Energy: AI requires so much energy that there are plans to use nuclear reactors
It is no secret that operating an AI data center, for the use of artificial intelligence, requires an immense amount of energy. Scientific American calculated that data centers would need 84.5 terawatts/h of electricity to use AI in 2023, more than a small nation uses. The situation will get worse with the introduction of AI for video creation, such as Sora.
To meet these enormous energy demands, experts are looking for alternative sources, as the BBC reports, including small nuclear reactors that could power individual data centers.
“Our industry needs to find another source of energy,” said Chris Sharp, CTO of Digital Realty.
Small and modular
For years, scientists have been developing small modular reactors ( SMRs ), which are small-scale power plants that can provide energy on site and thus dramatically reduce companies' dependence on the grid.
Despite a thriving industry dedicated to making them, there are still no commercial reactors in operation anywhere in the world, as the BBC notes .
This is an intriguing idea that has inspired some of the biggest players in the field of artificial intelligence to invest in this idea. Last year, job listings suggested that Microsoft was looking to roll out its own plan for SMRs, with the aim of using them to power its own AI data centers.
OpenAI CEO Sam Altman has also invested in a nuclear startup called Oklo, which is working on self-tuned SMRs. (Altman also suggested that AI is so energy-hungry that it requires an innovative power source.)
“Data centers are already power-hungry, but with AI we are reaching a new level of power requirements,” Michael Bluck, director of the Nuclear Engineering Center at Imperial College London, told the BBC.
The problem is that, although the solution appears logical, it is not immediately applicable, because, at this moment, there are none functioning SMRs in the West.
“ There are about 50 SMR designs out there,” Bluck added. “ The challenge is to build them in repeatable, factory-style units by standardizing production lines .”
A big obstacle to overcome is regulation. After all, nuclear energy carries some obvious risks. The U.S. Nuclear Regulatory Commission has been “ engaged in varying degrees of pre-application activities with several SMR designers in recent years,” it says on its official website.
In 2022, the US Nuclear Energy Commission approved the first SMR design, developed by an Oregon startup called NuScale Power. Two other projects are in an advanced state of authorization.
However, despite the clear momentum behind the idea, it is unclear whether SMRs will be the answer to our rapidly growing energy needs. Companies are desperate for ways to scale operations now, not in a few years. So, if this option is to become realistic, it had better be realized soon.
Thanks to our Telegram channel you can stay updated on the publication of new Economic Scenarios articles.
The article Energy: AI requires so much energy that nuclear reactors are being considered comes from Economic Scenarios .
This is a machine translation of a post published on Scenari Economici at the URL https://scenarieconomici.it/energia-la-ai-richiede-cosi-tanta-energia-che-si-pensa-di-utilizzare-i-reattori-nucleari/ on Sat, 17 Feb 2024 16:51:26 +0000.