OpenAI CEO Sam Altman has published detailed calculations on his blog about the electricity and water consumption of processing a single ChatGPT request. This data helps to estimate the real-world environmental footprint of a neural network.
OpenAI CEO Sam Altman has published detailed calculations on his blog about the electricity and water consumption of processing a single ChatGPT request. This data helps to estimate the real-world environmental footprint of a neural network.
In his blog, Sam Altman revealed that a single ChatGPT request consumes about 0.34 watt-hours of energy. For comparison, that’s about the same amount of energy an average electric oven uses per second, or as much as an energy-saving light bulb burns for a few minutes. This comparison helps to better understand the scale of the energy consumption, which is quite low, but noticeable on a global scale.
In addition to electricity, a neural network requires water — about 0.00032 liters per query, which is about one-fifteenth of a teaspoon. Water is used to cool servers in data centers that process millions of queries per day. These calculations allow us to see the real resource footprint of artificial intelligence, which is often thought of as «incorporeal» and not related to physical costs.
Altman emphasized that OpenAI aims to develop technologies that are as resource-efficient and environmentally responsible as possible. While the company is focused on creating the superintelligence of the future, it is also actively working to optimize existing systems to reduce their environmental impact.
The post also mentions that the technological development of artificial intelligence is accompanied by constant challenges — from energy consumption to privacy. At the same time, developers strive to make AI a tool that serves society, maintaining a balance between innovation and sustainable development.
Thus, the real numbers presented by Sam Altman challenge the notion of «carbon-free» digital services. They demonstrate that even artificial intelligence requires significant resources, but at the same time highlight the potential for further optimization and environmental responsibility in the industry.
We previously wrote about how scientists calculated how much water or electricity is needed to operate ChatGPT-4. We suggest comparing their calculations with the data provided by Sam Altman.
