex Rogers at the University of Oxford.

5.5 Embodied Emissions Data

Component-level embodied emissions data is derived from several sources. The silicon die coefficient by process node comes from the ACT semiconductor model of Gupta et al. (ISCA 2022). The CPU package base, the RAM coefficients and the assembly figure come from lifecycle assessment studies by Gröger, J. et al. at the German Federal Environment Agency, via an open life cycle inventory dataset. The GPU memory coefficient, board base and cooling solution term come from NVIDIA's HGX H100 and HGX B200 product carbon footprints. The fixed figures for power supplies, chassis, fans, motherboard and expansion cards come from Dell's critically reviewed life cycle assessments of its PowerEdge R740, R6515, R7515, R6525 and R7525 servers. NAND die densities by generation come from third-party semiconductor die analyses and published conference papers. Transport and end-of-life shares come from Dell, Seagate's product sustainability reports, Falk et al. 2025 for GPUs and Cisco's switch life cycle assessment for networking equipment. CPU die area data, including the IO die of chiplet processors, and CPU process nodes are sourced from third-party hardware specification databases. GPU die areas, process nodes and memory capacities come from manufacturer specification sheets and product pages, cited per GPU, and GPU thermal design power from manufacturer datasheets.

5.6 Water Consumption Data

Cloud provider Water Usage Effectiveness values are sourced from published sustainability reports by Google Cloud, Microsoft Azure, and AWS. Regional water intensity factors for electricity production are sourced from the World Resources Institute.

5.7 Cloud Infrastructure Research

Greenpixie researches cloud provider infrastructure across the source categories listed below.

Official service documentation and pricing pages are the primary source. These documents contain hardware specifications, resource allocation rules, capacity limits, scaling behaviours, and pricing structures that indicate the relationship between billing units and physical hardware. For each supported service, Greenpixie reviews the relevant documentation to identify which hardware components are involved in delivering each unit of work.

Pricing structures themselves carry significant hardware signals. The choice of billing unit, whether vCPU-hours, GB-seconds, requests, or GB-months, implies which hardware resources are consumed per unit of work. Pricing differentials between service variants, provisioned versus on-demand tiers, and regional price variations further inform the hardware profile.

Beyond official documentation, research draws on public technical disclosures made by cloud providers through conference presentations, published talks, engineering blog posts, and video content. Cloud architects and infrastructure engineers frequently discuss system design, hardware choices, and scaling architectures in public forums. These disclosures provide additional evidence for how services are built and what hardware underpins them.

5.8 Regulatory and Standards Sources

The methodology references the GHG Protocol Corporate Standard, the GHG Protocol Scope 2 Guidance, the GHG Protocol Scope 3 Standard, the GHG Protocol Product Life Cycle Accounting and Reporting Standard, ISO 14064-1, and ISO 14064-3. Regulatory context is drawn from the EU Corporate Sustainability Reporting Directive, the UK Streamlined Energy and Carbon Reporting framework, and the US Securities and Exchange Commission climate disclosure guidance.