But how much electricity does artificial intelligence consume?
According to Musk, artificial intelligence will soon surpass human capabilities, provided there is enough electricity to power data centers. The question of the (large) energy consumption of this technology is crucial: here's what we know
Elon Musk thinks that within just over a year new artificial intelligence models will surpass human intelligence, provided, however, that there is sufficient availability of electricity and components to power these technologies.
During an interview with Nicolai Tangen – CEO of Norges Bank Investment Management, the investment arm of the Norwegian central bank -, Musk declared that “towards the end of next year we will have an artificial intelligence that is more intelligent than any human being ”. According to him, within the next five years the capabilities of artificial intelligence will have surpassed those of humanity as a whole.
DO MUSK AND ALTMAN THINK THE SAME WAY?
As the Financial Times points out, other of Musk's predictions – for example about the spread of self-driving vehicles and the landing of a rocket on Mars: he owns companies active in both sectors – have not come to pass. However, the billionaire raised an interesting question, that of the large electricity needs of the data centers that power artificial intelligence models. An issue also addressed by Sam Altman, the CEO of OpenAI ( there is no bad blood between him and Musk ), according to whom in the future artificial intelligence will consume so much electricity as to require an energy breakthrough, such as nuclear fusion : Altman himself is an investor in Helion, a US nuclear fusion startup.
THE ELECTRICITY CONSUMPTION OF ARTIFICIAL INTELLIGENCE
According to the International Energy Agency, data centers, artificial intelligence and cryptocurrencies accounted for nearly 2 percent of global electricity demand in 2022, but this could double by 2026 and approach Japan's electricity consumption . Again according to the Agency's estimates, training an artificial intelligence model consumes more electricity than a hundred homes in a year.
A study reviewed by The Verge states that one hundred thousand NVIDIA HGX A100 server platforms at full capacity consume between 5.7 and 8.9 terawatt hours of electricity per year.
THE EMISSIONS
On top of this, there is the climate problem. Data centers powered by electricity generated from fossil fuels contribute to greenhouse gas emissions: Bloomberg wrote that data centers and electricity transmission networks annually release as much CO2 as Brazil. If desired, it would be possible to replace fossil fuels with clean ones, but artificial intelligence systems are growing at a much faster rate than the construction and connection times of clean energy plants.
DO I NEED GAS?
Furthermore, photovoltaic and wind farms generate electricity intermittently (depending on weather conditions, that is) and therefore may not be suitable to meet the massive and constant demand of data centers. For this reason, Mike Sommers – he is the CEO of the American Petroleum Institute, the US association representing the oil & gas sector – told Axios that artificial intelligence infrastructures “will need active energy twenty-four hours a day twenty-four. And that means we're going to need a lot more natural gas."
This is a machine translation from Italian language of a post published on Start Magazine at the URL https://www.startmag.it/innovazione/intelligenza-artificiale-consumo-energia-elettrica/ on Sun, 14 Apr 2024 05:53:15 +0000.