Two industrial powerhouses are making a strategic pivot that could reshape how artificial intelligence gets its power. Caterpillar is supplying natural gas generators to one of the world's largest planned AI data center campuses, while AtkinsRéalis is partnering with Nvidia to design nuclear-powered AI infrastructure. These moves signal that the real bottleneck for AI expansion isn't computing chips anymore; it's reliable, massive-scale electricity. The numbers explain the urgency. Global data center power consumption is forecast to nearly double from 415 terawatt-hours in 2024 to 800 terawatt-hours by 2026, driven almost entirely by artificial intelligence workloads. That represents a 165% increase in data center power demand in just two years. To put this in perspective, this growth trajectory would strain any traditional energy infrastructure built for gradual, linear increases in electricity needs. What Makes Nuclear Power the Ideal Solution for AI Data Centers? Unlike solar panels or wind turbines, which generate power intermittently, nuclear plants run continuously 24 hours a day, seven days a week. This stable, always-on baseload power is exactly what AI data centers require. These facilities cannot afford to shut down when the sun sets or the wind stops blowing. They need predictable, constant electricity to keep thousands of graphics processing units (GPUs) running without interruption. Nuclear energy also addresses the carbon footprint concern. As companies face pressure to reduce their environmental impact, nuclear power offers a low-carbon alternative to natural gas or coal-fired plants. This combination of reliability and sustainability makes it increasingly attractive to hyperscalers, the massive cloud computing companies that operate AI infrastructure at global scale. How Are Companies Accelerating Nuclear Plant Deployment for AI? - Digital Twin Simulation: AtkinsRéalis and Nvidia are using Nvidia's Omniverse tools to create virtual replicas of entire nuclear facilities before construction begins, allowing engineers to test designs, identify problems, and optimize layouts in software rather than discovering issues during physical construction. - Streamlined Design and Licensing: By using artificial intelligence to automate parts of the design process and simulate regulatory scenarios, the partnership aims to compress timelines that traditionally take years for permitting and approval into shorter cycles. - Leveraging Existing Expertise: AtkinsRéalis brings 70 years of nuclear engineering experience and proprietary CANDU reactor technology, providing a foundation of proven capability rather than starting from scratch with unproven designs. The traditional nuclear industry moves slowly by necessity; safety and regulatory compliance cannot be rushed. However, the AI energy crisis is creating unprecedented pressure to accelerate these timelines. AtkinsRéalis's partnership with Nvidia represents an attempt to use artificial intelligence itself to speed up the deployment of the infrastructure that will power AI systems. If successful, this could compress what normally takes five to ten years into a much shorter window. What Role Are Traditional Industrial Companies Playing? Caterpillar's involvement reveals another critical piece of the puzzle. The company is supplying G3500 natural gas generators to the Monarch Compute Campus in West Virginia, a facility being developed by Nscale and Microsoft. This 1.35-gigawatt campus is designed to scale to 8 gigawatts of AI computing capacity. Caterpillar has also signed a global framework agreement with Atlas Energy Solutions for gigawatt-scale generator deployments through 2029, indicating this is not a one-off project but a sustained, multi-year commitment. These deals position Caterpillar's power systems division as a core infrastructure provider for the AI boom. The company is transitioning from being known primarily for construction equipment and mining machinery to becoming a critical supplier of on-site power generation for data centers. This shift reflects a broader recognition that hyperscalers cannot rely solely on the existing electrical grid; they need independent, dedicated power sources to ensure their facilities have the energy they require. Why Is the Market Undervaluing These Opportunities? Despite the strategic importance of these deals, the market has been slow to recognize their value. AtkinsRéalis trades at a price-to-earnings ratio of just 5.73, a fraction of what investors typically pay for companies in high-growth sectors. This valuation gap exists despite the company holding a record-high backlog of 21.1 billion dollars, which grew 22.6% year-over-year. The disconnect suggests investors remain skeptical about whether nuclear-powered AI infrastructure can be deployed quickly enough to matter, or whether execution risks in the nuclear industry will prevent these ambitious timelines from being met. The fundamental tension is one of speed versus tradition. Nuclear projects are inherently slow, with lead times measured in years. Meanwhile, AI compute demand is accelerating exponentially. If the physical infrastructure isn't ready when the demand arrives, the entire value proposition collapses. This execution risk is likely why the market has not yet fully priced in the strategic value of these nuclear-AI partnerships. The coming quarters will be critical. A successful pilot project demonstrating that digital twin design can actually accelerate nuclear facility construction would serve as a powerful proof-of-concept. Such a project would provide concrete evidence that the AI-driven optimization can overcome the nuclear industry's traditional slowness and attract further contracts from AI developers seeking reliable, low-carbon power. Without visible progress on these pilots, the nuclear-AI convergence risks remaining a promising concept rather than an executable blueprint for the next generation of data center infrastructure.
