AI offers significant economic and societal benefits, but understanding its environmental footprint, especially during inference, is crucial. Google has developed a comprehensive methodology to measure the energy, emissions, and water impact of AI inference, specifically for Gemini prompts. Their findings indicate that a median Gemini text prompt uses 0.24 watt-hours of energy, emits 0.03 grams of CO2 equivalent, and consumes 0.26 milliliters of water. These figures are substantially lower than many public estimates and represent real operational efficiency at scale. Google's AI systems have become significantly more efficient, with the energy and carbon footprint of a median Gemini text prompt dropping by 33x and 44x respectively over a recent 12-month period. This efficiency is attributed to a full-stack approach encompassing custom hardware, efficient model architectures, optimized algorithms, and ultra-efficient data centers. The methodology accounts for factors often omitted in other calculations, such as idle machines, CPU and RAM usage, and data center overhead. By sharing this methodology, Google aims to promote industry-wide consistency in calculating AI resource consumption. They are committed to continued improvements in AI efficiency, recognizing the growing demand for AI services. This focus on efficiency is considered essential for the responsible development of AI.