Startup offers framework to estimate corporate AI emissions
As businesses increasingly adopt AI, tracking its environmental impact presents a significant challenge. Startup Watershed has released a new framework to help companies estimate their AI-related carbon emissions.
Watershed, a company focused on helping businesses track their emissions, has developed a framework to estimate the carbon footprint of artificial intelligence (AI) use. The increasing integration of AI into corporate operations and products brings with it an environmental impact, primarily through energy consumption, which has become difficult to quantify, especially with closed AI models that do not disclose their energy usage.
The framework calculates emissions based on data center infrastructure and the volume of AI tokens used, providing a figure in kilograms of CO2 per million tokens. "Companies are already tracking AI usage at the token level for cost management," stated John Bistline, Watershed's head of science. "The emissions math plugs into that same data. So this isn't asking companies to build something entirely new. Cost and sustainability go hand in hand here."
There is growing pressure from investors, auditors, and regulators for companies to disclose their AI emissions. This includes indirect, or Scope 3, emissions, such as those generated by employee business travel or the energy required for AI queries. Some regions, like California, already mandate Scope 3 reporting, and global standards bodies are considering similar requirements for cloud and AI services.
"AI may be a small part of most companies' footprints currently. But nobody expects that to stay the case for long," Bistline added. "The companies that build their measurement infrastructure now will be better prepared than those who wait." Quantifying AI emissions can also help corporations identify opportunities to reduce both their environmental impact and operating costs.
Watershed's framework provides an estimate, as precise measurement is not yet feasible. Many widely used AI models are proprietary, preventing independent energy consumption testing. While AI providers may not share all data, Watershed aims for them to disclose emissions per token. As more information becomes available, these estimates are expected to become more precise.