Big Tech just paid $1 billion for someone else to solve their power problem.
The Summary
- Valar Atomics closed a $1B Series B led by Sequoia, plus a $200M credit facility, to build what they call America's first nuclear "gigasites"
- The company is vertically integrating: they're building reactors, owning them, and selling power directly to data centers
- This is venture capital betting that AI training clusters need dedicated nuclear baseload, not grid power
The Signal
The math is simple. A frontier AI training run now burns 100+ megawatts for months. Valar Atomics just raised $1.2 billion to build dedicated nuclear plants next to data centers because the grid can't handle what's coming next. Sequoia didn't write a ten-figure check for climate virtue signaling. They wrote it because hyperscalers are already rationing compute.
The "gigasite" model is new. Traditional nuclear projects take 10-15 years, cost $10-30 billion, and sell into wholesale power markets. Valar is building modular reactors on accelerated timelines with a single customer locked in before ground breaks. They own the reactor, the land, and the transmission infrastructure. The data center operator gets guaranteed baseload power at fixed rates for 20 years.
"Big Tech outsourced their power problem to a startup that can move faster than utilities."
Why now? Three converging forces:
- Training clusters are getting denser. GPT-5 scale models need 500MW-1GW of continuous power. That's a small city's worth of electricity running 24/7.
- The grid can't scale fast enough. Utilities are telling hyperscalers to wait 5-7 years for new transmission capacity. AI labs are racing to AGI on 18-month cycles.
- Nuclear permits are accelerating. The NRC has approved three new small modular reactor designs since 2024. What used to take a decade now takes 3-4 years if you site it right.
Valar's vertical integration is the key insight. They're not selling reactors to utilities. They're not waiting for Exelon or Duke Energy to build gigawatts. They're becoming the power company, the developer, and the landlord. That's why Sequoia led this round. The margin structure looks more like infrastructure software than energy.
The credit facility matters too. Lenders don't put $200 million into construction debt unless there are signed offtake agreements. Valar likely has commitments from at least two hyperscalers already. They're just not announcing names yet.
The Implication
Watch for two follow-on moves. First, hyperscalers will start buying equity stakes in these nuclear developers to lock in capacity. Microsoft and Amazon already have nuclear partnerships. Google and Meta will announce theirs this year. Second, this model goes global fast. China and the Middle East are building AI clusters faster than the US. Whoever cracks modular nuclear deployment wins the next decade of AI infrastructure.
If you're building AI agents, price in energy scarcity. Model training costs are about to spike unless you have a direct power purchase agreement. The compute advantage goes to whoever secures gigawatts first.