
Hey Futurology! I got exclusive access to Copenhagen Atomics in Denmark to film their molten salt reactor technology.
The paradox: One 40-foot shipping container = enough power for 20,000 H100 GPUs.
Key specs:
– Thorium-based fuel cycle
– Passive safety (gravity-drain to dump tanks, no operator intervention)
– 1960s Oak Ridge technology, abandoned during the Cold War, now resurrected
– Designed for mass manufacturing (gigafactory model)
Trailer: https://youtu.be/PYYZ6fw0dZA
Full 20-minute documentary premieres tomorrow (Thursday) at 5pm CET / 11am ET.
I spent the day walking through their test loops, fuel salt production facility, and manufacturing floor. Happy to answer questions about what I saw.
I spent 9 hours inside a nuclear reactor startup building shipping container reactors to power AI [OC]
byu/yellow_gradient inFuturology
16 Comments
> Copenhagen Atomics plans to run its first nuclear chain-reaction at the Paul Scherrer Institute (PSI) in 2027.
from https://www.copenhagenatomics.com/
> enough power for 20,000 H100 GPUs.
How much is this in banana football fields squared?
Nice to see us finally building things like this to reduce our carbon footprint and ease the burden of high energy bills.
We’re not? It’s so people can make pictures of Taylor Swift with 3 tits?
The malevolent AI of the future will have its own energy sources. Check mate, puny humans.
The thing I’m instantly thinking of is what happens to this highly sensitive equipment should the startup fail, nuclear material is incredibly expensive to dispose of correctly and if the firm goes bust as many startups do then I want to know who’s footing the cleaning bill.
Cool! I look forward to seeing this. The US Department of Energy has opened up National Laboratory land for AI data center leases with the intent of building adjacent Small Modular Reactors (like those in the documentary). DOE has identified ten US based companies working on SMR technologies (source: I’m at a University partnering on a contract).
This sounds like an effort to drum up nuclear hype. Thorium is a painful fuel source – less efficient. And the power metrics are given in terms of graphics chips — maximum AI hype effort.
Small units like this have to got to be insanely expensive per $/kwh LCOE
> The paradox: One 40-foot shipping container = enough power for 20,000 H100 GPUs.Â
I’m obviously missing something, how is this a paradox?
The answer to the question at the end of the trailer “What could stop it”
Permitting. In speaking with the team there earlier this year that was the real world issue. The rest was just plain engineering.
Difficulty in getting even their own government to allow testing (thus PSI)
The NRC is incapable of regulating novel nuclear reactors designs so sadly even if your company manages to make it work and get customers enough to make the company work, this will never take off in the US.Â
I worked on a Nuclear Engineering graduate degree but quit because I just didn’t see a future for nuclear energy in the present regulatory climate. Best of luck to your endeavors. It sounds like you’re hitting all the marketing buzzwords for funding.Â
There’s a “nuclear Renaissance” about every 10 years or so, and it leads to a lot of white papers, research loops etc. But nothing comes from it, the money dries up either without breaking ground or cancelling stuff under production. More data points are collected.Â
I think MIT’s 2017 survey paper on the nuclear industry hit the nail on the head for the future of nuclear technology: stick with well established technologies, identical reactors instead of bespoke reactors and in geographies that require high power density. Basically this is how South Korea’s nuclear industry is doing so well well everyone else flounders. They have like 2 reactor designs, the old one, which they built all the reactors for a decade, and the new one they started building as a refresh after so many years.Â
> The paradox: One 40-foot shipping container = enough power for 20,000 H100 GPUs.
For how long?
So rather than use a new type of power source to help lower the average person’s energy bill you people want to help perpetuate more AI slop that helps exactly nobody except maybe, of course, scammers and people who already have more money than god? I hate this timeline.
This will become the norm.
Cheap and abundant energy can’t be a bottleneck to finding places for datacenters.
That is super awesome, looking forward to checking out the video, thank you!
This is such a waste of resources. So much power just for governments to use AI en Massé to monitor citizens and for those citizens to plagiarize real content to create slop
There have been *a lot* of smr-startups and lot of small reactor designs. These companies always fold when they move from napkin design to actual plans with actual calculations. CA have stuck around longer, but their communications are still incredibly fluffy: Did they explain, how they plan to manage the neutron economy in the reactor? Did they mention protactinium even once?Â