Despite most users only seeing quick responses, massive data centers and powerful servers consume vast amounts of electricity, reshaping global energy consumption patterns. In 2023, data centers accounted for 4.4% of total electricity use in the United States and about 1.5% globally. This percentage is expected to at least double by 2030, largely driven by the surge in artificial intelligence usage, according to a report. Energy consumption is not limited to training; every time millions of users send messages, a series of “inference” operations activate the trained model to generate responses. Although inference requires less energy per query compared to training, the sheer volume of requests results in a massive cumulative energy consumption. Simply put, the energy used for about 26 queries on some models is enough to power a typical microwave for three minutes or a 10-watt LED bulb for 12 minutes.
Despite advancements in efficiency and hardware, the unprecedented rise in AI adoption challenges efforts to maintain sustainable energy use. The problem is exacerbated by the lack of transparency about the exact energy consumption figures of AI models owned by tech giants. Researchers are calling for greater disclosure to enable users to make informed choices, and policymakers are urged to implement regulations that encourage energy accountability. As the digital age accelerates, understanding and addressing the true cost of AI-powered convenience will be crucial.
