How the US Nuclear Plant Fleet Is Leaning Into AI

This year, nearly every one of the 94 nuclear reactors in the United States has been given the opportunity to incorporate artificial intelligence, with most of them accepting, reports IEEE Spectrum. In August, California-based Atomic Canyon introduced NIVA, the Nuclear Industry Virtual Assistant, developed in collaboration with nuclear industry groups.

The assistant was tested in plants owned by Constellation Energy and is now accessible to the entire fleet of reactors across the U.S. Additionally, AI-for-nuclear startup Nuclearn claims its products are being utilized by more than 65 U.S. partners, including numerous traditional plants and an integration with NuScale Power Corp. focused on advanced small nuclear reactors.

Microsoft and Nvidia have teamed up on a project applying AI to optimize every stage of a nuclear plant's life cycle, from site permitting to continuous operations. According to Nuclearn's CFO and co-founder, Jerrold Vincent, the rapid adoption is due to two main factors. Firstly, technology companies betting on AI are also interested in reviving nuclear power to supply electricity for the growing number of data centers.

Secondly, AI is well-suited to tackle the nuclear industry's most pressing issues. The intricate regulations and complex systems for management and maintenance are precisely the tasks AI excels at. With aging hardware, an aging workforce, and the need to handle paperwork, Vincent believes machine intelligence is crucial for the industry's revival.

Currently, aside from electricity, the primary output of a nuclear reactor is paperwork. Nuclear plants generate vast amounts of documentation to prove regulatory compliance and record responses to any issues, from sidewalk cracks to reactor faults. Nuclearn's AI tools assist plants in efficiently searching through extensive data repositories to quickly file paperwork with federal authorities or determine past solutions to specific problems.

Vincent compares AI in the nuclear sector to ChatGPT, emphasizing its potential to streamline operations. However, both the Nuclearn and NIVA large language models currently report to humans and do not interact directly with power plant operations.