The First Passenger: AI Is Already Boarding the Nuclear Aircraft While It's Still Being Certified
- Rich Washburn
- 5 days ago
- 6 min read
Two days ago I published "The First Customer," arguing that military microreactor procurement would de-risk the technology and supply chain that eventually powers commercial AI infrastructure.
The progression was simple: military mission requirement → government-backed reactor deployment → operational proof → commercial AI crossover.
Then Crusoe and Aalo Atomics announced they're putting a modular AI data center beside an advanced reactor at Idaho National Laboratory. That's not merely adjacent to the article. It's practically the next paragraph.
The Pattern I Described, Showing Up in Real Time
Here's the chain from "The First Customer": Department of Energy reactor program → Idaho National Laboratory → Aalo reactor → Crusoe AI data center → commercial Crusoe campuses.
The handoff I described as sequential — "the military is the first customer, the commercial market is the second" — is already beginning to overlap. The national laboratory and federal reactor programs are supplying the test infrastructure, fuel access, technical expertise and accelerated authorization environment. Crusoe is showing up before that bridge is even finished and parking an AI factory on it.
The 2027 installation will initially pair Aalo's reactor program with a Crusoe Spark modular data center at INL as a proof of concept. The companies then intend to deploy Aalo's 50 MWe reactor plants across Crusoe campuses by the end of 2029.
AI may not be the first customer for advanced nuclear. But it's the first commercial passenger — and it's boarding while the aircraft is still being certified.
Who Crusoe Actually Is
If you don't know Crusoe, the origin story matters because it explains the strategy.
Chase Lochmiller is one of TIME's 100 Most Influential People in AI. Former quant at Jump Trading. Former GP at Polychain Capital. Climbed Everest. Then co-founded a company named after Robinson Crusoe — the guy stranded on an island who had to figure shit out with whatever was around him. That's literally what Crusoe did.
In 2018, Chase and Cully Cavness started in an oil field, capturing wasted natural gas to power bitcoin miners. The pitch was contrarian: bring the computer to the energy, not the energy to the computer. Seven years later, that "stranded energy" philosophy built one of the most important companies in artificial intelligence. Now there are reports that Crusoe is in talks to raise $3 billion at a ~$30 billion valuation, triple its price tag from October ($10B). Let that sink in. $30B. For a company that barely was in the AI conversation three years ago.
The Numbers
5+ gigawatts contracted. 40+ gigawatts in the pipeline. The 1.2 GW Abilene campus (Stargate for OpenAI) went from dirt to live servers in 11 months — the next fastest bid was 2.5 years. Customers include OpenAI, Meta, Microsoft, Oracle, and Google. Full vertical integration from "electrons to tokens": energy, data center construction, cloud platform, and software/memory optimization via the Atero AI acquisition.
And now nuclear. Baseload, clean, 24/7 power purpose-built for AI factories.
Why This Validates the Thesis From "The First Customer"
The final section of that article said: "The constraint used to be computing power. Then it was data. Then it was model capability. Now it's electricity."
Crusoe's entire strategy is the corporate manifestation of that sentence. It started with stranded flare gas, expanded into dedicated natural-gas generation, batteries and grid-management systems, and is now pursuing both microreactors and gigawatt-scale nuclear campuses. Its earlier Blue Energy agreement contemplated a Texas AI campus receiving bridge power from gas before transitioning to as much as 1.5 GW of nuclear generation.
The progression I described — military proves the technology, commercial market follows — isn't some distant 2035 crossover. It's an industrial pattern that was already snapping into place while I was writing about it.
Crusoe Is Not Really a Data Center Company
Most cloud companies begin with servers and work backward toward power: we need this many GPUs, therefore we need this much electricity. Crusoe began at the opposite end: here is stranded or underutilized energy — what valuable computation can we place directly beside it? That is the consequential idea. The original flare-gas operation wasn't merely a clever bitcoin-mining trick. It was the first implementation of a much broader architecture: move compute to energy rather than endlessly trying to move energy to compute.
Bitcoin happened to be the first workload capable of tolerating that arrangement. AI is the workload that turns it into an industrial empire.
Nuclear Changes the Geometry
Natural gas let Crusoe exploit energy that already existed in inconvenient places. Small modular nuclear potentially lets Crusoe manufacture high-quality energy exactly where it wants the compute. That is a completely different capability.
A dedicated reactor could give an AI campus 24/7 baseload generation, much less dependence on transmission availability, predictable long-term power economics, reduced exposure to grid congestion, campus placement based on fiber, land, cooling and customers — not merely utility capacity — and a standardized power block that can be replicated with a standardized data-center block. The real product is therefore not "a reactor beside a data center." It is a repeatable AI-factory unit: reactor + generation island + cooling + modular data halls + GPUs + cloud software. That is the data-center equivalent of turning individually engineered mainframes into standardized servers.
The Vertical Integration Is the Moat
Crusoe can increasingly optimize across boundaries that remain separate for almost everybody else: generation cost versus workload scheduling, reactor output versus compute utilization, cooling architecture versus server density, memory optimization versus GPU count, construction sequencing versus customer demand, inference placement versus latency and power availability.
Its Atero acquisition extends that integration into GPU memory and workload efficiency, meaning Crusoe isn't only trying to produce more electricity. It is also trying to extract more useful tokens from every installed GPU and every delivered megawatt. That is why "electrons to tokens" is more than marketing language. It describes the unit economics. A normal data-center operator sells megawatts and square footage. A cloud provider sells compute. Crusoe is attempting to optimize and monetize the entire conversion chain from fuel to intelligence.
The 50 MW Size Is Revealing
Fifty megawatts is not enormous by hyperscale standards. A 1.2 GW campus would require roughly 24 such plants before accounting for redundancy, auxiliary loads or capacity factors. But that may be precisely the point. A 50 MW standardized block is potentially large enough to support a meaningful inference or specialized training campus, while remaining modular enough to replicate. Multiple reactor-and-compute blocks could be added as demand materializes instead of waiting for one gigantic utility interconnection and transmission buildout.
The future may not consist only of a few gigantic 1–5 GW training campuses. It could also include a distributed layer of 50–200 MW sovereign, industrial and regional inference factories placed near cities, military installations, manufacturing facilities, research centers or national borders. Training may remain concentrated because it rewards enormous scale. Inference increasingly rewards proximity, reliability, sovereignty and predictable operating cost.
The Caveat
This remains advanced nuclear development — not an Amazon delivery schedule.
The Idaho project is intended as a demonstration, and the 2029 commercial deployments remain an announced goal. Reactor licensing, fuel availability, manufacturing, financing, local approvals, cooling, security and construction execution can all move those dates. Aalo has demonstrated real technical progress, including a recent zero-power criticality milestone, but moving from experimental validation to repeatable commercial 50 MWe plants is still a serious industrial undertaking.
The reported $3 billion raise at approximately a $30 billion valuation is also still described as talks, not a completed financing round. Reuters reported that investors expected the final valuation could come in below the headline figure. But even with those caveats, the direction is unmistakable.
The Pattern Was Already Snapping Into Place
I wrote "The First Customer" describing an industrial pattern: government-backed reactor programs create the proving ground, and commercial AI infrastructure follows once the technology is de-risked. Less than 48 hours later, Crusoe and Aalo announced a project that is almost a literal demonstration of that pattern. Federally supported reactor development at a national laboratory, with an AI data center showing up before the bridge is finished.
The AI infrastructure race is no longer fundamentally about who can buy the most GPUs. It is about who can create the most complete, repeatable and financeable machine for turning energy into intelligence. The military may be the first customer. The government may supply the proving ground. But AI is already lining up to be the first commercial passenger — and it isn't waiting for the aircraft to finish certification before it boards.
Crusoe started by finding stranded energy and bringing computers to it. Now it's moving toward creating the energy, building the campus, operating the cloud and optimizing the workload.
That is not merely vertical integration. That is the birth of the vertically integrated intelligence utility. And it started in an oil field, with two guys who named their company after a castaway. The pattern I described two days ago wasn't a prediction. It was already happening. I just didn't know Crusoe was about to prove it.
Rich Washburn is a technologist and strategist working at the intersection of AI, infrastructure, and capital. He is Managing Partner and Chief AI Officer at Eliakim Capital.
Comments