Glonce

Atomic Canyon's NIVA and Nuclearn bring AI assistants to U.S. nuclear plants

productsThe Glonce editorsIEEE Spectrum🔥55Glonce rating55Impact45Novelty45Relevance85Surprise40Players40Credibility70
Atomic Canyon's NIVA and Nuclearn bring AI assistants to U.S. nuclear plants

An IEEE Spectrum feature by Andrew Moseman describes how AI assistants are spreading through the U.S. nuclear fleet. Nuclear power is historically cautious and risk averse, yet this year nearly the entire fleet of 94 U.S. reactors has been offered the chance to integrate AI into its operations, and most have taken it. The article stresses that the AI stays away from plant operations: the Nuclearn and NIVA large language models used in the nuclear world report to humans and do not go anywhere near power plant operations.

The main products are two. In August, California-based Atomic Canyon launched NIVA, the Nuclear Industry Virtual Assistant, developed in cooperation with nuclear-industry groups. It was pilot-tested in plants run by Constellation Energy, which declined to be interviewed, and is now available to the entire U.S. reactor fleet. Separately, the startup Nuclearn says its products have gone to work with more than 65 U.S. partners, a total that includes many traditional plants and an integration with NuScale Power Corp., which is working toward advanced small reactors. The article also notes that Microsoft and Nvidia have collaborated on a project to apply AI to the entire life cycle of a nuclear plant, spanning site permitting, design, construction and continuous operations.

Nuclearn CFO and cofounder Jerrold Vincent gives two reasons for the rapid uptake. Technology companies banking their futures on AI also want a nuclear revival to power their data centers. And AI suits the industry's problems: intricate regulations and complex management and maintenance systems are the kind of tasks AI is good at, while the industry is staggering under aging hardware and an aging workforce. Vincent says customers arrive saying they have jobs open that they cannot fill, with regulatory work still to do.

Much of the work is paperwork. Plants produce piles of documentation to show regulatory compliance and to record how they handle any problem, from a crack in the sidewalk to a reactor fault. Nuclearn's products help plants search large data repositories to file paperwork with the federal government faster or find how a problem was solved before. Vincent: "If you think of it as ChatGPT for nuclear, it's actually not a bad place to start."

NIVA's origin, per Rob Austin, the Electric Power Research Institute's leader for the NIVA project: it started a year ago when Constellation chairman Joseph Dominguez challenged the nonprofit industry groups that hold information on regulatory compliance, maintenance, repair and mechanical history. He wanted that knowledge available with the convenience of ChatGPT, Copilot or Gemini, but secure. Three of the biggest such groups (the Institute of Nuclear Power Operations, EPRI and the Nuclear Energy Institute) moved in 2025 to create unified datasets, since information across the industry was scattered and not standardized. The task fell to Atomic Canyon, whose CEO Trey Lauderdale previously worked on AI in health care. After winning the NIVA contract in 2025, the company first downloaded 53 million pages of publicly available Nuclear Regulatory Commission data to train the model on nuclear operations, vernacular and regulations, then partnered with Oak Ridge National Laboratory to build the first nuclear-specific AI models. Lauderdale calls them sentence-embedding models, "a fancy way of saying, we taught AI how to seek the nuclear language."

Austin's example is a house-size, thousand-horsepower vertical cooling pump. Engineers can read EPRI maintenance guides and tens of thousands of incident reports held by the Institute of Nuclear Power Operations. The AI synthesizes those reports into a guide to the main pump failures and, crucially, shows its work so engineers can find the documentation behind its recommendations.

Data secrecy is a major hurdle. Nuclear data is largely proprietary or protected by federal law. Nuclearn offers only single-location solutions so a plant's data never leaves the site. NIVA can only search databases that already required plant-level security clearances, Austin says, so the same rules apply to using the assistant. The article adds that AI systems that chiefly help with documentation and data search are not subject to the strictest rules, which would apply if they interacted with components that control the plant.

Asked whether AI could one day move closer to operations, Vincent and Austin say that is far away, though they can envision a path. Austin points to welds, all of which must currently be inspected by a human. An EPRI pilot project is testing whether AI can look at the mountains of inspection data and flag areas of concern. A human reviews all of the AI's recommendations, and no power plant is allowing AI to take over its inspections. The author calls this a step toward using AI on direct plant operations. Lauderdale is blunter: he cannot see, in the short or long term, AI making decisions around plant operation, given the regulatory and risk nature, and expects humans in the loop for a very, very long time.

Key facts

  • Nearly the entire fleet of 94 U.S. nuclear reactors has been offered AI this year, and most have taken it; the source gives no exact count of adopters.
  • Atomic Canyon launched NIVA, the Nuclear Industry Virtual Assistant, in August; it was piloted at Constellation Energy plants and is now available to the whole U.S. fleet.
  • Nuclearn says its products work with more than 65 U.S. partners, including an integration with NuScale Power Corp.
  • NIVA's model was trained starting with 53 million pages of public Nuclear Regulatory Commission data, and Atomic Canyon partnered with Oak Ridge National Laboratory on nuclear-specific models.
  • The tools handle documentation and data search, not plant operations; an EPRI weld-inspection pilot keeps a human reviewing all AI recommendations.

Why it matters

Nuclear power is one of the most risk averse industries, so uptake across most of a 94-reactor fleet is notable in itself. The article ties it to a workforce problem: decades of stalled construction meant fewer people entered nuclear careers, leaving a skilled but quickly shrinking workforce asked to keep old reactors running and possibly start new ones as demand for electricity surges. Lauderdale argues the nuclear revival cannot happen without augmenting that knowledge-based workforce. There is also a loop with AI itself: tech companies building on AI want nuclear power for their data centers, and the industry says it needs AI to thrive.

Who it affects

Nuclear plant staff who file regulatory paperwork and troubleshoot equipment are the direct users. The industry groups INPO, EPRI and NEI supply the data behind NIVA. Utilities such as Constellation Energy and advanced-reactor developers such as NuScale Power Corp. are among the named customers or partners. Vincent says demand from advanced-reactor developers and new nuclear is growing while the existing workforce is spread thin.

How to use it

NIVA is now available to the entire U.S. reactor fleet. Workers write a query and the AI sorts through accumulated institutional knowledge, covering anything from filling out regulatory forms to procuring the right tools and materials. Nuclearn's Equipment AI lets engineers query equipment operations and returns answers drawing on decades of industry experience. Access follows existing rules: NIVA only searches databases that already required plant-level security clearances. The source gives no price or licence terms.

How solid is it

The article is a feature in IEEE Spectrum by Andrew Moseman, online communications editor at Caltech and a freelance contributor to IEEE Spectrum. Most of the substance comes from interviews with Vincent of Nuclearn, Austin of EPRI and Lauderdale of Atomic Canyon, all with a stake in the tools. The more-than-65 partners figure is Nuclearn's own claim, and the source does not name those partners except NuScale Power Corp. The source does not give the number of reactors that actually adopted AI; it says only that most have. Constellation declined to be interviewed. No accuracy, error-rate, time-saving or cost figures are given for NIVA or Nuclearn, and no regulator is quoted on AI use in plants.

Risks and caveats

Data secrecy is a major hurdle: nuclear data is largely proprietary or protected by federal law, which is why Nuclearn only offers single-location solutions and NIVA is limited to databases already behind security clearances. The article notes that AI systems chiefly helping with documentation and data search are not subject to the strictest rules, which would apply if they touched components that control the plant. The author says the EPRI weld-inspection pilot marks a step toward direct plant operations, but the people interviewed say that is far away, a human reviews all AI recommendations, and no plant is allowing AI to take over its inspections. The source does not say that any AI controls or makes decisions about reactor operations; it says the opposite. Public opinion of nuclear power, which soured decades ago and has only recently begun to recover, is described as another hurdle.

“So, I believe we’re going to have humans in the loop for a very, very long time.”

— Trey Lauderdale, CEO of Atomic Canyon