Estimating GHG Emissions from AI Use: Framework for Corporate-Level Measurement

Watershed has developed a framework that can help you produce credible and auditable estimates of greenhouse gas emissions stemming from your company's use of AI. The framework is comprised of three elements that are designed to make AI emissions measurement consistent and actionable: a comprehensive system boundary that includes model training, inference, data center overhead, and embodied hardware; a functional unit (kgCO2e per million tokens) that aligns with the metric companies already track for cost management; and a three-tier calculation approach that aligns with companies’ data quality. The framework also identifies practical levers for reducing AI emissions, including model selection, region selection, and prompt optimisation. This resource will be especially beneficial to sustainability, operations, and procurement professionals.