Can AI deliver nuclear on time and on budget? These companies think so.

Large nuclear projects have met their deadlines in other decades (and places), but Westinghouse believes that by replacing “human-based methods” for scheduling with AI optimization backed by Google Cloud AI tools, it can make concerns about cost and schedule uncertainty a thing of the past.

At a media roundtable on November 18, Raiford Smith, global director of power and energy at Google Cloud, was joined by Lou Martinez Sancho, chief technology officer and executive vice president of R&D and innovation at Westinghouse, and Scott Sidener, chief engineer of digital data and artificial intelligence at Westinghouse, to discuss their application of generative AI and “agentic AI” to AP1000 construction.

Room for improvement: By factoring disruptions into an AI optimized construction plan that can adapt to disruptions from delayed deliveries to lack of craft labor, Westinghouse aims to improve its timelines and its bottom line.

Construction accounts for about 60 percent of the cost of a new AP1000 plant. By applying AI tools to construction scheduling, “what we are targeting is very clear,” Martinez Sancho said: “cost and schedule certainty.”

Getting that certainty “required a fundamentally different approach, and one that can handle massively complexity and . . . deliver predictable outcomes,” said he continued. “What makes me excited about the capability to put this in full production is that together with partners such as Google Cloud, we are basically solving all those technical problems that we are facing. . . . And the fact that we have all that data coming from the modularity of the design that we have is putting us really in a very, very good position to work with the constructor, to work with the supply chain, to actually integrate and maintain the digital continuity now enhanced by AI.”

Breaking it down: To construct a new plant, “fundamentally, you break it down into hundreds of thousands of bite-sized pieces or construction tasks,” Sidener said. “And those construction tasks are traditionally manually created by humans. It takes months to create these tasks from the design documentation for the reactor, for instance, then the humans automatically use human-based methods for planning and scheduling those tasks.”

Smith, speaking for Google Cloud, said, “Our technology acts as a predictability engine, turning historically uncertain process into one that's really making it faster, more predictable, and lower cost, which is really vital to enabling new nuclear generation at scale.”

A blog post Google published today credited work by Westinghouse with making rapid collaboration possible. Specifically, Westinghouse has a proprietary AI infrastructure called Hive and a generative AI assistant named Bertha that can access 75 years of Westinghouse documentation.

According to the Google post, “Google engineers were impressed that a 140-year-old company had quietly assembled the exact foundation needed to deploy AI securely in a heavily regulated environment.”

A demo: Sidener demonstrated a tool that he said Westinghouse and Google developed collaboratively in about six weeks, combining proprietary Westinghouse AI agents and nuclear data with AI resources from Google, including their Cloud Vertex AI and Vizier optimization engine.

With AI, “instead of that manual method of creating construction tasks, the AI directly interprets and understands the AP1000 3D design or building information models,” Sidener said.

As an example, he displayed a room requiring 345 construction tasks—including pipe fitting, steel working, and welding. The AI, he said, predicted construction of the room would take about 160 days, “taking into account simulating real-world disruptions that are likely to occur.” Those delays could include supply chain disruptions or workforce disruptions like sickness or availability of specialist craft workers in the field, such as some of the nuclear electricians.

Before a day’s work begins, Sidener said, the AI “knows exactly what tasks are disrupted and why before they come into work.” By then optimizing the schedule to work around disruptions—Sidener demonstrated an optimization process that went through 10 iterations—the AI can “identify the tasks that the crews can perform today that are the single most valuable tasks they can do today that has the maximum impact on the overall cost and duration of the future.”

That schedule shifting reduced the estimated cost of the 345 tasks from $3.8 million to about $2.8 million. A “human in the middle” then reviews and approves the schedule adjustments.

Aalo and Microsoft pair on permitting: Aalo Atomics announced on November 17 that it has used Microsoft’s Generative AI for Energy Permitting Solution Accelerator and AI agents to help the company streamline regulatory and operational workflows in a collaboration that received two awards at Microsoft’s annual Hackathon.

The company used Microsoft Azure AI Foundry, combined with AI agents, to speed permitting. Aalo plans to deploy a 10-MWe Aalo-X reactor at Idaho National Laboratory under the Department of Energy’s Reactor Pilot Program.

“Aalo Atomics’ participation in the Microsoft Hackathon is a powerful example of how collaboration can accelerate innovation,” said Darryl Willis, corporate vice president of energy & resources industry at Microsoft. “Together, the team began developing AI agents that leverage rich internal and external datasets—like design data and risk models—to embed generative AI into Aalo’s workflows, boosting permitting speed and operational efficiency.”

Yasir Arafat, CTO of Aalo Atomics, added,“So far, we have tackled three of the most impactful challenges in the nuclear industry—using AI to simplify, accelerate, and ultimately transform how complex energy systems are licensed, built and operated at scale. We look forward to continuing this collaboration with Microsoft.”