According to a new study from Epoch AI, ChatGPT doesn't consume as much power as many people think. The energy ChatGPT consumes depends on how it is being used and what model of ChatGPT is answering the queries. The estimates from most sources say that answering one question with ChatGPT requires 3 watt-hours of power which is 10 times more than what it takes for Google search to answer one question. But Epoch AI says that ChatGPT using 3 watt-hours of power for one query is too high and OpenAI’s GPT-4 model only takes 0.3 watt-hours to complete one query which is even less than the power used by household appliances.
The environmental challenges with AI’s energy consumption have become a heated topic because AI companies are continuously expanding their infrastructures. Recently, about 100 organizations have signed an open letter which is urging the industry to ensure that new AI data centers which are being built do not harm the environment. So the industries are asked to rely on non-renewable energy sources.
The researcher on the study, Joshua You, said that there is a lot of discourse about how much energy is going to be used in the future but there hasn't been conversations which accurately portray AIs energy consumption right now. He said that AI using 3 watt-hours per query seems too high and it was based on some of his older research and rough calculations. OpenAI hasn’t shared how much its models use energy per query but You says that 0.3 watt-hour seems like an accurate figure. The analysis hasn't calculated the extra energy used by ChatGPT like for processing longer inputs or image generation.
AI is getting more advanced and it is going to require more energy for handling complex tasks in the future. By 2030, it is predicted that training cutting-edge AI models will require 8 GW of energy which is equal to eight nuclear reactors. Joshua You says that people who are concerned about AI’s energy footprint should stop using AI models like ChatGPT or opt for smaller models like GPT-4o-mini.
Read next: AI Chatbots Drive Higher Engagement and Transactions Than Search Traffic
The environmental challenges with AI’s energy consumption have become a heated topic because AI companies are continuously expanding their infrastructures. Recently, about 100 organizations have signed an open letter which is urging the industry to ensure that new AI data centers which are being built do not harm the environment. So the industries are asked to rely on non-renewable energy sources.
The researcher on the study, Joshua You, said that there is a lot of discourse about how much energy is going to be used in the future but there hasn't been conversations which accurately portray AIs energy consumption right now. He said that AI using 3 watt-hours per query seems too high and it was based on some of his older research and rough calculations. OpenAI hasn’t shared how much its models use energy per query but You says that 0.3 watt-hour seems like an accurate figure. The analysis hasn't calculated the extra energy used by ChatGPT like for processing longer inputs or image generation.
AI is getting more advanced and it is going to require more energy for handling complex tasks in the future. By 2030, it is predicted that training cutting-edge AI models will require 8 GW of energy which is equal to eight nuclear reactors. Joshua You says that people who are concerned about AI’s energy footprint should stop using AI models like ChatGPT or opt for smaller models like GPT-4o-mini.
Read next: AI Chatbots Drive Higher Engagement and Transactions Than Search Traffic
