A Nuclear Renaissance? Without AI it will be very hard...

Let’s picture a CEO announcing the opening of a new fab requiring 10,000 workers in a town of 3,000 inhabitants, children included. The optimism of the announcement would not resist the simple mathematical consideration that the town does not have enough workers to make the to-be-built fab operational. And this regardless of how high the demand for those fab’s products is estimated. As absurd as it may sound, this is what is happening in the nuclear industry today. 

Several countries are announcing plans to start, restart, prorogate or increase their nuclear programs; and many more are adding nuclear as an electricity source in their energy mix. At the same time the industry which is supposed to deliver keeps raising concerns about the lack of sufficient human resources to sustain these programs. Even in countries with a long-standing nuclear culture like France the difference between the number of nuclear specialists needed and those that the educational system can produce is staggering.  The problem is so acute that special acronyms such as NHCB (Nuclear Human Capacity Building) are used to describe it. However, in simple terms, the equation seems unresolvable despite the sexy acronym.

We believe Artificial Intelligence can be the solution. Because AI (and only AI) can move the problem from “nuclear human capacity building” to “nuclear processes capacity reduction”, changing the equation to dramatically reduce the number of specialists needed to perform identified tasks.

Instead of announcing a Nuclear Renaissance which would require thousands of specialists currently not available (and with no educational system capable of producing them), the objective should be to update industry methodologies and processes to be able to do more with AI-driven intervention.

This is exactly what Artificial Intelligence is, this is exactly what Artificial Intelligence does.

Artificial intelligence IS NOT Machine Learning, ChatGPT, predictive maintenance, automation, and is not robotics or mechanization. Artificial intelligence is a new way of doing things, it is a disruptive paradigm that can guarantee higher efficiency while maintaining 100% explicability when used in critical environments such as nuclear. 

 Strangely, the world of nuclear remains today quite impermeable to the potential benefits of Artificial Intelligence; use of AI is limited in nuclear if compared to other industries, there are very few specialists in both disciplines capable of identifying the right level of cooperation, and both nuclear scientists and companies remain statistically much more skeptical about AI or its benefits. In this article we will detail the innovative aspects of AI that can be more beneficial to specific nuclear processes and/or technologies, with the hope that this can contribute to bridge the distance and to bring closer two worlds that need to collaborate for their own long-term interest.

AI for Fission

Fission technologies are available today. Current nuclear production in the world is based on fission, and the next reactors to be built will still be based on fission though revisited by a modular approach (SMR) which bears significant advantages. More specifically, the initial investment per single reactor can be reduced, in the case of multi-unit installations, by the possibility to sell energy as soon as each unit is completed, thus impacting financial costs. Then of course smaller dimensions reduce mega-project complexities. And construction in series allows the experience curve to progress much faster. All this is forecasted to represent a significant reduction in the cost-per-Mw vs current large reactors.

Each SMR must in all cases undergo a complex licensing procedure in order to be built. Licensing is time-consuming and delicate, due to the complexity and number of reviews, analyses, and documents to be submitted to regulatory bodies for evaluation of the design as well as the location of the facility requires a lot of skilled manpower, from both sides of the process. Asynchronous communication through static documents, whether paper or digital, requires a lot of material to be updated manually at each stage of the project. And this leaves even less room for potential changes that may be needed during construction.

The starting point is an integrated platform to create, manage, modify, and share all the documentation required during the licensing phase of a reactor project, construction site, and environmental impact analysis. This is the digitalization phase. 

But AI goes much further: by training specific models on regulatory databases and previously approved licenses, it is possible to evaluate components and systems of the project and then assist in generating the final documents to be submitted to the competent authorities. By developing standardized tools and formats for data and information exchange, a single platform can be shared with the regulatory authority, ensuring transparency and traceability of all communications, project changes, and updates to engineering, safety, and other analyses.

This reduces risks and financial commitments during the licensing phase, which includes also project design and therefore can have a positive impact all along the supply-chain, as suppliers can be pre-approved using the same methodologies.

AI for Fusion

Artificial Intelligence holds transformative potential for nuclear fusion, especially with not merely engineering-related but fundamentally epistemological challenges in mind. Unlike traditional nuclear energy, fusion remains the “final frontier”—a domain where scientific exploration outpaces practical feasibility. While the metaphor of fusion as a "Star Trek" dream captures its visionary appeal, it cannot obscure the immense hurdles we still face in rendering it a scalable, economically sustainable energy source.

Today, many fusion research initiatives incorporate AI, but typically through conventional Machine Learning techniques. These approaches, though effective in specific applications such as plasma diagnostics or anomaly detection, often depend heavily on vast datasets and static training pipelines—an inherent limitation in a field where experimental data is scarce, costly, and time-consuming to generate.

The issue of data scarcity is being partially resolved by the use of compact accelerators capable of realizing several and enduring fusion processes on the desktop of a laboratory, instead of relying only on large and expensive Tokamaks or Stellarators. But even these advances can only provide a fraction of the data needed for a full simulation.

This is where AI goes beyond Machine Learning. The next breakthrough in fusion will come not from more data, but from smarter models. Emerging AI architectures, such as foundation models, reasoning agents, and self-supervised learning systems, offer a paradigm shift. These systems excel not by digesting terabytes of labeled inputs, but by learning autonomously, adapting in real time, and reasoning in abstract environments—qualities that align closely with the nature of fusion R&D. A model that can hypothesize, simulate, and refine its own experiments could accelerate progress dramatically, freeing researchers from the bottleneck of data dependency and allowing for more exploratory, imaginative scientific processes.

By integrating these advanced forms of AI into fusion laboratories—not only to interpret data but to orchestrate simulations, generate hypotheses, and optimize designs—we can unlock new paths toward the holy grail of clean, limitless energy.

Conclusion and disclaimer

The nuclear option remains divisive and subject to public debate, as it should be in democratic societies. Each country has the right to choose its path in terms of energy mix or technological neutrality. Nothing in this article should lead us to believe that our Institute has a definite policy on the matter.

This paper only intends to highlight the common ground in which AI and nuclear operate, as well as the commonality of issues such as social acceptance or operational Human-In-the-Loop. Furthermore, from a purely functional point of view, we have tried to illustrate the necessity for the nuclear industry, no matter the size of its future deployment, to take AI’s latest advancements into consideration from the beginning in order to deliver what it promises, and maybe even more.