Our AI Has an Invisible Footprint — Let's see How to Measure It

The environmental footprint of AI systems includes energy consumption, carbon emissions, water usage, and electronic waste. As AI engineers increasingly use third-party models via APIs, measuring this impact becomes more difficult because the computation is hidden on remote servers. A breakdown of Google's calculation for a single Gemini prompt reveals the methodology, which accounts for AI chip power, server overhead, and data center efficiency (PUE). This method highlights the key factors influencing an AI's environmental cost, particularly the carbon intensity of the local electricity grid. Engineers can use tools like EcoLogits and CodeCarbon to estimate these impacts for their own projects, promoting more sustainable practices.