Researchers at Hugging Face have revealed that the carbon footprint of generative AI, particularly text-to-video tools, is significantly underreported due to non-linear energy scaling. Their study indicates that energy consumption for video generation quadruples when the video’s length doubles. For example, a six-second AI-generated video uses as much energy as a microwave running for over an hour. This highlights the inefficiencies in current video production pipelines and the urgent need for more eco-friendly designs. Experts warn that generative AI is being deployed without fully understanding its environmental impact, with AI activities accounting for 20% of global datacenter power demand. Major tech firms, like Google, are investing heavily in AI infrastructure, resulting in increased carbon emissions—Google reported a 13% rise attributed to generative AI services. Suggested mitigation strategies include intelligent caching and pruning inefficient data, though the effectiveness of these measures remains uncertain.
Source link
