Small-Reactor Stocks Are Soaring on AI Hype. But None Has a Reactor Running Yet. Buy, or Bubble?
Updated September 15, 2026
So I went down a rabbit hole trying to buy a piece of the AI-nuclear boom, and I hit a wall on step one. Not a single one of these small-modular-reactor companies has a reactor actually making power. The demand under them is real — data centers are starving for electricity, and everybody knows it. But you're buying a stock today for power that, best case, shows up around 2030. Right now it's mostly momentum. The reactors are still on paper.
The part that stopped me cold: nobody's flipped one on yet
I figured at least one of these companies had a working reactor and the stock was pricing in the next ten. Nope. Every name in this basket is pre-commercial. The most established one has a reactor design the government has actually certified, which sounds like a big deal — and it is — but it still doesn't have a signed, funded U.S. order to point to. The buzzier one, the microreactor company that had Sam Altman on its board early, is building its first little powerhouse at a national lab in Idaho and hasn't sold a watt of commercial power.
Here's the picture that stuck with me. One of these companies booked about as much revenue last quarter as a decent taco truck clears over a good summer — while the market still tags the whole thing at billions. That's not a business you're buying. It's a promise, priced like a business.
Why the story is so easy to swallow
The pitch writes itself, and that's exactly why it's dangerous. AI keeps growing, so data centers need more power, so somebody has to build clean power that runs day and night, so small reactors get built, so these companies make money. Every link in that chain is true on its own. I couldn't poke a hole in any single step.
The hole is the clock. The analysts I read don't see real small-reactor adoption picking up until around 2030, and the big energy agencies peg the first commercial units in that same neighborhood — while the AI power crunch is happening right now, this quarter, with shovels already in the ground for gas plants and grid upgrades. And nuclear has a nasty habit of running late. One flagship U.S. project already slid its target years to the right. Russia's floating reactor took more than a decade to make power instead of the couple years promised. Nuclear runs on nuclear time. The stock chart runs on hype time. Those two clocks are not friends.
The size problem that never makes the thumbnail
Then there's the thing nobody puts in the flashy video. A single giant AI campus can crave the output of several full-size power plants. One of these starter reactors makes a sliver of that. To actually feed a hyperscale data center you'd need a whole row of them, built and licensed and running — and we just covered how none are running.
The deals everyone waves around are softer than they look, too. A handshake to deliver power two decades out. A campus partnership. A letter of intent. Mostly non-binding intentions, not orders with real money locked in. And a commenter on one of these stock videos said the quiet part out loud: what happens to a reactor company if the AI buildout it's betting on turns out to be the bubble? Because then you're not making one bet. You're stacking two on top of each other and hoping both land.
So — buy, or bubble? And where I'd go next
Straight answer from my homework: it's kind of both, and which one it is depends on your stomach and your calendar. These aren't vaporware. The tech is real, the electricity demand is real, and the government is pushing hard to get reactors built faster. If the timelines hold, some of these names could look cheap later.
But the price today already assumes a lot of things go right, on a schedule the nuclear industry has almost never hit. That's the definition of momentum — and momentum cuts both ways. This whole group already gave back most of its run this year while the AI story only got louder, which tells you the market is starting to separate the story from the business. If you buy, go in knowing you're buying a 2030-ish payoff with a lot of bumpy road in between, not a power company.
I went way deeper on how this whole grid scramble actually fits together — where the gas turbines, the transformers, the substations, and yeah, these little reactors really sit in the AI power story — and put the whole thing on video. If you want the rabbit hole with pictures instead of just my notes, come watch it and subscribe over on the Byte Bungalow channel. That's where I'm untangling the rest of it.
Common questions
Does any small modular reactor company actually have a reactor running yet?
Why are nuclear stocks like Oklo and NuScale falling in 2026 if AI needs so much power?
Can one small reactor actually power an AI data center?
What happens to these stocks if the AI boom slows down?
Which is more established, Oklo or NuScale?
Keep reading
The Data Center Is Already Running. Can a Town Ever Force It to Shut Down?
The vote's over and the data center is humming. Can a town still force it offline? The levers that actually work aren't the ones you'd guess.
Sleep deprivation is banned as torture — so why can a data center legally hum you awake every night?
Keeping a person awake is a recognized form of torture. A 24/7 data center hum does the same thing and only answers to a zoning decibel line. Here's the gap.
The Boring Way to Bet on the AI Data-Center Boom (Why Pros Skip the Reactor Stocks)
Newcomers chase flashy AI-power and nuclear startups. Experienced investors keep pointing at boring regulated utilities instead. Here's the reason why.
Everyone's Using 'Government Money' to Pick a Nuclear Stock. I Followed the Money and It Went Somewhere Weird.
Does federal backing make one small-reactor startup the safer AI-nuclear bet? I traced the actual DOE checks. They didn't land where the comment threads think.
Why Can't You Just Vote the Data Center Down? Sometimes You Can.
In about half the states, residents can force a data center onto the ballot and vote it down. In the rest, one legal word quietly takes that option away.
A Data Center Promises Helium Cooling, No Water. Should That Settle Your Vote?
Helium-cooled data center, no water? The water worry can shrink — but the grid, your power bill, and a scarce gas don't. What to check before the vote.
Sources
- fool.com — Oklo down ~45% and NuScale ~36% in 2026 despite the AI-nuclear narrative; both remain pre-commercial. NuScale is the only company with a US NRC-certified SMR design (77 MWe VOYGR l
- ebc.com — NuScale generated $31.5M revenue last year (down from $37M), net loss $355.8M; stock ~78% below Oct peak of $53.43. IEA expects renewables + natural gas to cover near-term data-cen
- thebulletin.org — Natrium target pushed from "5-7 years" (2020) to 2031; est. cost $9.4B for 345 MW. Russia's KLT-40S took 13 years from construction start to power vs expected 3. China HTGR took >2
- fool.com — Oklo + NuScale shed a combined 85.3% this year. Bank of America doesn't see real-world SMR adoption until 2030 at earliest, meaningful adoption 2035-2040. NuScale reported just $75
- tradingview.com — Oklo 2026 10-Q: revenue $1.21M, net loss $48.54M, EPS ($0.28). Consensus FY2026 sales ~$1.548M. First criticality of Groves Isotope Test Reactor Aug 5, 2026 with DOE startup author
Byte Bungalow
I do the homework out loud.
If this is the kind of thing you want checked against the documents instead of the hype, the video's above — and there's a new one every week.
Subscribe on YouTube →By Byte Bungalow. Home power and home tech, checked against the documents instead of the hype. Independent commentary; not affiliated with any manufacturer, utility, or builder named here. Not professional electrical advice.