Companies that have dug deep into AI have fallen in love with nuclear power for its promise of 24/7 electricity. Meta, Google, and Microsoft have all made deals with startups or reactor operators. But does the nuclear industry love AI back?

Yes, with caveats.

No one is proposing to let an AI run a reactor, but power companies are increasingly interested in the technology’s potential to tighten things up on the business side, Bradley Fox, co-founder and CEO of Nuclearntold Read.

Fox and Jerrold Vincent started Nuclearn to capitalize on that interest. The company says its AI tools are being used in more than 65 nuclear reactors around the world.

It recently raised a $10.5 million Series A round led by Blue Bear Capital with participation from AZ-VC, Nucleation Capital, and SJF Ventures.

Nuclearn got its start when the founders were working at the Palo Verde Nuclear Generating Station just west of Phoenix. They had been experimenting with ways to streamline various repetitive tasks first from a data science perspective then with more advanced AI models.

Soon, other reactors took note, Fox said. “Can you help us do the same thing you’re doing for Palo Verde but for my plant?” they asked him.

That interest coincided with the COVID pandemic. “We both were kind of bored after work,” Fox said. “We’re like, hey, let’s work on a startup.”

Nuclearn has developed models trained on nuclear industry-specific terminology. The startup can train custom models for utilities and power providers that request it, and while its software runs in the cloud, it can also help reactors set up hardware on site if their security protocols require it.

The startup’s software can generate routine documentation that reactor employees then review and sign off on.

“Most AI in the industry now, the (Nuclear Regulatory Commission) considers it a tool. It’s the same way as if you’re going to use Excel or Mathematica or some type of engineering software,” Fox said. “Liability always falls with a person.”

Reactor operators can set thresholds for how much gets automated depending on their level of comfort and their confidence in how well the model can tackle the problem.

“If the model doesn’t know or if we’re unsure, based on the setting you select, it’ll send it back to the right people and get a double check,” Fox said. “We tell the customers, ‘Think of this as the junior employee.’”