10 times more powerful than Google
Each request to OpenAI's chatbot, which is capable of generating all sorts of responses to natural language queries, consumes 2.9 watt hours of electricity.
According to the International Energy Agency (IEA), this number is 10 times higher than when searching on Google.
OpenAI claims that ChatGPT now has 300 million weekly users making a total of one billion requests per day.
In addition to ChatGPT, the pioneering company that aims to bring generative AI to the masses in 2022, there are thousands of other chatbots in operation.
Electricity consumption is greater than France and Germany
Artificial intelligence (AI) wouldn't work without data centers that store massive amounts of information and computing power.
According to research by consulting firm Deloitte, by 2023, data centers will account for nearly 1.4% of global electricity consumption.
But with massive investments planned in generative AI, data center electricity consumption is expected to reach 3% by 2030.
The figure is equivalent to the annual consumption of France and Germany combined, according to auditing firm Deloitte.
The IEA predicts that data center electricity consumption will increase by more than 75% by 2026 compared to 2022 levels.
Generates a lot of CO2
Researchers at the University of Massachusetts Amherst estimated in 2019 that training one of the large language models (LLMs) that power chatbots would produce about 300 tons of CO2.
This figure is equivalent to the output of 125 round-trip flights between New York and Beijing.
Waste a lot of water
In addition to energy, AI also consumes water, especially to cool computer hardware.
According to conservative estimates by researchers at the University of California Riverside and the University of Texas at Arlington, GPT-3 needs about half a liter of water to generate between 10 and 50 responses.
Overall, the increased demand for AI is expected to consume between 4.2 and 6.6 billion cubic meters of water annually. According to the 2023 study, this is four to six times Denmark's annual water consumption.
Electronic waste pile
According to a study from the journal Nature Computational Science, about 2,600 tons of electronic waste such as graphics cards, servers, and memory chips will be generated from AI applications by 2023.
Researchers estimate that number will reach 2.5 million tons by 2030, equivalent to about 13.3 billion discarded smartphones.
And like many other computer hardware, AI devices including chips also require rare metals to manufacture.
Mining such metals, often in Africa, can involve processes that are heavily polluting.
