Summarize this content to 2000 words in 6 paragraphs in Arabic Stay informed with free updatesSimply sign up to the Artificial intelligence myFT Digest — delivered directly to your inbox.Informed readers already know that lots of energy will be needed to fuel hoped-for progress in machine learning, large language models and various other stuff that tech-industry salesmen classify as “artificial intelligence”. Barclays has a nice summary out today about AI’s energy demand: Most of the world’s more than 11,000 registered data centres are not yet involved in any kind of AI-related activity. Combined, their consumption of electricity (excluding cryptocurrencies) is about 1.0%-1.5% of the world’s total, according to a mid-2024 report from the IEA2.However, data centres’ energy demands could change dramatically in coming years, thanks to the dissemination of AI. According to a June 2024 analysis from Barclays Research — which uses a bottom-up approach based on utilities’ forward-looking supply contracts — annual demand to power data centres in the US could grow by a range of 14%-21% every year through 2030. That would imply US data-centre demand roughly tripling by 2030, from 150-175 terawatt hours (TWh) in 2023 to as much as 560 TWh — equivalent to 13% of current US electricity demand.It’s notable that the data-centre boom came at a very convenient time for the commercial real estate market, but that’s perhaps a different story. From the bank: Headlines out of Virginia have already shown how “bottlenecks” can arise in energy supply, especially as data-centre construction tends to cluster in certain crucial regions. And the only energy source that really meets the demand? It seems that’s mainly nuclear energy: If the internet has to be flooded with AI slop, we might as well get some clean energy out of it.
