Developing an Impact Framework of Planetary Justice Impacts of AI

Written By Sara Marcucci

The AI Supply Chain Impact Framework is a tool designed to examine the full lifecycle of AI systems—from raw material extraction through manufacturing, model training and deployment, to disposal. It maps out the environmental and social impacts at every stage, helping to uncover the often hidden network of processes that contribute to AI’s overall footprint.

We developed this framework because we realized that behind every AI system lies a hidden, often overlooked network of processes—ranging from mining and manufacturing to training, deployment, and disposal—that affect both our environment and communities. Too often, discussions about AI focus only on the code and algorithms, ignoring the tangible impacts such as the strain on natural resources, worker conditions, and local community disruptions.

This methodology is aspirational—it outlines our ideal vision of what we want to know about an AI system’s true impacts. Focusing on local effects, such as the direct impacts of mining or assembly on nearby communities, we hope to establish a clear, manageable foundation. Over time, as more data becomes available, this approach will allow us to expand our understanding and build a more comprehensive picture of the AI supply chain's environmental and socio-political footprint.

Why We Built This Framework

Our goal is to ask the tough questions about where the materials for AI come from, who makes the hardware, how energy and water are used, and what happens when AI systems are no longer needed. We believe that understanding these questions is a vital first step in addressing the true cost of AI—not just in terms of dollars, but in its impact on our planet and people. Even when we don’t have all the answers, raising these questions is key to pushing for more transparency and better data in the future.

How We Developed It

We started by reviewing existing research—such as the paper “Estimating the Carbon Footprint of BLOOM, a 176B Parameter Language Model“ by Sasha Luccioni, Sylvain Viguier, and Anne-Laure Ligozat—and broke down the AI supply chain into clear stages. For each stage, we developed questions that cover both the measurable (like energy use and emissions) and the less quantifiable (like worker safety and community well-being). Along the way, we faced challenges such as missing or proprietary data and the complexity of assigning responsibility for impacts when the AI system is only a part of a larger process. We tackled these issues by including explicit prompts to note data gaps, suggest proxy figures, and attribute responsibility proportionately.

What the Framework Covers

The framework is organized into several stages:

  • Raw Material Extraction: Where and how are the necessary minerals and resources obtained?

  • Materials Manufacturing: How are these raw materials processed into usable components?

  • Equipment Manufacturing: How is the physical hardware built, and what are its environmental footprints?

  • Model Training: What are the impacts of running energy- and water-intensive training sessions?

  • Model Deployment: How does running the AI system affect local communities and resource use?

  • Disposal/End-of-Life: What happens when the hardware is outdated or no longer needed?

A screenshot from the Framework.

For each stage, we’ve included questions that address both environmental impacts (like water use, energy consumption, and emissions) and social effects (such as labor conditions, community displacement, and cultural impacts). We also stress the importance of documenting where data is missing and explaining why, so that every assessment is as transparent as possible.

Limitations

We know that this framework isn’t perfect. Many aspects of the AI supply chain are still hidden behind proprietary walls or incomplete reporting. That’s why we openly acknowledge these challenges and invite you to help us fill the gaps. The framework is a living document—it will change, improve, and grow as more data and insights become available.

Next Steps: Stakeholder Engagement

Now that we have launched the framework, our next step is to actively engage with stakeholders. We will begin testing the framework through interviews, workshops, and collaborative sessions with researchers, industry practitioners, community representatives, and policymakers. These insights and feedback will be critical in refining the tool, addressing uncertainties, and ensuring that the framework remains a living, evolving resource.

If you have questions, comments, or ideas for improvement, please use this link to join our collaborative document. Your feedback will help us refine this tool and make it as useful and transparent as possible for everyone—from researchers and policymakers to community members and advocates.

Download the Framework here!

Sara Marcucci