While AI systems are known to spew wrong information and make up facts, Google and Westinghouse Electric are now pressing generative AI models into service to transform how nuclear reactors are constructed and optimize their operation.
The two companies aim to pair Westinghouse’s “nuclear-specific” HiVE GenAI system and its “nuclear-aware” large language model (LLM), bertha, with Google Cloud’s technology and expertise to streamline construction of new nuclear plants and improve the performance of current reactors through “data-driven AI insights.” The electric gear maker claims its nuclear AI technology is trained on “75 years of proprietary data, knowledge, and expertise,” according to the deal announcement.
Westinghouse interim CEO Dan Sumner claimed the AP1000 pressurized water reactor design is the only fully licensed, construction-ready modular reactor available today, and therefore represents the best way to add extra atomic energy sources to the US grid.
Ironically, the urgent need to add extra generating capacity to the grid is because of the ever-increasing number of datacenters that are popping up in the US to develop and train ever more sophisticated AI models.
Google has also inked an agreement worth over $3 billion to supply up to 3,000 MW of hydroelectric power to the US grid. The deal with Brookfield Asset Management is a 20-year Power Purchase Agreement (PPA) for Brookfield’s Holtwood and Safe Harbor hydroelectric facilities in Pennsylvania to support Google’s operations across the PJM Interconnection and Midcontinent Independent System Operator (MISO) power regions.
A recent Deloitte Insights report estimated that the energy required by AI datacenters in America may be more than 30 times greater in a decade.
According to Westinghouse, there are currently six AP1000 units in full commercial operation – two in the US and four in China – plus three planned in Poland, two in Bulgaria, and nine in Ukraine. The company says the reactor design has passive safety systems that will cause it to shut down in the event of a malfunction.
The pair claimed they have already achieved a proof-of-concept of the technology, getting it to generate and optimize AP1000 modular construction work packages. This involved the Westinghouse WNEXUS digital plant design platform combined with HiVE AI and Google Cloud technologies, said to include Vertex AI, Gemini, and BigQuery.
Westinghouse said that HiVE and bertha will be used to optimize new deployments of the AP1000 reactor, as well as the smaller AP300 small modular reactors, and the even tinier eVinci microreactors. The company believes its AI solutions are also ready to support power plant operations.
The eVinci is a reactor design currently under development, with funding from the US Department of Energy, expected to generate 5 MW of power for eight years.
Let’s hope that Google and Westinghouse don’t find that their AI solutions need constant checking by humans to make sure they don’t mess up, and that they don’t get things wrong about 70 percent of the time. ®