Nuclear's SpaceX Moment

The breakthrough may not be a better reactor. It may be discovering that reactors can be treated like products.

On June 18, in the Utah desert, a nuclear reactor called Ward 250 went critical.

That sentence sounds considerably more ominous than it actually is.

Nobody grew a third arm. Utah remains approximately where we left it. And, as far as I know, there were no surprise mushroom clouds added to the weather forecast.

But something important happened.

https://www.youtube.com/watch?v=FSVSDDAtUd4

Ward 250, built by Valar Atomics, successfully completed a zero-power fueled criticality demonstration under the Department of Energy's Reactor Pilot Program. It became the first DOE-authorized reactor built outside a national laboratory.

Cool milestone.

But I don't think that's the story.

The story may be that nuclear power is beginning to undergo the same structural transition that transformed the commercial space industry.

This might be nuclear's SpaceX moment.

We Already Know How to Split Atoms

The physics isn't new.

We've been producing controlled nuclear fission for more than 80 years. The Navy has operated nuclear-powered ships for generations. Commercial reactors have supplied grids around the world for decades.

The problem with nuclear hasn't been convincing uranium to cooperate.

It's been everything surrounding it.

Traditional nuclear plants became enormous civil-engineering projects with enormous capital requirements, enormous regulatory timelines, enormous construction risk and an unfortunate tendency for the word "billion" to appear several times before anyone generates a watt.

Every plant was practically a cathedral.

Designed for a particular location. Built by armies of contractors. Financed over decades. Connected into giant utility systems.

And when something took twelve years instead of seven, everybody involved got another twelve years older.

That model works.

It just doesn't look particularly compatible with an economy suddenly asking:

"Can you get me another 500 megawatts by Tuesday?"

Welcome to AI infrastructure.

We've Been Watching the Customer Arrive

This part didn't happen overnight.

Back in 2024, I wrote about Microsoft moving to restart Three Mile Island and Google signing agreements around advanced nuclear power.

At the time, the interesting part wasn't nuclear itself.

It was who suddenly wanted it.

Microsoft doesn't have a sentimental attachment to steam turbines.

Google isn't collecting reactors because they look nice beside the server racks.

AI created a new kind of customer: one that needs enormous quantities of reliable power, twenty-four hours a day, and increasingly cannot wait for the traditional electrical grid to provide it.

Since then, the pattern has only become clearer.

I recently wrote about the military becoming The First Customer for microreactors.

The military has exactly the sort of problem these machines solve: remote or islanded power, vulnerable fuel logistics, mission-critical uptime and no interest in discovering that the local utility has a five-year interconnection queue.

Then came The First Passenger.

AI began climbing aboard before the nuclear aircraft had even finished certification.

Data centers and reactor developers aren't merely talking about each other anymore. They're beginning to design infrastructure around each other.

That distinction matters.

Because demand was never enough.

Someone still had to figure out how to make the reactors.

Enter Ward 250

Valar's Ward 250 is a high-temperature gas reactor using TRISO fuel.

On June 18, it reached criticality in Utah under the DOE program. By July 4, four separate advanced reactor projects had achieved zero-power criticality through DOE programs — Antares, Valar, Deployable Energy and Aalo. A fifth, Oklo's Groves test reactor, followed in August.

That alone is remarkable considering the historical pace of nuclear development.

But then came August 3.

Valar announced a $1 billion Series B led by Sequoia Capital, along with a $200 million credit facility.

And the company's stated objective for that money is the sentence that interests me most:

Move from demonstrating an integrated reactor system to producing fleets of them.

There it is.

Not another reactor.

Fleets.

That's not nuclear-project language.

That's manufacturing language.

The SpaceX Comparison

SpaceX didn't invent rockets.

NASA had some pretty good rockets.

What SpaceX fundamentally changed was the industrial model surrounding the rocket.

Build.

Test.

Fly.

Break something.

Figure out why.

Build another one.

Repeat.

That sounds obvious today because we've watched rockets land vertically on drone ships enough times that we've become spoiled.

It was not obvious when billion-dollar launch vehicles were treated like Faberge eggs with engines.

SpaceX turned launch systems from national projects into increasingly standardized, vertically integrated, rapidly iterated commercial products.

And NASA helped create the environment where that could happen.

NASA stopped needing to own every bolt.

It could define the mission, provide technical expertise, establish safety requirements, purchase services and allow private companies to compete over how the hardware actually got built.

DOE appears to be experimenting with a nuclear version of that philosophy.

Its Reactor Pilot Program explicitly created an accelerated DOE authorization pathway for privately developed advanced reactors outside the national laboratories.

That doesn't mean regulation disappears.

Nor should it.

This is still nuclear fission. "Move fast and break things" requires slightly different punctuation when the thing being broken contains enriched uranium.

But the institutional model is changing.

And Valar Is Acting Like a Hardware Company

Here's another interesting breadcrumb.

Earlier this year, the U.S. government put an unfueled Ward reactor aboard a C-17 and flew it from California to Utah.

A nuclear reactor.

On an airplane.

Which is the kind of sentence that makes the 1970s Nuclear Regulatory Commission spontaneously develop chest pains.

But that's precisely the point.

This machine was designed to be transported.

Valar is also participating in DOE's Fuel Line Pilot Program to establish TRISO fuel fabrication capability for Ward 250 and potentially other high-temperature gas reactors.

Reactor.

Fuel.

Manufacturing.

Deployment.

Vertical integration.

Now the SpaceX analogy gets considerably more interesting.

Because SpaceX didn't revolutionize launch by drawing a prettier rocket and sending the blueprints to the traditional aerospace supply chain.

It pulled critical pieces inward.

Engines. Avionics. Software. Structures. Manufacturing.

Control more of the stack and you can iterate the entire stack faster.

Valar appears to understand the same principle.

The Reactor Isn't the Product Yet

And this is where everyone should keep both feet on the ground.

Ward 250 reaching criticality does not mean Valar has solved commercial nuclear power.

The initial demonstration was zero-power criticality. It proved that the reactor could sustain and control the nuclear chain reaction. That's an essential milestone, but it's not the same thing as operating a commercial plant at full thermal and electrical output for years. Those tests matter enormously.

Materials matter.

Heat exchangers matter.

Turbomachinery matters.

Fuel availability matters.

Maintenance matters.

Economics matter.

Licensing matters.

Reliability matters.

Nuclear history contains plenty of brilliant reactor physics attached to machinery that developed more personality than its operators would have preferred.

Atoms are surprisingly cooperative.

Bearings occasionally aren't.

So Ward 250 isn't Falcon 9.

It certainly isn't Starship.

If we're going to abuse the analogy properly, it might be closer to Falcon 1.

Small.

Early.

Far from proving the entire economic model.

But enough to demonstrate that a different way of building the thing might actually work.

That's the Threshold I'm Watching

Because if these companies succeed, the biggest nuclear breakthrough of this generation may not be some exotic new reactor physics.

It may be manufacturing.

Standardize the reactor.

Standardize the fuel.

Standardize the containment.

Standardize the controls.

Standardize the deployment process.

Build one.

Learn.

Build the next one faster.

Learn again.

Then stop thinking about nuclear facilities as bespoke megaprojects and start thinking about nuclear generation as manufactured infrastructure.

That's a fundamentally different industry.

And AI couldn't have arrived at a more convenient time.

Compute Has Changed the Economics of Power

I've written repeatedly that compute is becoming constrained less by chips than by the infrastructure surrounding them.

You can order GPUs.

You can raise capital.

You can pour concrete.

Then someone asks where the next 200 megawatts are coming from and suddenly everybody becomes extremely interested in transmission engineering.

The AI industry doesn't merely need electricity.

It needs dispatchable electricity in specific places on specific timelines.

That creates economic pressure unlike anything the nuclear industry has seen in decades.

A hyperscaler doesn't necessarily care whether its electrons came through 400 miles of transmission line.

It cares that there are enough of them.

Constantly.

Which means the future data center may increasingly become its own little industrial ecosystem:

Compute. Cooling. Power generation. Storage. Networking.

Put the reactor behind the meter and suddenly the data center isn't merely consuming infrastructure.

It is infrastructure.

That was the direction I was pointing toward when Google began making nuclear commitments in 2024.

It became clearer when the military emerged as an early customer.

Clearer again when AI started becoming the first commercial passenger.

Now we're getting another breadcrumb.

Someone is trying to build the aircraft on an assembly line.

Nuclear's SpaceX Moment

I don't know whether Valar becomes the SpaceX of nuclear.

Maybe it's Aalo.

Maybe Oklo.

Maybe Radiant.

Maybe a company nobody is paying much attention to yet.

SpaceX itself wasn't obviously SpaceX when Falcon 1 was blowing up.

That's not really the important question.

The important question is whether the industry has crossed into a different development regime.

Private capital is arriving.

Customers are waiting.

Government is creating accelerated demonstration pathways.

Fuel production is being rebuilt.

Multiple reactor architectures are reaching criticality.

And companies are openly talking about fleets rather than projects.

Those are very different ingredients from the nuclear industry of twenty years ago.

Maybe Ward 250 ends up being a historical footnote.

Maybe it becomes something much bigger.

But I suspect we're going to look back on this period and realize the important breakthrough wasn't simply that another reactor went critical in the Utah desert.

It was that the entire idea of what a nuclear reactor is started changing.

From facility to machine.

From project to product.

From construction schedule to manufacturing cadence.

From one enormous bet to something you might eventually order in multiples.

SpaceX didn't invent the rocket.

It industrialized the hell out of it.

Nuclear may finally be getting ready to do the same.

And if AI keeps consuming electricity at the rate we're currently planning for, it has about a billion reasons per financing round to figure it out.

Just try not to call the first nuclear-powered AI cluster Skynet.

Some jokes really don't need that much fuel.

---

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.