2024-07-11 16:01:11
Los language models on which they are based IA Generative algorithms require enormous computing power to train using thousands of millions of datas, which requires powerful servers.
Then, every time a user sends a query to ChatGPT or any other generative AI, it spins up servers located in a data center.
These servers consume electricitythey heat up and need to be cooled with systems that in turn require energy.
According to International Energy Agency (IEA), data centers typically use about 40% of their electricity to power servers and 40% to cool them. Several studies indicate that a survey of ChatGPT requires on average 10 times more energy than a simple Google search engine query.
The rise of AI since 2022 led to giants of Internetas Amazon, Google y Microsoftto invest massively in the creation of data centers around the world.
In its environmental report, Google, the internet giant, internet,highlights the increase in energy consumption in its data centres, as well as the rise in emissions linked to the construction of new data centres and the modernisation of existing ones.
How much energy does AI consume?
Before the AI rush, Data centers accounted for around 1% of global electricity consumption, according to the IEA.
If you add AI and the cryptocurrency sector, the data centers consumed almost 460 TWh of electricity in 2022, or 2% of total global production, according to the institution.
This figure could double in 2026 to reach 1,000 Twh, which would be equivalent to the electricity consumption of Japan, warns in a report.
Alex de Vries, economist at the Free University of Amsterdam, developed a model of the electricity consumption needed for artificial intelligence alone based on the company’s sales projections Nvidia, whose processors are essential for training AI models.
If Nvidia’s sales estimates for 2023 are correct, and if all servers are running at full capacity, they could consume between 85.4 and 134 Twh per year, what a country like Argentina, write in an article.
“The figures I put in my article were quite cautious at first because they don’t take into account processes such as refrigeration needs,” he said.
Nvidia’s sales beat its forecasts last year, so the numbers could be even higher, he added.
Related
2024-07-11 16:01:11