As more companies embed artificial intelligence in their operations, the environmental impact of AI remains largely unmeasured, especially since many use closed AI models that do not reveal their energy consumption. Watershed, a startup specializing in emissions tracking, has developed a framework enabling businesses to estimate their AI-related carbon emissions. Their model factors in data center infrastructure and measures emissions per million AI tokens used. By integrating emissions data with existing AI usage tracking, companies can better align sustainability goals with cost management. Investors, auditors, and regulators are increasingly pressing firms to disclose these emissions, which currently fall under Scope 3 indirect emissions, such as business travel. While AI’s share of total corporate emissions might be modest now, it’s expected to grow rapidly, making early adoption of measurement tools crucial. Watershed’s approach not only estimates emissions but links them to actionable choices like AI model selection, query region, and prompt design, all of which affect energy use and carbon output. The challenge remains that precise energy data is scarce, largely due to closed proprietary models and ‘commercially sensitive’ information being withheld. Nevertheless, Watershed hopes companies will share emissions-per-token ratios, improving estimation accuracy over time and potentially highlighting efficiency gains in AI technology.
Back
