AI for nuclear plants? ORNL supercomputer's new task is no sci-fi - it's a clean energy win

A nuclear artificial intelligence company will use the world's fastest supercomputer at Oak Ridge National Laboratory to train its AI tool, which could save the heavily regulated nuclear power sector millions of dollars and untold hours of work time.

Atomic Canyon, a California-based startup introducing AI to nuclear power plants, will use the Frontier supercomputer in partnership with ORNL.

The company says its tool will help nuclear power plant engineers and operators search through their own data and millions of documents from the Nuclear Regulatory Commission, widely considered the world's most rigorous nuclear oversight agency.

Its long-term goal is to create a generative tool like ChatGPT that could help nuclear companies apply for expensive construction and operating licenses, co-founder and CEO Trey Lauderdale said.

"How lucky are we? We get to use the world's fastest supercomputer to solve one of the biggest problems that society has ever faced, which is helping to enable nuclear power as a clean source of energy," Lauderdale told Knox News. "This is a dream come true."

Several AI developers have used Frontier, which is the fastest computer ever made, capable of more than a quintillion calculations per second. Frontier has topped the list of the world's fastest computers five times in a row and made Time's list of the 200 best innovations of 2023.

Atomic Canyon will train its AI-powered search tool on Frontier after being awarded time on the computer by the U.S. Department of Energy. Scientists around the world can apply to use Frontier's unprecedented processing power for free to solve a range of scientific problems, as long as the results are available to the public.

If Atomic Canyon had to pay for time on a system like Frontier, it would cost tens of millions of dollars, a central roadblock for other AI startups. The company is self-funded by Lauderdale and Kristian Kielhofner, co-founder and chief AI architect.

Safety is key as nuclear AI starts small

Like the AI sector itself, Atomic Canyon has moved rapidly since it was founded just six months ago. In March, the company launched Neutron, its first free search tool trained on complex nuclear terms.

Neutron sets itself apart from other AI search toll with its training on 52 million pages of documents publicly available from the Nuclear Regulatory Committee. Atomic Canyon's next tool with have expanded capabilities, thanks to Frontier.

The idea for integrating AI into the nuclear sector started when Lauderdale moved to San Luis Obispo, California, with his wife and kids and learned they would live just 10 miles away from Diablo Canyon, the last nuclear plant in the state.

After confronting his misconceptions about the safety of nuclear power, Lauderdale began meeting Diablo Canyon employees.

"Nuclear, by far ... is the most regulated space I've seen in my life," Lauderdale said. "Everything that happens in a nuclear power plant is documented. Every action that's taken, every change in process."

Lauderdale's previous company, Voalte, was the first to bring iPhones into hospitals for clinical communications, something health care professionals loyal to their BlackBerry phones once considered unthinkable. Some nuclear professionals see using advanced AI as unthinkable, too.

"AI has to prove itself to the nuclear industry. You have to earn your way up that kind of risk profile," Lauderdale said. "Searching through documents is an incredible place to start, but over time, we're going to see artificial intelligence in a safe, reliable way make its way throughout the entire nuclear supply chain."

AI and nuclear need each other, AI CEO says

As more advanced nuclear companies work through the long process of getting approval and existing power plants extend their lifetimes, Atomic Canyon will help them locate previous NRC approvals. This is similar to how a lawyer would search for court cases as precedent.

This could also be good news for AI, which requires power-intensive data centers.

"I don't know any other source of power that is going to give you reliable, safe, clean gigawatts of power outside of nuclear," Lauderdale said. "It is the clear choice."

The Diablo Canyon plant provides 17% of California's zero-carbon electricity and nearly 9% of the state's total electricity supply, according to Pacific Electric and Gas. In Tennessee, two Tennessee Valley Authority nuclear plants provide 45% of the state's electricity.

Atomic Canyon wants to be an integral part of what many have called the "nuclear renaissance," a resurgence of nuclear power as the most powerful source of clean energy and the clearest path to global net-zero energy goals. AI could help the industry overcome cost and time overruns.

The Knoxville and Oak Ridge region is a critical hub for nuclear technology with 150 nuclear companies and a history of groundbreaking nuclear innovation at Oak Ridge National Laboratory dating back to the Manhattan Project.

"I have never worked on something so exciting in my entire life," Lauderdale said. "I think nuclear power is the right technology to help us battle climate change, and I think artificial intelligence and leveraging AI in a safe, reliable efficient format will be an unlock to help our current nuclear power fleet run more effectively."

The U.S. joined more than 20 countries in a commitment to triple their nuclear fleets by 2050 at the COP28 climate summit in Dubai late last year. That requires the U.S. to go from 100 gigawatts to 300 gigawatts of nuclear power, equivalent to building 100 Hoover Dams.

Daniel Dassow is a growth and development reporter focused on technology and energy. Phone 423-637-0878. Email daniel.dassow@knoxnews.com.

Support strong local journalism by subscribing at knoxnews.com/subscribe. 

This article originally appeared on Knoxville News Sentinel: ORNL supercomputer Frontier to train nuclear AI tool