Water, electricity: the staggering energy cost of AIs like ChatGPT

© Pexels/Pavel Danilyuk

The operation of ChatGPT and its rivals relies primarily on gigantic data centers that are constantly running to respond to our requests. To get a clearer picture, the Washington Post has published a very informative infographic. The journalists have notably taken up the work of Shaolei Ren, a researcher at the University of California, in order to evaluate the water and electricity costs of these technologies. They then cross-referenced this data with various reports on energy consumption in the United States.

Gigantic water and electricity consumption

For example, a 100-word response generated by AI requires a little more than a bottle of water. Adding up the millions of prompts tends to overheat data centers. To protect them, this heat must be transported out of the buildings via cooling towers. According to environmentalists quoted by the American media, data centers are often among the largest consumers of water in the cities where they are installed.

Concerning electricity, the example given is even more impressive. A single 100-word message from the chatbot is equivalent to the consumption of 14 LED bulbs lit for an hour.

This is particularly obvious in the state of Virginia, which has the largest concentration of data centers in the world. On site, citizens have mobilized to denounce this energy drain and these very noisy buildings that generate relatively few jobs, while destroying the value of their homes.

200% Deposit Bonus up to €3,000 180% First Deposit Bonus up to $20,000

Tech giants defend themselves

Criticized, the big technology companies indicate that they have taken measures to rectify the situation. For example, Craig Cincotta, CEO of Microsoft, declared that his company “remains committed to reducing the intensity with which we withdraw resources”. He adds that the Redmond firm is working on methods of cooling data centers that will “completely eliminate water consumption” .

Studies have already pointed to this danger

This is not the first time that the energy-consuming side of these chatbots has been pointed out. This is the case, for example, of Alex de Vries, an employee of the central bank of the Netherlands and creator of the Digiconomist website. This portal allows the general public to be measured and alerted about the energy impact of new technologies.

According to him, AI could follow the same trajectory as cryptocurrencies: “Quoted by the New Yorker, the specialist thus underlined: “I saw that it had a similar capacity and that it had the potential to have a similar growth trajectory in the years to come, and I felt immediately compelled to make sure that people were aware that it is also an energy-intensive technology.”.

Finally, let us mention this study carried out in 2023 by American scientists. They estimated that training GPT-3 in Microsoft data centers required 700,000 liters of fresh water, which is equivalent to the amount of water needed to cool a nuclear reactor.

Coincidence or not, the tech giants are eyeing this energy source and innovations to ensure the supply of their data centers. To find out more, don’t hesitate to consult our analysis here.

📍 To not miss any Presse-citron news, follow us on Google News and WhatsApp.

[ ]