Can nuclear power the AI factory? Crusoe and Aalo find out

Crusoe and Aalo Atomics announced a strategic partnership on 30 July, with plans to power a Crusoe Spark modular data centre at Idaho National Laboratory using Aalo’s nuclear technology from 2027. It is being pitched as the first nuclear-powered “AI Factory” data centre, though the more useful story isn’t the superlative. It’s what has to happen between now and then for the deal to actually work.

Aalo reached criticality on its Aalo-X reactor this summer, becoming one of only four companies to do so under the US Department of Energy’s Reactor Pilot Program, launched under an executive order that gave participants an aggressive deadline to get reactors operating. Matt Loszak, Aalo’s chief executive, called it a turning point for the company’s commercial ambitions.

“As one of only four companies to reach criticality by President Trump’s deadline for companies in the DOE Reactor Pilot Program, we are now well positioned for not only the full power operation of our 10 MWe Aalo-X reactor, but also the deployment of the Aalo Pod to power commercial data centres,” Loszak said.

Criticality means the reactor achieved a sustained fission reaction, an engineering milestone. It is not the same as commercial licensing, and it doesn’t mean Aalo can simply bolt a reactor onto any customer’s site tomorrow.

The DOE pilot programme gave Aalo a faster regulatory pathway specifically for a test reactor on federal land at INL, working alongside the DOE rather than solely through the Nuclear Regulatory Commission’s standard commercial licensing process. Scaling the 50MWe Aalo Pod to Crusoe’s wider data centre footprint by 2029 will mean navigating NRC licensing for reactors sited off government land, a process that has historically taken years rather than months, even for designs the DOE has already blessed.

Aalo’s pace so far has been genuinely unusual for the nuclear sector, but the pilot programme was built to compress the easiest part of the regulatory journey. The harder part, commercial-site licensing at scale, still lies ahead.

Where the money sits

Neither company has published pricing, and that’s not unusual this early. But the structure of the deal tells you something about how Crusoe is thinking about long-term energy costs. Crusoe isn’t buying reactors outright. It’s testing a proof of concept at INL first, with the 2027 Spark deployment acting as a live pilot before any commitment to deploy Aalo Pods and 50MWe XMR plants across its own data centres by the end of 2029.

That sequencing matters for anyone modelling long-term power costs against AI compute demand. A behind-the-meter or on-site nuclear source, if it works, offers something PPAs with grid-connected renewables can’t: power that isn’t subject to interconnection queues or transmission constraints, delivered at a fixed cost profile once the reactor is built. Cully Cavness, Crusoe’s co-founder and chief strategy officer, framed it as a reliability and scale question rather than a straightforward cost play.

“The next generation of AI infrastructure demands power that is reliable, abundant, clean and at scale, and Aalo’s nuclear reactors are built to meet that need,” Cavness said.

The economics only really get tested once Aalo has to build and finance multiple 50MWe units at commercial sites, rather than a single demonstration unit on a DOE campus with federal cost support baked into the pilot programme.

Capital cost overruns have sunk more nuclear projects than technical failure has, and factory-built modular reactors are still an unproven manufacturing model at the volumes Aalo would need to hit its 2029 target.

A pattern, not a one-off

This is Crusoe’s second publicly announced nuclear tie-up. The company previously partnered with modular nuclear developer Blue Energy on a Texas project, and it has separately built out a large AI campus in Abilene alongside conventional power sources for Oracle and OpenAI’s Stargate build-out, plus roughly 1.6 gigawatts of capacity now under contract to Meta. Read together, the pattern looks less like a single bet on Aalo and more like Crusoe hedging across multiple reactor technologies and vendors while its gas and grid-connected capacity does the heavy lifting in the near term.

That’s a rational posture given how early-stage the SMR sector remains. Rolls-Royce SMR is working through licensing in the UK and Czech Republic, ARC Clean Technology has signed with Deep Atomic to target North American data centres, and Google has separately gone the corporate PPA route with Kairos Power and the restart of Duane Arnold. Crusoe’s approach, running a live pilot with Aalo while keeping other options open, mirrors how hyperscalers have generally treated nuclear so far: real capital, but staged and diversified rather than all-in on one design.

The International Energy Agency has tracked the SMR pipeline tied to data centre operators nearly double, from 25 gigawatts at the end of 2024 to 45 gigawatts today, so Crusoe and Aalo are not operating in isolation. What sets this deal apart is the timeline. 2027 for a live demonstration and 2029 for fleet-wide deployment is fast even by the standards of an industry that has recently rewarded speed over caution.

Whether that schedule survives contact with NRC licensing, supply chain realities for enriched fuel, and the capital markets’ patience for pre-revenue nuclear manufacturers is the question worth watching.