Google Releases AI Energy, Carbon Measurement Methodology

Google has released a comprehensive methodology to measure the energy, water, and carbon emissions of its AI models, aiming to improve efficiency and reduce environmental impact. The initiative addresses the growing energy demands of artificial intelligence, which is increasingly applied in fields such as medicine, energy, autonomous systems, and quantum computing.

AI has the potential to help tackle major societal challenges, from improving student learning and diagnosing cancer earlier to enhancing transportation and cybersecurity systems and predicting wildfire paths for first responders. Realizing this potential, however, requires robust energy infrastructure, efficient energy use, and innovative technology solutions.

Google is approaching these challenges through investments in new infrastructure, smarter and more resilient grids, and the expansion of clean energy sources. The company is also focusing on maximizing efficiency across its operations, from custom-built hardware to the software and models running in its data centers.

Historically, comprehensive data on the environmental impact of AI inference has been limited. Google’s new methodology helps close this gap by providing detailed measurements of the energy and carbon footprint of its AI models. Over the past year, improvements in model efficiency have led to a 33-fold reduction in median energy consumption and a 44-fold decrease in carbon footprint per Gemini Apps text prompt, while delivering higher-quality responses. The energy consumed per median prompt is now equivalent to watching television for less than nine seconds.

These advancements build on Google’s broader commitment to data center efficiency. In 2024, the company reduced data center energy emissions by 12% even as electricity consumption grew 27% year-over-year due to business and service expansion.

Google emphasized that transparency and shared methodologies are critical for driving progress in AI efficiency. The company hopes that its study will support global efforts to develop sustainable AI at a time when energy, sustainability, and scientific discovery are increasingly intertwined.