Researchers at the University of Washington have developed an artificial intelligence system that automatically calculates the environmental impact of manufacturing electronic devices, achieving accuracy comparable to human experts while reducing assessment time from months to about one minute. The breakthrough, published in Nature Electronics, addresses a critical gap in corporate transparency around the sustainability claims that dominate tech marketing today.

Why Is Device Carbon Footprint Data So Hard to Find?

When you shop for a flight on Google Flights, you see a quick carbon comparison between itineraries. But when you buy a new laptop or smartphone, you won't find similar sustainability information, even though electronics manufacturing carries substantial environmental costs. The reason is simple: calculating a device's carbon footprint is extraordinarily difficult and time-consuming, even for environmental experts.

A typical smartphone contains hundreds of chips and components, each produced with different emissions profiles. Much of this data isn't publicly available, forcing human experts to spend days or even months manually hunting through spreadsheets, product databases, and technical documents to piece together a complete picture. This opacity creates a perfect environment for what researchers call "AI greenwashing," where companies make vague sustainability claims without clear evidence to back them up.

"Recent studies have shown that people are willing to pay more for more sustainable devices. So there's growing demand for this information. But a phone, for example, is made of hundreds of chips and other components, and producing each of those causes varying amounts of emissions. Since that data isn't public or sometimes not even measured, human experts can spend days, even months manually gathering information for LCA. Instead we designed multiple AI agents that work together to automatically find this data and produce comparable estimates in about a minute," said Vikram Iyer, a UW assistant professor in the Paul G. Allen School of Computer Science and Engineering.

Vikram Iyer, Assistant Professor, University of Washington

How Do These AI Agents Actually Work?

The University of Washington team built a system using two AI agents that mimic the roles of human environmental assessors. The first agent acts as an analyst, defining what information needs to be gathered and reviewing results for accuracy. The second agent functions like an engineer, scraping publicly available data from product descriptions, images, technical documents, and even unconventional sources like FCC databases and iFixit teardown posts.

The two agents work in a loop, with the analyst setting the scope and the engineer gathering information, then the analyst reviewing and requesting additional searches if needed. Once they've compiled a complete parts list, the system references life cycle assessment (LCA) databases to convert component data into carbon estimates. The system achieved an average error rate of 5 to 19 percent, matching the accuracy of assessments conducted by human experts.

The researchers also developed a shortcut for common devices. They discovered that products with similar specifications, like screen size and processor type, cluster around similar carbon values because only a handful of companies manufacture specialized components for the entire industry. This "nearest-neighbors" approach allows the system to estimate carbon footprints for devices without complete data. When tested on missing emissions factors, the AI method achieved an average error of 23 percent, compared to 143 percent for human experts using traditional single-match approaches.

Steps to Reduce AI's Own Environmental Impact

• Use Smaller Models: The researchers deliberately chose compact AI models that consume far less energy than general-purpose large language models, reducing the computational overhead of running carbon assessments.
• Cache Previous Results: The system first checks whether a device's carbon footprint has already been calculated, allowing it to skip expensive AI computations if the data exists.
• Optimize Inference Loops: When the system does need to run AI models repeatedly, the total energy consumption is comparable to brewing a single cup of tea, demonstrating that even complex assessments can be made environmentally reasonable with careful design.

What Does This Mean for Corporate Sustainability Claims?

The timing of this research is significant. As AI becomes more embedded in economic and public life, the environmental consequences are increasingly shaped by the stories companies tell about their technologies. Researchers at RC Trust and Leiden University have examined whether some sustainability claims made by major tech companies could be challenged under European Union consumer protection law.

The tension is stark: large technology companies increasingly promote AI as environmentally beneficial, a tool for optimization and long-term sustainability. Yet independent research points to substantial environmental costs. Data centers require vast amounts of energy and water. Semiconductor production consumes resources and causes pollution. AI hardware contributes to electronic waste. And AI systems are used in areas that directly support environmentally harmful activities, including fossil fuel exploration and production.

"Sustainability claims shape how consumers, policymakers, investors, journalists, and the wider public understand the role of AI in society. If a company presents itself as moving towards carbon neutrality while its emissions continue to rise, or if it promotes vague claims about 'AI for sustainability' without clear evidence, such communication can influence public trust and consumer choices," noted Rachel Griffin, Postdoctoral Researcher at RC Trust.

Rachel Griffin, Postdoctoral Researcher, RC Trust

EU consumer protection law already prohibits misleading commercial practices, and recent updates have introduced stricter rules on greenwashing. Common forms of "sustainable AI" messaging that could fall within this legal framework include vague and unsubstantiated environmental claims, broad sustainability statements that only relate to part of a company's activities, and carbon neutrality claims based on offsetting rather than actual emissions reductions.

What's Next for Automated Carbon Assessments?

The University of Washington team plans to collaborate with companies to help automate their sustainability workflows. Many large corporations maintain dedicated sustainability teams that perform life cycle assessments, and automating this process could free those teams to focus on actually reducing product carbon footprints rather than hunting for data.

The research demonstrates that transparency doesn't require perfect information. With AI agents handling the tedious work of data collection and synthesis, companies can provide consumers with meaningful carbon comparisons in the same way flight booking sites do. This shift from opacity to accessibility could reshape how people evaluate the true environmental cost of the devices they buy, and it may finally give teeth to the sustainability promises that dominate tech marketing today.