ChatGPT Might Not Be as Energy-Intensive as Previously Thought

12.02.2025

Recent research by Epoch AI, a nonprofit AI research institute, challenges the widely held belief that ChatGPT consumes a significant amount of energy. Contrary to the commonly cited figure of 3 watt-hours per query, which is said to be ten times more than a Google search, Epoch’s analysis suggests that a typical ChatGPT query uses only about 0.3 watt-hours, far less than many household appliances. This lower energy usage, according to Joshua You, the data analyst behind the study, shows that ChatGPT’s energy demands aren’t as burdensome as previously assumed, especially when compared to daily household activities like heating, cooling, or driving.

The study also addresses concerns surrounding AI’s environmental impact, especially as the technology becomes more widespread and infrastructure demands grow. Epoch’s analysis questions previous studies that may have overestimated energy usage, pointing out that earlier calculations were based on outdated models and assumptions about the hardware OpenAI uses. While Epoch’s figure is an estimate, it highlights the need to reconsider the energy costs associated with AI, as these technologies continue to evolve.

However, You notes that while ChatGPT’s current energy consumption is relatively low, this could change as AI models grow more advanced. As AI becomes more powerful, its training and usage will likely require more energy, particularly as new models are developed to handle more complex tasks. The increasing deployment of AI systems globally will also contribute to higher energy demands, as AI data centers expand to accommodate the growing need for computation.

AI’s rapid scaling presents an energy challenge, with predictions indicating that data centers could require vast amounts of power in the near future. A report from Rand suggested that by 2030, training frontier AI models might demand energy equivalent to that of several nuclear reactors. While OpenAI has made strides toward creating more energy-efficient models, the ongoing demand for more complex reasoning models—those that take longer to process tasks—will likely lead to higher energy consumption.

For those concerned about their energy footprint when using AI, You advises more selective usage of models like ChatGPT, using smaller, more efficient models when possible. While these smaller models may not handle every task, they offer a more energy-conscious alternative for those looking to minimize their environmental impact.