Your laptop might politely sip electricity at your desk, but it has already feasted on a banquet of fossil fuels—through extraction, refining, shipping and assembly—before you even took it out of the box.
Recent global surveys show that 70% of consumers are willing to pay more for environmentally friendly electronic devices, but they lack the data necessary to make an informed choice.
To fulfill this need, Tingyu Cheng, assistant professor of computer science and engineering at the University of Notre Dame, teamed up with researchers at the University of Washington (UW), Brown University, and Northeastern University. Together, they developed a multimodal AI tool that estimates the carbon footprint of electronic devices. The tool’s estimates differ from human experts’ calculations by approximately 19%. This is an impressive degree of accuracy when you take into consideration the fact that human experts differ among themselves by about the same margin and require weeks instead of less than a minute to deliver an answer. The team’s research was published in Nature Electronics.
“Previously, there was no way a customer could get a ‘cradle-to-factory gate’ analysis of a device’s carbon use, or life cycle assessment [LCA],” said Cheng. “To calculate it manually, you have to know every single material inside a device down to the gram, plus exactly where it was mined and shipped from. Usually, you just don’t have that data. So, how do you know if an iPhone is worse for the environment than any other device?”
By leveraging recent advancements in smart text (LLMs), vision-language models, and AI agents, the team bypassed the need for proprietary corporate blueprints and factory spreadsheets. Instead, the tool requires only a product’s name to calculate a device’s embodied carbon footprint—that is, the emissions caused by the product’s creation before it’s even plugged in. Companies can also use the LCA tool early in a product’s design phase to identify possible environmental hotspots in its life cycle.
When some of the factors necessary to calculate LCA are unknown—for instance, when details about a cutting-edge chip or a laptop display are kept as corporate secrets—the tool’s AI agents analyze comparable components in publicly available databases.
In addition to accurately tracking and reporting the carbon footprint of devices, the team’s multimodal tool effectively communicates these results to end users. To support this goal, the team developed a related browser-based interactive system called Living Sustainability, which can scan device descriptions, identify individual components, and estimate the carbon cost associated with each component, as well as the carbon costs of assembly and shipping. This related work was published in IMWUT 2025.
Operating AI systems, even those with environmentally friendly goals, can make a sizable carbon footprint of their own. To improve the system’s efficiency, it uses smaller, task-specific AI models and saves every calculation it has ever performed into a shared database, which retrieves old answers from memory rather than burning computer energy to recalculate them from scratch.
“As the Internet of Things expands, we need to design more eco-friendly electronics and make sure the carbon footprint responsible for producing those electronics stays minimal,” said Cheng.
—Karla Cruise, Notre Dame Engineering