AtkinsRéalis and NVIDIA set to accelerate build of nuclear-powered AI factories

The two companies will explore how they can combine AtkinsRéalis’s CANDU pressurised heavy-water reactor design technology with NVIDIA’s blueprint for designing and operating gigawatt-class AI factories.

Global AI infrastructure demand is rapidly outpacing available power supplies, and nuclear power is emerging as a leading low-carbon baseload for gigawatt-class AI factories, AtkinsRéalis said. With its CANDU technology, AtkinsRéalis said it is “well positioned to work alongside NVIDIA’s Omniverse DSX Blueprint for next-generation hubs supporting dedicated AI computing workloads”.

NVIDIA launched Omniverse DSX Blueprint last year. It maximises GPU productivity and energy efficiency across the compute fabric with a single digital twin environment that can be simulated, optimised and used for operations, NVIDIA said.

The pair have provided some detail on how they will work together. First, AtkinsRéalis will examine how the standardised nuclear power solutions within its CANDU portfolio can be integrated with large-scale AI data centre requirements. NVIDIA will contribute strategic leadership in compute, networking and AI, including digital twin and high-fidelity simulation tools that could help modernise nuclear delivery and support licensing pathways.

Second, NVIDIA Omniverse, agentic AI, and large language and visual language model technologies can streamline engineering workflows, enhance collaboration, accelerate construction schedules, and improve safety and performance outcomes across nuclear, power and data centre programmes.

And third, AtkinsRéalis will explore how its capabilities in power distribution, cooling systems, resilience, modular engineering and construction efficiency can support the deployment of NVIDIA DSX-based AI infrastructure.