Two years ago, Trey Lauderdale asked not what nuclear power could do for artificial intelligence, but what artificial intelligence could do for nuclear power.
The value of atomic power stations to provide the constant, zero-carbon electricity many data centers demand was well understood. What large language models could do to make building and operating reactors easier was less obvious. His startup, Atomic Canyon, made a first attempt at answering that by creating a program that could make the mountains of paper documents at the Diablo Canyon nuclear plant, California’s only remaining station, searchable. But Lauderdale was thinking bigger.
In September, Atomic Canyon inked a deal with the Idaho National Laboratory to start devising industry standards to test the capacity of AI software for nuclear projects, in much the same way each update to ChatGPT or Perplexity is benchmarked by the program’s ability to complete bar exams or medical tests. Now, the company’s effort is going global.
On Wednesday, Atomic Canyon is set to announce a partnership with the United Nations International Atomic Energy Agency to begin cataloging the United Nations nuclear watchdog’s data and laying the groundwork for global standards of how AI software can be used in the industry.
“We’re going to start building proof of concepts and models together, and we’re going to build a framework of what the opportunities and use cases are for AI,” Lauderdale, Atomic Canyon’s chief executive, told me on a call from his hotel room in Vienna, Austria, where the IAEA is headquartered.
The memorandum of understanding between the company and the UN agency is at an early stage, so it’s as yet unclear what international standards or guidelines could look like.
In the U.S., Atomic Canyon began making inroads earlier this year with a project backed by the Institute of Nuclear Power Operators, the Nuclear Energy Institute, and the Electric Power Research Institute to create a virtual assistant for nuclear workers.
Atomic Canyon isn’t the only company applying AI to nuclear power. Last month, nuclear giant Westinghouse unveiled new software it’s designing with Google to calculate ways to bring down the cost of key components in reactors by millions of dollars. The Nuclear Company, a startup developer that’s aiming to build fleets of reactors based on existing designs, announced a deal with the software behemoth Palantir to craft the software equivalent of what the companies described as an “Iron Man suit,” able to swiftly pull up regulatory and blueprint details for the engineers tasked with building new atomic power stations.
Lauderdale doesn’t see that as competition.
“All of that, I view as complementary,” he said.
“There is so much wood to chop in the nuclear power space, the amount of work from an administrative perspective regarding every inch of the nuclear supply chain, from how we design reactors to how we license reactors, how we regulate to how we do environmental reviews, how we construct them to how we maintain,” he added. “Every aspect of the nuclear power life cycle is going to be transformed. There’s no way one company alone could come in and say, we have a magical approach. We’re going to need multiple players.”
That Atomic Canyon is making inroads at the IAEA has the potential to significantly broaden the company’s reach. Unlike other energy sources, nuclear power is uniquely subject to international oversight as part of global efforts to prevent civilian atomic energy from bleeding over into weapons production.
The IAEA’s bylaws award particular agenda-setting powers to whatever country has the largest fleet of nuclear reactors. In the nearly seven decades since the agency’s founding, that nation has been the U.S. As such, the 30 other countries with nuclear power have largely aligned their regulations and approaches to the ones standardized in Washington. When the U.S. artificially capped the enrichment levels of traditional reactor fuel at 5%, for example, the rest of the world followed.
That could soon change, however, as China’s breakneck deployment of new reactors looks poised to vault the country ahead of the U.S. sometime in the next decade. It wouldn’t just be a symbolic milestone. China’s emergence as the world’s preeminent nuclear-powered nation would likely come with Beijing’s increased influence over other countries’ atomic energy programs. As it is, China is preparing to start exporting its reactors overseas.
The role electricity demand from the data centers powering the AI boom has played in spurring calls for new reactors is undeniable. But if AI turns out to have as big an impact on nuclear operations as Lauderdale predicts, an American company helping to establish the global guidelines could help cement U.S. influence over a potentially major new factor in how the industry works for years, if not decades to come.
A chart showing the trio of South America nations' share of next year's oil production. EIA
