RTS Nuclear + AI Briefing, Issue 03 · August 2026
In August 2026, generative AI moved out of the pilot deck and into nuclear operations, licensing, and dispatch. An AI assistant grounded in NRC guidance shipped into operating fleets, TerraPower made its storage system the pitch to hyperscalers, the Department of Energy put a 10x target on AI-assisted licensing, and NuScale put nuclear-specific AI to work on its own engineering.
The common thread is where the technology now sits: inside regulated workflows that a control room, a licensing team, and a dispatcher have to defend. That is exactly where traceability stops being optional.
Atomic Canyon puts an AI assistant inside nuclear operations
Atomic Canyon released NIVA, a Nuclear Industry Virtual Assistant that lets plant staff query operational records, NRC guidance, and technical standards in plain language. It was developed with EPRI, INPO, and NEI, and runs on the Neutron platform and FERMI models trained on Oak Ridge’s Frontier supercomputer. Constellation Energy is among the first users.
The RTS perspective
The value is not the chatbot. NIVA is grounded in NRC guides, licensing basis documents, and INPO operating experience, the same corpus a control room has to defend. An assistant that cites the source record turns “the AI said so” into a traceable answer an inspector can follow.
The open question is configuration management: who owns the model’s knowledge cutoff, how updates get validated, and how a wrong answer is caught before it reaches a procedure. The assistant retrieves and drafts. A qualified human still signs. That is Human in Command®.
Source: Neutron Bytes
TerraPower makes molten-salt storage its pitch to data centers
TerraPower’s Natrium plant pairs a 345 MW sodium-cooled fast reactor with molten-salt thermal storage. The reactor runs flat while stored heat boosts turbine output when demand spikes. Meta committed to eight Natrium plants in January 2026, and the first plant in Kemmerer, Wyoming received NRC construction approval in March.
The RTS perspective
The storage is the real story, and it is an operations story more than a reactor story. Holding the reactor flat while shaping output to a swinging AI load is what makes nuclear look dispatchable to a hyperscaler. Schedule realism still governs: today’s data center deals are options on late-decade capacity, not power now. Buyers will want contracts that guarantee storage performance, and auditors will ask how the storage controls tie back to the licensing basis.
Source: TechCrunch
DOE’s Project Prometheus puts AI on the licensing path
The Department of Energy launched Project Prometheus, a $60 million, three-year program to apply AI across nuclear design, licensing, manufacturing, and operations, with more than $200 million in industry cost-share. Idaho National Laboratory leads a 32-partner consortium including X-energy, TerraPower, Oklo, NVIDIA, AWS, and Microsoft. The targets: 10x faster design and licensing, 50 percent lower deployment timelines and cost, and 3x faster manufacturing and fuel fabrication.
The RTS perspective
Those are the right targets, and the ones the NRC will scrutinize hardest, because a faster application is worthless if the reviewer cannot trust how it was produced. Every generated claim has to trace to a qualified basis and a named reviewer. Build the evidence trail now, so that when these tools reach production the workflow already shows what the model saw, what it wrote, and who approved it.
Source: Neutron Bytes
NuScale puts nuclear-specific AI to work with NPX and Nuclearn
NuScale Power formed a strategic alliance with Nuclear Promise X (NPX) and Nuclearn to deploy nuclear-specific AI across engineering and knowledge management. In a proof of concept, the AI cut the time to find relevant information by as much as 80 percent, with traceability back to controlled sources.
The RTS perspective
The 80 percent number will get the headlines, but the operative phrase is traceability back to controlled sources. A reactor designer lives and dies by configuration control. An AI that retrieves from the current, approved engineering basis is useful. One that cannot prove which revision it read is a liability.
Source: NuScale Power
Notes from NECX 2026 in Dallas
RTS Founder and Chief Nuclear Officer Sam Rieck attended the Nuclear Energy Conference and Expo in Dallas, August 24 to 27, hosted by ANS and NEI. Three takeaways:
- Regulatory modernization has moved from talk to practice. Licensing reform is now a working agenda, and teams building defensible evidence trails today will move through it fastest.
- AI is consolidating around shared, governed tools. Sessions on software qualification and the NIVA project showed the industry converging on qualified, auditable AI rather than scattered pilots.
- Delivery and workforce are the binding constraints. Schedule and skilled trades, not reactor physics, will decide who deploys on time.
What this signals
- AI has moved from the pitch deck into operations, licensing, and dispatch, where it faces real inspection.
- Near-term demand sits in the operating fleet, where document volume is a daily cost.
- Data center deals signed today are options on late-decade capacity.
- The NRC will judge AI-assisted work by its evidence trail, not its speed.
- The fuel cycle and safeguards are drawing federal capital, so AI advantage has to reach beyond the reactor.
The measure is not how fast the model runs. It is whether the result holds up under review.
Talk to RTS
One conversation. No pitch deck. Schedule a roadmap session or browse RTS research.
Comments are closed