QAZAQ GREEN.  Google has released detailed technical data on the energy and environmental impact of its Gemini AI model operations, providing the first comprehensive methodology for measuring AI inference footprint at scale.

According to the company's research paper, a median Gemini Apps text prompt consumes 0.24 watt-hours of energy, emits 0.03 grams of carbon dioxide equivalent, and uses 0.26 milliliters of water. The energy consumption is equivalent to watching television for less than nine seconds.

The methodology accounts for full system dynamics including idle machine capacity, CPU and RAM usage, data center overhead, and cooling water consumption. Many existing AI energy calculations only measure active processing power, which Google estimates at 0.10 Wh per prompt - less than half the actual operational footprint.

Google reports significant efficiency improvements over the past 12 months, with energy consumption per prompt decreasing by 33 times and carbon footprint dropping by 44 times while maintaining response quality. These gains result from architectural improvements including Mixture-of-Experts models that activate only necessary components for each query, reducing computations by 10-100 times.

The company's custom TPU processors show substantial efficiency gains, with the latest Ironwood generation consuming 30 times less energy than the first publicly available TPU. Google's data centers operate at an average Power Usage Effectiveness of 1.09, among the industry's most efficient ratings.

The research addresses growing concerns about AI's environmental impact as adoption scales globally. Google's methodology aims to establish industry standards for measuring AI resource consumption, as current estimates vary widely due to incomplete measurement approaches.