Running nuclear power plants is a labor-intensive affair, with plants employing more than 500 workers per plant. And it requires a host of processes, such as condition report analysis, outage planning, and regulatory documentation — each of which can take teams of highly skilled people weeks to complete.
As delicate and essential as these processes are, Jerrold Vincent, CFO and co-founder of Nuclearn, believes that they’re also well-suited to be automated by artificial intelligence. The startup, founded in 2021, has created an AI platform dedicated to nuclear operations. And today, it announced that it has raised $10.5 million in Series A funding, led by Blue Bear Capital, with additional investments by SJF Ventures, AZ-VC, and Nucleation Capital.
“Because nuclear is so documentation- and process-heavy, there’s a really unique opportunity to use AI and machine learning to make plants run more efficiently,” Vincent told Latitude Media. “The majority of costs for running a nuclear power plant [aren’t] fuel or equipment. It’s actually people.”
And those people are in short supply. The nuclear sector is facing the same labor shortages that challenge many other energy sectors. This will be exacerbated by an expected wave of retirements; in 2022, the Global Energy Talent Index estimated that around 25% of nuclear workers were over 55 years old.
At the same time, nuclear is going through something of a renaissance, prompted by the booming data center sector’s hunt for abundant baseload power, which has led hyperscalers such as Microsoft and Google to invest heavily in nuclear solutions old and new. The increase in demand is expected to put additional strain on the dwindling workforce.
Nuclearn aims to use AI and machine learning to support, especially the work of engineers who perform documentation-heavy, repetitive work. Using advanced tools, Nuclearn can help workers write and classify work orders and engineering evaluations, assess whether new tests or changes to a facility are in compliance or require approval by the Nuclear Regulatory Commission, and renew licences, among other things. It does so using both generic models trained on publicly available data, and customer-specific models trained on the needs and data sets of specific power plants.
For example, the startup’s first product, called Corrective Action Program AI, or CapAI, automates the mandatory systems for identifying, assessing, and resolving issues. “Historically, at a plant, there’s been a team of people to do that job,” Vincent said. “With our platform, we can do it automatically, saving them time and effort.” Despite the fact that all nuclear power plants are required to have a corrective action program, the product requires customer-specific training because all the programs look slightly different, and general modeling techniques can’t achieve the needed levels of accuracy.
Nuclearn’s platform is already deployed in more than 65 reactors globally. The startup uses an annual subscription business model, which Vincent says isn’t common in the industry, but that customers like because it’s transparent and easy to use.
Nuclear challenges
Because of how the nuclear industry is structured — and its safety and security needs — it comes with some unique challenges for an AI platform like Nuclearn. For one, many of the documents Nuclearn uses for training its models are “ancient,” according to Vincent, and require “a lot of work on data cleanup.”
As Ernst Sack, founding partner at Blue Bear Capital, explained, the industry “has had very little innovation, especially in software that’s been applied to it.”
Additionally, the things Nuclearn automates are often complicated, multi-step processes involving many different documents. “When you’re doing that kind of work, you can’t just provide an answer to a nuclear engineer and have them accept that and sign off on it,” Vincent said. “You need auditability and visibility into the process [and] the reasoning that goes through it, so that they can review, adjust as needed, and really trust that kind of output.”
And in the U.S., access to sensitive nuclear energy information is restricted under federal law to U.S. citizens, and in many cases also to lawful permanent residents. “What that means in practice is that a lot of sites will not send that data outside of their network; or if they do, there are very strict requirements on where that data is hosted, which make most cloud providers basically a no-go for that kind of data,” Vincent said.
This means that Nuclearn owns its own hardware, which it keeps in a co-location data center in Phoenix, and a small part of its newly-announced fundraise will go into its expansion.
“It’s not something many startups do because they use cloud, and [hardware] comes across as an expense,” Vincent said. “But we plan our hardware strategy in a way that allows us to have the hardware we need to train the latest and greatest models for our hosted customers, and do so with a cost-effective approach.”
