This August 2025 paper presents Google's comprehensive methodology for measuring the environmental impact of AI inference workloads in a large-scale production environment. It addresses a critical gap in existing research by accounting for the full stack of AI serving infrastructure, including active AI accelerator power, host system energy, idle machine capacity, and data center overhead. The paper reveals that a median Gemini Apps text prompt consumes significantly less energy, carbon emissions, and water than many prior public estimates. Furthermore, it highlights Google's efforts in software efficiency and clean energy procurement, which have led to substantial reductions in the environmental footprint of AI serving over the past year. The authors advocate for a standardized, comprehensive measurement framework to accurately compare AI models and incentivize further efficiency gains across the industry.Source:https://arxiv.org/pdf/2508.15734

Ver todos los episodios