By AFP - Agence France Presse
The environmental impact of generative AI in numbers
By Daxia ROJAS
The rise of generative artificial intelligence (AI) is accompanied by growing fears about the technology's ecological footprint, one of the main issues discussed at a global summit in Paris on February 10 and 11.
Here are some important figures on the situation at the beginning of 2025:
Each request made to OpenAI's chatbot, which is capable of generating all kinds of answers to queries in natural language, consumes 2.9 watt-hours of electricity.
This is ten times more than the equivalent value for a Google search, according to the International Energy Agency (IEA).
OpenAI claims that ChatGPT now has 300 million weekly users who make a total of one billion requests every day.
In addition to ChatGPT, which spearheaded the emergence of generative AI in the public consciousness in 2022, there are thousands of chatbots.
A survey carried out by French researchers Ifop revealed that 70% of 18-24-year-olds in the country said they used generative AI.
In the United States, a Morning Consult survey found that 65% of 13 to 17-year-olds used generative AI, with the figure close to half for the general population.
Generative AI wouldn't work without data centers that host vast reserves of information and computing power.
In 2023, data centers were responsible for almost 1.4% of global electricity consumption, according to a study by consultancy Deloitte.
But with the massive investments planned for generative AI, this figure is expected to reach 3% by 2030, or 1,000 terawatt hours (TWh).
Deloitte said that this figure is comparable to the combined annual consumption of France and Germany.
The IEA predicted an increase of more than 75% in data center energy consumption by 2026, compared to 2022 levels, to 800 TWh.
American consultancy Gartner said that the high energy demand meant that up to 40% of data centers built for AI applications could face electricity shortages by 2027.
Training one of the large language models (LLMs) that power chatbots generate around 300 tons of greenhouse gas carbon dioxide, researchers at the University of Massachusetts Amherst estimated in 2019.
That's about the same output as 125 round-trip flights between New York and Beijing.
Two years later, researchers at Oxford University estimated the figure at 224 tons for a single training session of OpenAI's GPT-3 model.
Developers need to train thousands of models to advance their technology.
Despite these estimates, the researchers say that judging the overall greenhouse emissions of generative AI is a challenge.
Experts and institutions have pointed to the lack of information on how the models are produced, as well as the absence of global measurement standards.
In addition to energy, generative AI also consumes water, especially to cool the computer hardware.
The GPT-3 requires around half a liter of water to generate between 10 and 50 responses, according to a conservative estimate by researchers at the University of California Riverside and the University of Texas at Arlington.
Overall, the increase in water demand from AI is expected to be 4.2 billion to 6.6 billion cubic meters (155 billion to 233 billion cubic feet).
That's four to six times Denmark's annual water consumption, according to the same 2023 study.
Around 2,600 tons of electronic waste, such as video cards, servers, and memory chips, emerged from generative AI applications in 2023, according to a study in the journal Nature Computational Science.
The researchers extrapolated this figure to 2.5 million tons by 2030 if current trends continue and nothing is done to limit waste.
That would be the equivalent of around 13.3 billion discarded smartphones.
And like much computer hardware, AI equipment, including chips, requires rare metals to be manufactured.
Mining these metals, usually in Africa, can involve highly polluting processes.
dax/tgb/js
Comments