Tech companies are banking on nuclear power to supply the massive energy demands of their AI ambitions. However, data centers need electricity now—and the nuclear sector has a reputation for being slow.
Trey Lauderdale believes AI could be the key to accelerating the nuclear industry.
Lauderdale’s interest in nuclear energy was sparked locally in San Luis Obispo, California, where he frequently met residents working at the nearby Diablo Canyon Power Plant. “They’re coaching our flag football team,” he noted.
Inspired by Local Insights, Lauderdale Turns to AI to Tackle Nuclear Industry’s Document Overload
Through these conversations, he discovered that nuclear plants are overwhelmed by documentation—Diablo Canyon alone holds about 2 billion pages, he said. Drawing on his background as a healthcare entrepreneur, Lauderdale suspected AI could help streamline this massive paper trail.
He launched Atomic Canyon roughly 18 months ago, initially self-funding the venture. The startup uses AI to assist engineers, maintenance teams, and compliance officers in quickly locating the documents they need.
In late 2024, Atomic Canyon secured a contract with Diablo Canyon. That breakthrough triggered interest from other nuclear facilities. “That’s when I realized it was time to raise capital,” Lauderdale said.
The company has now closed a $7 million seed round led by Energy Impact Partners, as exclusively shared with TechCrunch. Other investors include Commonweal Ventures, Plug and Play Ventures, Tower Research Ventures, Wischoff Ventures, and several early angel investors.
Early AI Challenges Highlight the Complexity of Nuclear Language for Atomic Canyon’s Team
When Atomic Canyon got off the ground, its AI engineers initially struggled—early tests with various models produced disappointing outcomes. “We quickly realized the AI starts hallucinating when it encounters nuclear terminology,” Lauderdale explained. “It just hasn’t seen enough examples of those acronyms.”
However, training a new AI model demands immense computing power. To solve that, Lauderdale secured a meeting with Oak Ridge National Laboratory—home to cutting-edge nuclear research and the world’s second fastest supercomputer. The lab was intrigued by the concept and granted Atomic Canyon 20,000 GPU hours of computing time.
Atomic Canyon Uses RAG and Sentence Embeddings to Make Nuclear Documents Searchable and Reliable
Atomic Canyon’s AI models leverage sentence embeddings, which are well-suited for indexing and organizing large document sets. The system makes nuclear plant documents searchable using retrieval-augmented generation (RAG). RAG combines large language models with direct references to source documents, helping to generate accurate responses and minimize hallucinations.
For now, Atomic Canyon is focused on document search, largely because it carries less risk.
“We’re starting with generative features like document title suggestions, where mistakes might be annoying but not dangerous,” Lauderdale explained. “It won’t compromise safety at the plant.”
Looking ahead, he imagines Atomic Canyon’s AI eventually drafting the initial versions of documents, complete with citations. “There will always be a human involved in the process,” he emphasized.
Lauderdale hasn’t set a timeline for that goal. For now, search remains the core priority. “Search is the foundational layer—you have to get that right first,” he said. And with the sheer volume of documents across the nuclear industry, “we’ve got a long runway just focusing on search.”
Read the original article on: TechCrunch
Read more: Crystal-Based Cooling Could Prevent Future Devices From Overheating