Episode Summary
Executive Summary: Trey Lauderdale, CEO of Atomic Canyon, explains how his healthcare-tech background led him to build AI tools for nuclear power. Atomic Canyon’s first commercial on-site generative AI deployment at Diablo Canyon focuses on document search and workflow support, not operating reactors. The conversation covers why nuclear needs AI to scale, the company’s on-prem security architecture, partnerships with national labs and industry groups, and the startup’s broader roadmap for licensing, construction, and operations.
Main Topics: Atomic Canyon’s mission and product focus (Priority: 5/5): Atomic Canyon is building generative AI tools for nuclear power plants to improve efficiency, document retrieval, and workflow support while staying away from direct reactor control. Trey Lauderdale’s entrepreneurial path (Priority: 5/5): Lauderdale describes his transition from healthcare IT/digital health into nuclear, drawing parallels between highly regulated industries and the opportunity to modernize archaic workflows. Why nuclear needs AI (Priority: 5/5): The discussion centers on administrative burden, workforce shortages, and document-heavy processes in nuclear as reasons AI can materially improve productivity and scalability. Technical architecture and security (Priority: 5/5): The company’s solution is fully on-premise, using nuclear-specific embedding models and local H100 hardware, with strict security and permission controls to satisfy plant requirements. Initial deployment at Diablo Canyon (Priority: 5/5): Atomic Canyon’s first product solved document search for Diablo Canyon, where employees needed a better way to find and summarize information across fragmented legacy systems. Benchmarks, expansion, and future roadmap (Priority: 4/5): Atomic Canyon is working with Idaho National Laboratory on Nuclear Bench to establish AI performance benchmarks for nuclear workflows and expand into licensing, construction, and international markets. Funding, hiring, and market opportunity (Priority: 4/5): The company recently raised a $7 million seed round and is actively hiring AI/software engineers, positioning itself to scale across the U.S. fleet and beyond.
Key Arguments: Nuclear power and AI are mutually reinforcing: AI needs massive clean power, and nuclear needs AI to reduce costs, improve workflows, and scale faster. The best near-term nuclear AI use case is not autonomous reactor operation but document search and knowledge retrieval, because it solves a real operational bottleneck without introducing undue risk. Off-the-shelf LLMs are not reliable enough for nuclear data because nuclear terminology, acronyms, and document formats are highly specialized and can cause hallucinations. A fully on-prem architecture is essential in nuclear because plant data is sensitive, export-controlled, and subject to strict security and permissioning requirements. AI can help nuclear scale by streamlining licensing, construction, and operational workflows, where paperwork and administrative burden are enormous. Nuclear’s current workforce shortage means the industry will need AI augmentation to build and operate new reactors at the pace required by rising electricity demand. Atomic Canyon’s credibility comes from executing a concrete deployment at Diablo Canyon and then using that success to open broader industry opportunities. A fresh outsider perspective can be an advantage in a conservative industry, provided the company quickly hires nuclear experts and respects domain constraints.
Data Points: Company age: About 2 years old - Trey says Atomic Canyon is just coming up on its two-year anniversary. Prior healthcare deployment scale: Hundreds of thousands of nurses and doctors - Volt, Lauderdale’s earlier company, was deployed broadly in hospitals. NRC document corpus: 53 million pages - Atomic Canyon downloaded NRC public documents to train nuclear-specific models. Oak Ridge compute access: 20,000 GPU node hours - Oak Ridge National Laboratory provided compute for model development. Model performance improvement: 40% better - Atomic Canyon says its sentence embedding models outperformed off-the-shelf search embeddings by this margin in benchmarks with Oak Ridge. Diablo Canyon staffing: 1,300 head-of-household jobs - Lauderdale cites the plant’s local economic importance in San Luis Obispo. Diablo Canyon power output: 2 gigawatts - He describes the plant as producing roughly 2 GW of power continuously. California power share: 10% - Lauderdale says Diablo Canyon supplies about 10% of California’s power. U.S. nuclear fleet: 54 plants / 94 reactors - He references the current U.S. reactor fleet when discussing market size. International expansion opportunity: Roughly 200 reactors - Lauderdale estimates the addressable international market for current software. Recent seed round: $7 million - Atomic Canyon closed a $7M seed round in May. Hiring growth: Engineering team expected to at least double in the next year - He says hiring is focused on back-end, full-stack, and front-end developers.
Pivotal Quotes: "We do not say AI should run a nuclear power plant. We are not there yet." — Trey Lauderdale: He clarifies Atomic Canyon’s safety-first approach and near-term product scope. "Boring is good. We love boring in nuclear." — Trey Lauderdale: Explaining why document search is a valuable first product in a risk-averse industry. "The only way we're going to succeed as an industry is through the augmentation of our work staff with artificial intelligence." — Trey Lauderdale: His closing argument about nuclear workforce shortages and AI’s role in scaling the sector.
Implications: Atomic Canyon’s approach suggests the most viable nuclear AI near term is operational assistance, not autonomy. If successful, it could speed licensing, reduce paperwork, and help the nuclear industry scale its workforce and deployment pipeline.