Project By

Measure AI's Carbon Footprint with Purpose

A standardized specification extending the Software Carbon Intensity methodology to measure the carbon emissions of AI systems throughout their lifecycle.

AI systems have become increasingly resource-intensive, yet there's no consistent way to measure their environmental impact. SCI for AI changes that by providing the first consensus-based standard that makes AI's carbon footprint transparent, comparable, and actionable. Help refine this transformative specification with your feedback and experience as we revolutionize how we build and deploy sustainable AI.

Leading organizations collaborating to standardize AI carbon measurement

Avanade BCG Cisco Google Microsoft NTT DATA Siemens UBS Accenture ACIS Almaviva Amadeus Ausy-Technologies (Randstad) Aveva Bloomberg Boavizta CAST Chalmers University of Technology ClimateAction.tech Cloud Sustainability Watch Code from Finland Concordia University Digital Emissions DXC EDSP ECO Electricity Maps Energy Web Envite Futurewei German Informatics Society GitHub GitLab Globant Goldman Sachs GovTech Singapore Green Software Foundation GrUSP HARC HSBC IBM Indeed Indian Institute of Management Mumbai Infocomm Media Development Authority Interfima International Institute of Information Technology Leaders for Climate Action Linnaeus University Liquid Reply Manchester Metropolitan University Mastercard Mercedes-Benz Tech Innovation Metropolitan Police mondragon.edu Nomura Research Institute, Ltd OpenUK Renewable Energy Laboratory (NREL) Responsible AI Schneider Electric Scope3 Sentry Shell Supercritical Sustainable Digital Infrastructure Alliance Texas State University The Explorers The Green Web Foundation Thoughtworks Time for the Planet Trust.ai UCL Universidad Politécnica de Madrid Università degli Studi dell'Aquila University of Amsterdam University of Bristol University of Edinburgh University of Huddersfield University of Limerick University of Salento Virtasant WattTime Xero ZutaCore
Avanade BCG Cisco Google Microsoft NTT DATA Siemens UBS Accenture ACIS Almaviva Amadeus Ausy-Technologies (Randstad) Aveva Bloomberg Boavizta CAST Chalmers University of Technology ClimateAction.tech Cloud Sustainability Watch Code from Finland Concordia University Digital Emissions DXC EDSP ECO Electricity Maps Energy Web Envite Futurewei German Informatics Society GitHub GitLab Globant Goldman Sachs GovTech Singapore Green Software Foundation GrUSP HARC HSBC IBM Indeed Indian Institute of Management Mumbai Infocomm Media Development Authority Interfima International Institute of Information Technology Leaders for Climate Action Linnaeus University Liquid Reply Manchester Metropolitan University Mastercard Mercedes-Benz Tech Innovation Metropolitan Police mondragon.edu Nomura Research Institute, Ltd OpenUK Renewable Energy Laboratory (NREL) Responsible AI Schneider Electric Scope3 Sentry Shell Supercritical Sustainable Digital Infrastructure Alliance Texas State University The Explorers The Green Web Foundation Thoughtworks Time for the Planet Trust.ai UCL Universidad Politécnica de Madrid Università degli Studi dell'Aquila University of Amsterdam University of Bristol University of Edinburgh University of Huddersfield University of Limerick University of Salento Virtasant WattTime Xero ZutaCore

What is SCI for AI?

SCI for AI extends the globally adopted Software Carbon Intensity (SCI) ISO specification to address the unique characteristics of artificial intelligence systems. It provides a standardized methodology for calculating carbon emissions rates across the entire AI lifecycle, from data preparation and model training to deployment and inference.

Unlike simple energy metrics or carbon offsets, SCI for AI creates a comprehensive score that incentivizes real emission reductions. By making the true carbon cost of AI transparent and comparable, it transforms sustainability from an abstract goal into a measurable, optimizable metric that drives innovation in efficient AI architectures and influences strategic decisions across industries.

Why SCI for AI Matters

As AI becomes ubiquitous across industries, its environmental footprint grows exponentially, but measurement remains fragmented and inconsistent.

Industry Impact

SCI for AI provides the first consensus-based, standardized approach to measuring AI's environmental impact. This standardization drives innovation in efficient AI architectures, influences procurement decisions, and helps organizations meet sustainability commitments. By revealing the true carbon cost of AI development and deployment, it enables meaningful comparisons between different systems and approaches, transforming how organizations think about AI investments.

Business Benefits

Reduce operational costs through improved computational efficiency and optimized cloud resource consumption

Prepare for future carbon pricing and regulatory requirements with ISO-compatible measurement standards

Gain a competitive advantage through transparent sustainability metrics for AI products and services

Make informed trade-offs between model performance and environmental impact with clear, actionable data

Build stakeholder trust through demonstrable commitment to responsible AI development

Environmental Impact

SCI for AI directly addresses the growing carbon footprint of AI systems by providing metrics that incentivize real reductions rather than offsets. The specification reveals the full picture of AI emissions, from data preparation through training to inference, exposing impacts in early stages that often dwarf inference costs. This visibility encourages practices like model optimization, efficient architectures, and carbon-aware computing that significantly reduce AI-related emissions by enabling informed choices about when, where, and how AI systems operate.

Understanding AI's Carbon Lifecycle

SCI for AI measures emissions across every stage of AI development and deployment, revealing optimization opportunities throughout the entire lifecycle.

Inception

Scoping the problem and setting constraints

Design and Development

Where major emissions accumulate through training

Deployment

Integrating AI into production systems

Operation and Monitoring

Inference, orchestration, and ongoing maintenance

End of Life

Decommissioning systems and handling data

AI's Hidden Emissions

Traditional approaches often focus solely on inference costs, missing the significant carbon footprint of training and data preparation. SCI for AI provides comprehensive lifecycle coverage, including often-overlooked stages like data engineering and system integration. This holistic view enables organizations to identify and address the true sources of AI emissions.

Transformative Capabilities

SCI for AI brings unprecedented clarity to AI sustainability through innovative features designed for real-world application.

Comprehensive Coverage

Measures emissions from data preparation through end-of-life, capturing impacts others miss

AI-Native Design

Supports all AI paradigms: ML, deep learning, generative AI, and emerging technologies

Clear Boundaries

Precise definitions for measuring different AI systems with appropriate functional units

Engineering Focus

Incentivizes direct optimizations rather than relying on carbon offsets

Industry Consensus

Developed with input from major players with royalty-free IPR for broad adoption

A New Architecture for Energy Intelligence

Building on Industry Collaboration

In early 2025, AI experts from GSF member organizations participated in a series of workshops hosted by the Software Standards Working Group. These sessions were designed to define the GSF approach to AI measurement as well as evaluate existing metrics. The outcomes laid the groundwork for creating the SCI for AI specification.

The purpose of this specification is to assist AI practitioners in understanding and reducing the carbon footprint of AI systems. By making informed choices about model design, computational efficiency, and deployment strategies, practitioners can minimize emissions while maintaining performance.

Navveen Balani

Software Standards Working Group Chair

Accenture / Green Software Foundation

Explore Further

Deep dive into SCI for AI methodology and related resources

Software Carbon Intensity (SCI) Specification

Explore the parent specification that SCI for AI extends with AI-specific applications

Green Software Patterns for AI

Discover proven patterns for reducing AI carbon emissions in your applications

Shape the Future of Sustainable AI

Your expertise and experience can help refine this transformative standard

Read the Specification

Read the full specification and share your feedback

Join the Software Standards Working Group (Members Only)

Collaborate with industry experts shaping this specification

Visit the Directory

Get in touch with project leads

Inquiry about becoming a GSF member

Reach out directly to the GSF team

Development Timeline

  1. Q4 2024

    Proposal

  2. Q2 2025

    Pre-Draft

  3. Q3 2025

    Draft

  4. Q4 2025

    Consistency Review & SC Ratification

  5. Q1 2026

    Publication & ISO-readiness approval

  6. Q2 2026

    ISO Submission

Project Leadership

Part of the Software Standards Working Group

Navveen Balani

Software Standards Working Group Chair

@Accenture

Henry Richardson

Software Standards Working Group Chair

@Watt Time

About the Green Software Foundation

SCI for AI is developed and maintained by the Green Software Foundation (GSF), a nonprofit consortium under the Linux Foundation dedicated to building a future where software has zero harmful environmental impact. With steering members including Accenture, Avanade, BCG X, Cisco, Google, Microsoft, NTT Data, Siemens, and UBS, GSF unites industry leaders to create practical standards and specifications that make sustainable computing achievable at scale.

SCI for AI builds on GSF's proven expertise in carbon measurement, including the ISO-certified Software Carbon Intensity (SCI) specification already in production globally, and open-source tools like the Impact Framework and Carbon Aware SDK that enable organizations to measure and reduce their software emissions.