How much power will AI use, how much strain will it place on infrastructure, and how many additional nuclear power plants will we need to build?
As AI becomes more widespread, attention is also turning to its environmental impact.
AI Energy Use: The Hidden Cost of ChatGPT Queries - IEEE Spectrum
https://spectrum.ieee.org/ai-energy-use
When using an AI system, advanced calculations are performed in data centers somewhere in the world, consuming a significant amount of electricity for the calculations. According to Google's own calculations, the median power consumption of Gemini is 0.24Wh per prompt, carbon dioxide emissions are 0.03g, and the water consumption for machine cooling is 0.26ml. This is equivalent to the amount of energy consumed while watching television for 9 seconds or 5 drops of water.
Each Gemini uses the same amount of electricity as watching TV for 9 seconds and 5 drops of water.
Additionally, OpenAI CEO Sam Altman stated in a blog post in June 2025 that the average ChatGPT query would require approximately 0.34Wh of power, roughly the same as the power consumed by an oven in just over one second. OpenAI usage statistics calculated that an average user would make 25 queries per day, meaning each user would consume 8.5Wh of power per day.
OpenAI claims that ChatGPT serves 700 million weekly users and processes 2.5 billion queries per day. Assuming an average query consumes approximately 0.34 Wh, ChatGPT consumes 850 MWh of electricity per day across all users, equivalent to enough power to charge 14,000 electric vehicles.
Furthermore, 2.5 billion queries per day translates to approximately 900 billion queries per year, and if ChatGPT's user base continues to grow, it is likely to exceed 1 trillion queries per year. Using the same formula to calculate annual energy consumption, ChatGPT's annual consumption is estimated to be 310 GWh, equivalent to the annual electricity supply of approximately 29,000 American households.
Those who believe AI will continue to develop further predict that the average number of queries per day will increase dramatically over the next five years. A report by a research team on sustainable energy management at France-based Schneider Electric estimates that the overall energy consumption of generative AI will reach 15 TWh in 2025 and 347 TWh in 2030.
Even Stargate , a $500 billion (approximately 73 trillion yen) AI infrastructure construction plan led by OpenAI, has a total scale of 5GWh, which would require several Stargate-class data centers.
According to IEEE estimates, if the energy consumption of the generation AI were to reach 347 TWh, it would require 44 additional nuclear power plants.
According to Google, Gemini's energy consumption has been reduced by 33 times and its total carbon emissions by 44 times in the past year. Therefore, it is unclear what the energy consumption of generative AI will be in 2030, but since AI is expected to continue to expand, how to deal with increasing energy consumption and the challenges of more efficient energy management and technological innovation are becoming important issues to watch.
Related Posts:
in Software, Posted by log1e_dh