The OpenAI app icon displayed together with different AI functions on a smartphone.
Jonathan Raa | Nurphoto through Getty Photographs
Know-how giants are turning to nuclear power to energy the energy-intensive knowledge facilities wanted to coach and run the huge synthetic intelligence fashions behind right this moment’s generative AI functions.
Microsoft and Google are among the many companies agreeing offers to buy nuclear energy from sure suppliers within the U.S. to convey extra power capability on-line for its knowledge facilities.
This week, Google stated it will buy energy from Kairos Energy, a developer of small modular reactors, to assist “ship on the progress of AI.”
“The grid wants these sorts of unpolluted, dependable sources of power that may help the construct out of those applied sciences,” Michael Terrell, senior director for power and local weather at Google, stated on a name with reporters Monday.
“We really feel like nuclear can play an vital position in serving to to satisfy our demand, and serving to meet our demand cleanly, in a approach that is extra across the clock.”
Google stated its first nuclear reactor from Kairos Energy can be on-line by 2030, with extra reactors going reside by way of 2035.
The tech large is not the one agency seeking to nuclear energy to understand its AI ambitions. Final month, Microsoft signed a cope with U.S. power agency Constellation to resurrect a defunct reactor on the Three Mile Island nuclear energy plant in Pennsylvania, whose reactor has been dormant for 5 years.
The Three Mile Island plant was the placement of essentially the most severe nuclear meltdown and radiation leak in U.S. historical past in March 1979, when the lack of water coolant by way of a defective valve precipitated a reactor to overheat.
Why they’re turning to nuclear
Tech firms are below strain to seek out power sources to energy knowledge facilities — a key piece of infrastructure behind modern-day cloud computing and AI functions.
Many builders lease out servers outfitted with GPUs (graphics processing models), which might usually be too costly to personal outright, from so-called cloud “hyperscalers” — corresponding to Amazon, Microsoft and Google.
These tech giants have benefited from a surge of curiosity in generative AI functions corresponding to OpenAI’s ChatGPT. However that improve in demand has additionally led to an unintended impact: correspondingly giant spikes within the quantity of power required.
International electrical energy consumption from knowledge facilities, synthetic intelligence and the cryptocurrency sector is anticipated to double from an estimated 460 terawatt-hours (TWh) in 2022 to greater than 1,000 TWh in 2026, in line with a analysis report from the Worldwide Vitality Company.
Researchers on the College of California, Riverside, revealed a research in April final 12 months that discovered ChatGPT consumes 500 milliliters of water for each 10 to 50 prompts, relying on when and the place the AI mannequin is deployed. That equates to roughly the quantity of water in a typical 16-ounce bottle.
As of August, there have been greater than 200 million folks submitting questions on OpenAI’s well-liked chatbot ChatGPT each week, in line with OpenAI. That is double the 100 million weekly energetic customers OpenAI reported final November.
Environmental opposition
Nuclear power is not with out its controversy. Many local weather activists oppose such provides, citing their hazardous environmental and security dangers, and the truth that they don’t supply a real supply of renewable energy.
“Nuclear energy is extremely costly, hazardous and gradual to construct,” the local weather charity Greenpeace says on its web site.
“It’s sometimes called ‘clear’ power as a result of it would not produce carbon dioxide or different greenhouse gases when electrical energy is generated however the actuality is that it’s not a believable different to renewable power sources.”
Proponents of nuclear power, however, say that it provides an almost carbon-free type of electrical energy and is extra dependable than renewable sources like photo voltaic and wind.
“Whether it is constructed and securitized in the proper approach, I do suppose nuclear is the long run,” Rosanne Kincaid-Smith, chief working officer of Northern Knowledge Group, a world knowledge heart supplier, informed CNBC at a tech convention in London final week.
“Individuals are petrified of nuclear due to the disasters we have had previously. However what’s coming, I simply do not see conventional grids being the sustainable energy that is ongoing within the improvement of AI,” Kincaid-Smith added.
Whereas Northern Knowledge Group is not utilizing nuclear power — neither is it actively exploring plans to make use of nuclear as an influence supply for its AI knowledge facilities — the agency does need to “contribute to that dialog as a result of it is vital for the broader ecosystem, the broader financial system,” Kincaid-Smith informed CNBC.
– CNBC’s Pippa Stevens contributed to this report