When new environmental challenges emerge, we don’t wait for the industry to give us the answers. We test new approaches, learn what works, and share those insights so others can move forward with us. When we couldn’t adequately measure the energy used by our cloud computing, we developed a new, industry-leading approach that inspired peers to take similar action. When shipping emerged as the largest source of emissions from our marketplace, we became the first global ecommerce company to measure, disclose and offset our shipping emissions.

Today, we’re presented with a new challenge: measuring and addressing the environmental impact of our use of Artificial Intelligence (AI).

How we’re meeting this moment

Etsy exists to Keep Commerce Human—to ensure that the real people behind the products on our marketplace have the opportunities they need to reach buyers around the world and grow their businesses. While AI presents a meaningful opportunity to strengthen our mission – from making it easier for buyers to discover items to building tools that help sellers grow their business – realizing this opportunity responsibly means understanding and addressing AI’s potential impacts. One of those is environmental.

AI uses significant energy and other resources, yet companies using AI often lack the data needed to understand their impact or identify effective ways to reduce it. It is difficult to manage household energy use without knowing which appliances use the most electricity; the same is true of AI. Better measurement helps companies understand how different models, tools, and usage patterns affect energy use and emissions.

To help address this gap, we partnered with Sustainable AI Group (SAIG) to develop and pilot an open-source methodology for estimating the energy use and emissions associated with large language models (LLMs). Etsy helped review, refine, and test the approach, and will be its first corporate adopter.

The methodology translates token usage - the units of text an AI model processes - into estimated energy use and emissions by model. It gives companies a more practical, granular way to understand their AI footprint and assess the impact of their decisions while provider data continues to improve.

A framework for the future

Measurement is the first step in a broader effort to reduce the environmental impact of AI. Our longer-term aim is to reduce emissions from the AI and cloud services we use, as part of our wider sustainability work. But those impacts span models, data centers, and providers, so progress requires action both within Etsy and across the industry.

To guide our sustainable engineering efforts, we’ve developed a framework aimed at grounding our actions where we can move the needle. We’ve designed it to be resilient, continuing to adapt with us as technology and our use of AI evolves.

• Catalyzing industry action: Alongside our work with SAIG, we participate in The Business Council on Climate Change (BC3) and BSR’s Environmental Impacts of AI coalitions. With other members, we are working toward a shared approach to measuring and disclosing AI’s environmental impact and engaging suppliers. A common approach can give AI providers clearer, more consistent requests for sustainability data—making those requests easier to fulfill and company disclosures easier to compare— and can help focus action where it can have the greatest impact.

• Influencing the AI supply chain: As a large-scale marketplace, we have meaningful purchasing power. We’re leveraging that to invest in lower-carbon technologies. For example, we’re deploying a part of our Emissions Budget to support a green concrete project aiming to help reduce the footprint of cloud and AI infrastructure. We’ve also begun asking AI suppliers sustainability questions informed by the SAIG approach, aligning our requests with peers while establishing a baseline to guide future engagement.

• Optimizing our own use of AI and machine learning: We’re implementing engineering practices that are intended to reduce our compute waste. This includes forming a sustainable engineering working group that is building and testing tools to help employees optimize AI and cloud applications. We’ve also shared internal guidance aimed at helping employees use AI and cloud applications more efficiently. These efforts support our target to keep our Google Cloud emissions intensity flat through 2027.*

You can find a deeper dive on our optimization efforts and Etsy’s history of leadership in sustainable engineering in our engineering blog, Code as Craft.

As our use of technology continues to evolve, our approach to addressing its impact will also evolve. We’re committed to continuing to build solutions to better measure our environmental impacts, advance more sustainable practices, and not only serve Etsy’s community, but create a positive impact on the world beyond it.

[^1]: Measured as tonnes of location-based carbon dioxide equivalent per $1M in sales.