Generative AI, particularly for video applications, incurs significant energy costs that can affect both users and the environment. While training large language models demands considerable electricity, video models like OpenAI’s Sora app require even more power. A study by Sasha Luccioni from Hugging Face reveals that video diffusion consumes 30 times the energy of image generation and 2,000 times more than text generation. Creating a single AI video can use around 90 Watt-hours—equivalent to running a 65-inch TV for approximately 37 minutes. As AI tools grow in popularity, concerns rise over the strain on the US electrical grid, prompting industry and government initiatives for AI infrastructure investments. The need for efficient AI models is critical, and companies must prioritize transparency about their energy consumption to enable users to make informed choices. As generative AI continues to evolve, understanding its environmental impact is essential for sustainable technology adoption.
Source link
