Nuclear got a $1 billion vote from Sequoia, and AI's appetite is why

Here's what gets me about this one. The same week we're all arguing about whether AI's compute buildout has gone off the rails, Valar Atomics closed a $1 billion Series B led by Sequoia at a reported $6 billion valuation. The money is big, but it's not the headline. The headline is that a startup that builds miniature nuclear reactors just watched one of the most disciplined firms in venture put a whole billion on the table. And the reason the round even exists is AI.

What Valar is actually building

Valar makes small modular reactors, the miniaturized, factory-built power plants meant to be cheaper and faster to stand up than traditional ones. In June the company says its Ward 250 reactor powered an Nvidia Blackwell system, and it signed a deal with Nvidia to develop a waterless 30MW AI factory. That last part matters, because data centers drink enormous amounts of water for cooling, and a reactor that can run an AI workload without any of it removes a constraint most people never think about.

The company is also talking about speed in a way that sounds almost too confident. It took two years to build one core, seven months to take another reactor critical, and it frames the pitch as a shrinking tick rate, moving toward producing tens, then hundreds, then thousands of reactors a year. Bloomberg reported the $6 billion figure Valar didn't disclose itself.

The pattern behind the round

Valar is not alone. Antares raised $470 million to build reactors for the US military, and X-energy pulled $1 billion through an IPO. When you stack those up, the money is chasing the same thing: power, and lots of it, exactly where the AI training and inference clusters want to live.

I keep coming back to the tension. A few months ago the conversation about AI infrastructure was mostly about chips and cash, about who was spending how many billions on data centers. Now the bottleneck is quietly moving to the other end of the wire, to how you generate the electricity those clusters will draw. The $750 billion capex question I wrote about last month framed the spend. This is the physical layer under that spend, the part that has to exist before any of the chips turn on.

Why I'm watching this

I genuinely don't know how to feel about the pace. A factory that claims it will eventually crank out thousands of reactors a year sounds like a brochure until you remember how slowly every previous nuclear project has moved, and how hard it is to prove safety matters at that scale. But the capital is real, the Nvidia partnership is real, and the timing is not accidental. When venture money flows this hard into power generation, it usually means someone upstream is already committed to the demand.

The free AI model explosion I covered on this blog is pushing usage up. Free usage and cheap compute push people to run more. More running means more thermals in real warehouses. Valar is essentially a bet that this loop keeps spinning, and that nuclear is the only answer with enough scale and low enough carbon to keep up.

I'll be watching whether that Ward 250 really powers a full Nvidia Blackwell cluster in production, and whether the factory-built promise bends the timeline that has crushed every nuclear startup before it. Right now the story is money agreeing with the physics. That's not the same as the physics agreeing back.