According to Professor Shaolei Ren, a scientist from the University of California in Riverside, the increase in microsoft water consumption is due to “significant investments in the generative AI …”
In their latest environmental responsibility reports, Microsoft and Google reported water consumption pea last year. In question: generative artificial intelligence technology (AI).
6.4 billion liters of water, or 2,500 Olympic pools: it is Microsoft’s water consumption in 2022, an increase of 34% compared to 2021.
At Google, the increase is 20% over the same period, according to the environmental responsibility reports of these firms, cited by journalists from the Associated Press Agency (AP).
What pushes tech giants to “swallow” so much water? Professor Shaolei renters this “to significant investments in the generative AI”.
Concretely, ultra-powerful “superordinators” learn to imitate human writing: we speak of “the development of a linguistic model”. The AI must then train by identifying regularities in a large set of texts (corpus) written by man.
However, these analyzes (or “calculations”) consume electricity and generate heat. To cool the superordinators, the data centers have “cooling towers”.
When the outside temperature is low or moderate, the ventilation is enough. But when the thermometer displays more than 29.3 ° C outside, it is necessary to pump water.
In an article which should be published in the course of the year, the Professor RENSE team estimates that Chatgpt engulfs 500 milliliters of water – the equivalent of a small bottle of water of 50 cl – each time you ask him a series of 5 to 50 questions.
“This fork varies according to the location of its servers and the season”, and the estimate takes into account the cooling of data centers, but also indirect water consumption that companies do not measure – for example to cool the power plants which feed the superordinators with electricity.
In response to questions about the preservation of water resources, Microsoft said it was investing in research to measure the EIA energy imprint and carbon footprint “while working on means of making major systems more effective, both in terms of training and application”.
