NVIDIA and the US government want to build nuclear power plants twice as fast and 50% cheaper.
Idaho National Laboratory has agreed to collaborate with NVIDIA on how to bring new nuclear reactors to production sites in the US faster. The key tool is artificial intelligence . The partners expect to cut development time by at least half while simultaneously reducing operating costs by more than 50%.
The agreement is part of the US Department of Energy's Genesis Mission initiative, which is building a large-scale scientific computing platform to accelerate research, safety, and energy development. One of the program's components is codenamed Prometheus. The message is straightforward: nuclear energy must be built faster, operate safer, and be less expensive.
Prometheus envisions AI being used throughout the entire reactor lifecycle, from the initial drawings to plant operations. This includes not only design but also licensing, manufacturing, construction, and operation. Currently, these stages take years due to the volume of calculations, documentation reviews, and approvals. The initiative's creators want to automate the most labor-intensive engineering tasks without removing humans from the process. The model is called "human-in-the-loop": algorithms generate calculations and options, while decisions are made by engineers and regulators.
Another reason for acceleration is the growing demand for electricity from data centers, which power the AI infrastructure. This is intended to be a two-way street: AI helps design and commission reactors faster, while nuclear power supplies computing platforms and the data centers themselves.
The technical foundation of the project is generative models, digital twins, and agent-based operating scenarios. A digital twin is a detailed computer model of a reactor that can be used to simulate operational modes and emergency scenarios even before construction begins. These models are planned to be trained using INL's long-term data: laboratory experiments, archival studies, and pilot plant operation data. The goal is simple: to test the system's behavior in various modes in advance, minimizing the need for rework during construction.
NVIDIA is taking over the computations. The company is providing hardware and software capable of quickly running complex models on graphics processing units (GPUs). Several core computing programs, including MOOSE, BISON, Griffin, and Pronghorn, will be ported and optimized for these accelerators. Engineers use them to model what happens inside the reactor: how the fuel behaves, how materials heat up and age, and how neutrons move. If these calculations are run on GPUs, then what used to take weeks to calculate can be run significantly faster. This will also provide more power for more detailed simulations, meaning fewer surprises during construction and operation.
The US Department of Energy plans to use supercomputers to train large AI models and run complex calculations. To ensure that the digital twins are not just pretty pictures, they will be verified against real measurements at INL facilities. Among the sources of such data are a neutron radiography facility and the MARVEL project, which is related to microreactors. MARVEL has not yet been launched, but it is expected to be used in the future as one of the suppliers of experimental results.
The project isn't just about computation. A significant part of the plan revolves around regulatory matters: the initiative's creators hope that AI will help speed up the preparation and review of licensing materials, thereby shortening a stage that today often takes the longest and is the most expensive. They hope to expand the program and include more participants, from private reactor developers and energy companies to investors and other national laboratories.
If the plan works, the US will be able to deploy new reactors faster while simultaneously strengthening the infrastructure that powers its own artificial intelligence industry .
Idaho National Laboratory has agreed to collaborate with NVIDIA on how to bring new nuclear reactors to production sites in the US faster. The key tool is artificial intelligence . The partners expect to cut development time by at least half while simultaneously reducing operating costs by more than 50%.
The agreement is part of the US Department of Energy's Genesis Mission initiative, which is building a large-scale scientific computing platform to accelerate research, safety, and energy development. One of the program's components is codenamed Prometheus. The message is straightforward: nuclear energy must be built faster, operate safer, and be less expensive.
Prometheus envisions AI being used throughout the entire reactor lifecycle, from the initial drawings to plant operations. This includes not only design but also licensing, manufacturing, construction, and operation. Currently, these stages take years due to the volume of calculations, documentation reviews, and approvals. The initiative's creators want to automate the most labor-intensive engineering tasks without removing humans from the process. The model is called "human-in-the-loop": algorithms generate calculations and options, while decisions are made by engineers and regulators.
Another reason for acceleration is the growing demand for electricity from data centers, which power the AI infrastructure. This is intended to be a two-way street: AI helps design and commission reactors faster, while nuclear power supplies computing platforms and the data centers themselves.
The technical foundation of the project is generative models, digital twins, and agent-based operating scenarios. A digital twin is a detailed computer model of a reactor that can be used to simulate operational modes and emergency scenarios even before construction begins. These models are planned to be trained using INL's long-term data: laboratory experiments, archival studies, and pilot plant operation data. The goal is simple: to test the system's behavior in various modes in advance, minimizing the need for rework during construction.
NVIDIA is taking over the computations. The company is providing hardware and software capable of quickly running complex models on graphics processing units (GPUs). Several core computing programs, including MOOSE, BISON, Griffin, and Pronghorn, will be ported and optimized for these accelerators. Engineers use them to model what happens inside the reactor: how the fuel behaves, how materials heat up and age, and how neutrons move. If these calculations are run on GPUs, then what used to take weeks to calculate can be run significantly faster. This will also provide more power for more detailed simulations, meaning fewer surprises during construction and operation.
The US Department of Energy plans to use supercomputers to train large AI models and run complex calculations. To ensure that the digital twins are not just pretty pictures, they will be verified against real measurements at INL facilities. Among the sources of such data are a neutron radiography facility and the MARVEL project, which is related to microreactors. MARVEL has not yet been launched, but it is expected to be used in the future as one of the suppliers of experimental results.
The project isn't just about computation. A significant part of the plan revolves around regulatory matters: the initiative's creators hope that AI will help speed up the preparation and review of licensing materials, thereby shortening a stage that today often takes the longest and is the most expensive. They hope to expand the program and include more participants, from private reactor developers and energy companies to investors and other national laboratories.
If the plan works, the US will be able to deploy new reactors faster while simultaneously strengthening the infrastructure that powers its own artificial intelligence industry .
