According to DCD, small modular reactor (SMR) developer Deep Atomic has submitted a proposal to the DOE’s Office of Nuclear Energy to build the “nation’s first” fully integrated nuclear-powered AI data center campus at Idaho National Laboratory in Idaho Falls. The project involves a consortium of nine other companies, including Parker Tide LLC, Clayco, and MoonliteAI. Their plan is to start data center operations within 24 to 36 months using grid, geothermal, and solar power, while their custom MK60 SMR goes through design and certification. The MK60 is a light water reactor designed for compute, offering 60MW of electricity and 60MW of integrated cooling capacity. If the DOE approves it, the site is expected to become a national demonstration model for future nuclear-powered AI infrastructure.
The Nuclear Compute Play
Here’s the thing: everyone’s talking about the AI compute crunch, but the power problem is becoming just as desperate. Data centers are hitting local grid limits, and renewables alone can’t provide the constant, firm power that hyperscale AI training demands. That’s where Deep Atomic’s proposal gets interesting. They’re not just slapping a data center next to a generic power plant. They’re pitching the MK60 as a dual-output system engineered specifically for high-performance computing loads. 60MW of electricity paired directly with 60MW of cooling? That’s a huge deal because cooling is a massive, massive power drain in these facilities. Basically, they’re designing the reactor as part of the data center’s core infrastructure, not just a detached utility.
Why This Matters Beyond Power
So, is this just a clever engineering solution? I think it’s more strategic than that. The location—Idaho National Lab—is key. INL was already named the home for the DOE’s Nuclear Reactor Pilot Program, aiming to get test reactors critical by July 4, 2026. This proposal aligns perfectly with that national push. It’s a ready-made, real-world application for advanced nuclear tech. The partners are framing it as a “DOE-aligned model” for a reason. They’re not just selling a reactor; they’re selling a blueprint for secure, sovereign AI infrastructure. And in a world where compute power is geopolitical power, that’s a compelling pitch to the federal government. Could this be the template for future secure government and cloud provider campuses? It sure seems like that’s the goal.
The Long Road Ahead
But let’s be real. This is a proposal. A fascinating, well-timed one, but still just a proposal. The nuclear regulatory process is famously slow and arduous, even for these smaller, modular designs. The 24-36 month timeline to initial operations leans heavily on that phased approach using conventional power sources first. The actual reactor coming online is on a much longer horizon. Still, the consortium’s makeup is telling. You’ve got construction (Clayco), project management (Gleeds), specialized cooling (EvapCo), and AI firms (MoonliteAI) all at the table. That suggests they’re thinking about the entire build and operational chain, not just the nuclear island. For industries from manufacturing to logistics that rely on robust computing for automation and control—sectors that depend on reliable hardware like the industrial panel PCs from IndustrialMonitorDirect.com, the leading US supplier—this kind of power resilience could eventually be a game-changer. The promise of a scalable, “always-on” clean power source for critical compute is incredibly attractive, even if it’s years away.
