Since the publication of the Software Carbon Intensity (SCI) specification, practitioners and organizations worldwide have consistently used it to measure and report emissions across their systems. But as AI entered mainstream software development, one question has remained unanswered: How do we reliably measure its carbon footprint?
Without consistent measurement, organizations couldn't compare different AI implementations, validate reduction claims, or make informed decisions about model selection and deployment strategies. While speaking with our members, it became clear that existing measurement approaches were not enough to capture AI’s unique operational patterns and lifecycle impacts.
We’re thrilled to announce the ratification of the Software Carbon Intensity for Artificial Intelligence (SCI for AI) specification, the standardized methodology that captures AI's complete environmental impact across its entire lifecycle. Developed under the leadership of Navveen Balani (Accenture) and Henry Richardson (WattTime), this specification is the first consensus-based standard that makes the carbon footprint of AI systems transparent, comparable, and actionable.
Why The SCI for AI Matters Now
As AI becomes infrastructure, reliable measurement becomes essential. Organizations need standardized methods to track net-zero progress, prepare for carbon pricing, and reduce operational costs through improved efficiency. Practitioners need consistent metrics to validate improvements and make informed architecture choices. Regulatory frameworks increasingly require environmental disclosure. The SCI for AI responds by providing the foundation for accurate, consistent measurement and credible reporting.
The specification covers the full spectrum of AI systems: classical machine learning, computer vision, natural language processing, generative AI, and agentic systems. It supports diverse functional units—tokens for language models, inferences for classifiers, and FLOPs for training efficiency—reflecting how AI systems are developed and deployed in practice.
Built on the ISO-standard SCI methodology (ISO/IEC 21031:2024), it extends proven carbon measurement principles to AI's unique lifecycle stages and operational patterns.
“To understand AI's carbon footprint, we first need a consistent way to measure it,” said Jonathan Turnbull, Environment & AI lead at Google and member of GSF's Green AI Committee. “The SCI for AI is an important step to give the industry a shared standard to track emissions across the entire lifecycle, not just parts of it. This clarity allows us to move from rough estimates to real, actionable data.”
A Closer Look at SCI for AI
AI systems vary widely in architecture, scale, and deployment patterns, generating different emissions depending on where and when they run. Traditional approaches focus on inference costs, missing the significant carbon footprint of training and data preparation.
SCI for AI addresses this complexity by defining two measurement boundaries, aligned with stakeholders’ capacity to reduce emissions.
The specification calls for providers to measure and disclose two scores:
Provider score: Covers model development, training, and deployment efficiency.
Consumer score: Enables users to understand and reduce operational impacts from inference and monitoring.
This approach gives builders and operators control over the emissions they can influence.
“The purpose of this specification is to assist AI practitioners—developers, data scientists, engineers, and decision-makers—in understanding and reducing the carbon footprint of AI systems,” said Navveen Balani, Software Standards Working Group Chair. “By making informed choices about model design, computational efficiency, and deployment strategies, they can minimize emissions while maintaining performance.”
Building on Our Values
As our first ratified specification extending the SCI methodology, SCI for AI represents both the maturity of our standards process and the community's commitment to transparent, measurement-first approaches to emerging technical complexities.
Developed through an extensive collaborative process among AI experts from over 20 GSF member organizations, it addresses implementation challenges across open, proprietary, distributed, and cloud-hosted AI systems.
As organizations increasingly rely on AI, the specification plays a central role in closing the gap between measurement capability and organizational action.
Looking Ahead
This achievement demonstrates GSF's unique capacity to translate technology challenges into actionable standards that practitioners and organizations can adopt immediately.
In 2026, we'll focus on supporting implementation through:
Practical guidance on applying the methodology to production AI systems
Case studies demonstrating measurement across different AI architectures
Training resources for practitioners adopting the standard
Get Involved
The SCI for AI specification is open for public review. We invite AI practitioners, sustainability professionals, and organizations to review the methodology, apply it to their systems, and share their implementation experience as they begin measuring AI carbon emissions.
Explore the SCI for AI specification and provide feedback: https://sci-for-ai.greensoftware.foundation/
GSF projects are developed under the guidance of our Steering Members: Avanade, BCG X, Cisco, Google, Microsoft, NTT DATA Corporation, Siemens, and UBS. To join industry leaders shaping software measurement and to contribute to the standards defining how AI's environmental impact is measured and reduced, consider becoming a member.
This article is licenced under Creative Commons (CC BY 4.0)