Atomarine: Nuclear Data Centers at Sea | Hacker News Reader

Atomarine: Nuclear Data Centers at Sea

atomarine.co

42 pointsndr77 comments

The one thing I enjoy about our AI goldrush driven economy is all the cartoonish cash grabs coming out of the woodwork to try and part investors from their money.

1. 4-5x faster deployment than on land

- Not that this will ever be built but if it were, I assume Newport News or Electric Boat would be fitting the nuclear reactor (unless they cheap out and go with an import). Given they are booked solid until the 2040s somehow this company is even slower than the nuclear industry at building land based plants.

2. 1.5 GW deployed per year

- Starting in which year?

3. 1.1 PUE seawater cooling, modeled.

- Well there you have it, they modeled it, so it's basically already built.

4. 20-40 year infrastructure life

- Remind me how many 20 year old servers are currently running AI workloads? Oh, I guess that just means the ships and the nuclear reactor. I mean I guess that's good, it's not like a building on land would last that long...

Oh but I guess they will just be using LNG to start with, thank goodness that is alot easier to get than nuclear. In fact now is a really good time to start this as I heard there's some companies looking to offload a few LNG tankers somewhere in the gulf for a steal!

I wonder if the business model is to operate on the high seas, where sanctions do not apply….

I'm not sure what sanctions you are thinking about but generally if you want to be beyond the legal reach of a nation state or group of nation states, you just set up a server in a different nation state that doesn't enforce the laws you want to avoid.

Current nuclear-powered ships are run by the world’s premier naval powers and protected by their own stealth or a huge array of armed escort ships.

Have these guys factored in the cost of a task force of protective missile frigates, subs, and destroyers at each site?

That's a problem for the future, no need to worry investors with these silly questions.

Hey if you can externalize that cost, more money left for you and the investors !!!

I doubt Russian nuclear ice breakers have naval escort. Until recently they operated within the coverage of on-shore long range SAM systems (but most of these SAM systems were moved elsewhere recently e.g. to protect oil refineries so Northern Sea Route is no longer protected).

I don’t think that anyone would want to steal an icebreaker, but a datacenter full of valuable equipment and interesting information is a more likely target.

Would a marine raid by ship be lower risk than a smash and grab by panel-van on land?

I imagine land is a lower bar to entry, accessible to a much larger criminal base, but security response times and road surveillance is tough.

I'm too much of a land-lubber to know how easy it is vanish from a coastal crime scene by ship. The sea... nowhere to hide but very big. Can satellites track you these days? If wanting to nab a container or two that's a big ship. Not much you could put in a minisub or sub-surface thing. A James Bond villain could cut the supports and drag the whole thing to the hollowed out mountain of their island lair.

I need Michael Mann to make "Heat but on a boat" to educate me.

Why would a carrier battle group be a prerequisite for this? Capital ships will have escorts anyways, the propulsion system doesn't matter.

Nuclear submarines often operate without any escort. Russian nuclear icebreakers and floating SMRs also do not have heavy military escorts.

> Nuclear submarines often operate without any escort.

Correct. That's where the GPs comment regarding stealth comes into play.

Attempt a hostile takeover and see how long the response takes. It doesn't matter how far away the assets are if they can either make it to your position before you accomplish your goal or else credibly hunt you down after the fact. I don't think "theft of nuclear material from a russian naval icebreaker" is a realistic possibility.

Don't be a killjoy, the bloom of harebrained ideas around AI should be celebrated!

of course they did not calculate it! it will come later when they want to calculate real costs, they will release it after investor update

Forget security (which worries me less), the legalities of this are notoriously tricky (especially if the ship might ever leave the waters of one nation), and the economics are even worse.

I wouldn’t touch civil maritime nuclear with a barge pole, and if I did, the clear win is decarbonizing big cargo ships.

Agreed, the largest container ships put out some serious pollution, going nuclear (a decades old proven technology for shipping) would be a massive improvement

But the people championing nuclear data centers are not trying to solve climate change

Proliferation resistant reactors should probably be used.

Like.... LFTR

Basically, the normal operation of the reactor doesn't allow the production of weapons-useful nuclear isotopes.

You'll probably also need a smaller footprint reactor

Like.... LFTR

And you'll want something that can be turned on, shut off, and has natural anti-meltdown abilities in case hurricanes or other disasters threaten the ractor

Like ... LFTR

LFTRs make readily-available (liquid-form) weapons-grade Uranium-233 during protactinium separation. They're the least proliferation resistant reactor known to man.

LFTR has a far larger core than a PWR. Neutron diffusion length in water is 5 cm. In LFTR graphite moderator it's 16.

Many reactors including modern SMR LWRs can cool themselves passively without backup power. Especially when floating in seawater.

Highly recommend doing PWRs for this application.

Disclaimer: I'm trying to do this with PWRs.

U-233: is this inaccurate?

https://www.quora.com/Why-do-so-many-people-claim-thorium-br...

I'm no expert on U enrichment, but if there's even just a little U-232 that mixes in with the U-233 then the gamma rays get pretty hard to handle? I guess the breeding process might produce a lot of pure U-233, but again to the point of the quora response... not a lot of U-233 was used in "real" weapons.

I've never heard of MSRs being "large". The ORNL one fit in a closet, but that wasn't a breeder from thorium AFAIK. The penetration doesn't matter much if you have a blanket of Thorium to intercept neutrons.

"Gas today, nuclear tomorrow" in the power section, with a LNG tanker docked to it.

I think the word "today" in that phrase is extremely generous.

How do they deal with saltwater corrosion, do the barges go in drydocks after a couple of years? What about weather do they have their own propulsion do they need to be towed to avoid bad weather? Can't you put the same gear in containers next to the shore and avoid the whole weather issue?

With the current capabilities of drone warfare is literally putting ecological bombs near to the shores. Awesome!

I thought the same... huge security risks (or huge security costs ?).

Genuinely, have these guys done anything other than AI generated ship pictures? The top down ones look like they didn’t even use a good model.

I’d be stunned if anyone takes this seriously

There are some very stupid VC folks around. I have spoken to some, and watched them shart millions away on the back of a powerpoint and some stage presence

Stupid is uncharitable. They lock into the small picture, like some details about it, and get carried away. Then again, it's their job to get carried away.

How do I get in front of these people? I have legit business ideas, and it’s frustrating to hear how easy it is to get VC money. I want a piece of that cake.

Sealand[1] is becoming attractive real estate again...

[1] https://sealandgov.org/

Naa. That's just too rotten. Deep cons need https://en.wikipedia.org/wiki/Condeep

I ran the numbers for this about six months ago and it works out a lot better than all the other alternatives.

The fact you're surrounded by coolant, can use intrinsically safe versions of high pressure reactors and have low latency between multiple markets with one datacenter is a big plus.

That you're not competing with humans for electricity or water is another.

That said you'd want to build a semi submersible platform with the nuclear reactors in the pontoons and the superstructure able to house container based datacenter units. The only place with enough experience to build these is South Korea which isn't a deal breaker per se, but Samsung can pretty much eat your lunch if these get successful.

So I take it you’re a marine engineer or similar?

No, but I did have my father check the maths and he built oil rigs for 30 years.

Can you share your work?

Trust but verify

If you'd like to buy the report drop an email on my profile to talk about price.

Write up an ai generated biz plan, shop it around quick. make the necessary human pyramid, get it on the stock market. it'll be as irresistible as peanut butter & chocolate. (ai+nuclear.. with a dollop of cheaper than spacex AI buildout). Doesn't matter than solar+batteries might be a BIT safer. (even goat Elon has said we don't need nuclear as we already have a free one that is at a safe distance)

Avoiding legal jurisdiction, regulation and tax by being offshore.

Bring on the exploding drone boats.

You also lose the protections given by legal jurisdictions.

Also regulation and tax by being offshore have much easier ways to be bypassed and which are already being done effectively so by utilizing already existing legal structures and jurisdictions I think.

Why the website and the images look like AI-generated?

because they are :))

This reminds me of the Pangeos megayacht from a few years back.

https://www.cnn.com/travel/article/pangeos-terayacht-turtle/

TL;DR: Pangeos was a concept for a floating free city that trended a few years back, but was never built for obvious reasons, just like how these floating data centres will never be built. Presumably some people were bilked into paying money to the Pangeos scammers, just like there will be those who "invest" in this.

The problem is that it costs a lot more to put a nuclear reactor into a permanently sea-worthy form-factor than it does to build on land and put up with some pesky taxes. There are plenty of places with lax regulations (environmental and otherwise) that will be happy to let you build whatever size data centre you want if you're willing to build the power generation too.

Sure, lets heat up the oceans even more.

The sensible way would be to use the waste heat for district heating, but let's heat the fish instead

It's a drop in a bucket. Global warming is due to ~5e18 Wh of excess heating. A nuke dumping 5e12 Wh of additional heat into water makes no difference at all.

Drops add up, just like using gas, oil, coal...

No, they really don't in this case. IIRC a sustained gigawatt of output over a 24 hour period is roughly equivalent to the sunlight absorbed by a 5 square kilometer patch of ocean in the tropics over the same period of time. The oceans absorb an almost incomprehensible amount of solar energy on a daily basis.

Wasn't there a Microsoft idea to build data centres in shipping containers a few years ago, again using the ocean for cooling?

What happened to that?

Given the expense of the latest AI chips, or just the RAM for them, the price of land, cooling and power is not top of the expenses bill, so all talk of sea-based or space-based data centres has to be regarded as having jumped the shark.

As of yet I have not seen the use of server racks as structural architectural features. They always seem to need a vast shed that has its own supports, when server racks could also be the walls and supports for much of the building, or some other structure, maybe a road bridge.

The price of Datacenter are the AI chips but the bottleneck in creation of them is power, given how much power exhaustive they are (and rightfully so) people protesting against it because its them who pays the price.

That being said, the shipping container idea is decent but the bottleneck is still power not so much cooling, so unless someone figures out a way to have an actual nuclear power reactor, there are issues.

Also oh the issue isn't that there aren't nuclear power reactors but that they aren't large enough. Some nuclear powered datacenters exist in some experimental settngs but these new Datacenters are trying to take a magnitude more and they can't wait for the production of those things and further research.

>Microsoft idea to build data centres in shipping containers a few years ago, again using the ocean for cooling

Project Natnick!

It was a success, the research concluded and it is no longer active.

Article: https://news.microsoft.com/source/features/sustainability/pr...

Wiki: https://en.wikipedia.org/wiki/Project_Natick

In June, China opened a $236 M wind powered underwater data center. Article references another commercial underwater data center they opened in 2023.

https://www.wired.com/story/china-opens-worlds-first-wind-po...

It was a success? then why it not scaled up instead of shut down

So Much land. Even non civilized lands, Even near sea for piping water. What does this idea solve? Or just adds problems like, Non secure assets/nucclear on the sea.

freedom from laws

You can not move datacenter built on land. But you can tow this datacenter to new direction if needed. Just like "Howl's Moving Castle" :)

If this is just designed for AI economics. The end output of these datacenters are the tokens which are mostly relatively really small data and is not sensitive to latency.

> What does this idea solve?

The obvious answer is NIMBYs.

This is a really ambitious and difficult idea. It would be awesome if they can pull it off!

Since most people live near a coast this could power more than just data centers.

Am I the one who's out of touch, or is this absolutely insane? Let me summarize the state of the AI black hole right now:

Data centers! Everyone wants more data centers! But how do I get in on the craze? I think the brightest investors of our generation are hungry for data centers placed in weird places. Space data centers, underwater data centers, data centers in the Earth's core and on a space elevator! I got it - sea data centers! Now, to make this idea not sound crazy, we'll just pretend there's literally not a single shred of buildable land on the entire planet. So now we've established that land data centers suck, all we need to do is to figure out the easy parts, such as shipbuilding, propulsion, creating our own power grid (no worries, we'll just pretend that designing, building, maintaining and scaling a state of the art nuclear reactor at sea is easier than building a power line on land or even just constructing a normal land power plant), repairs, staffing, scaling construction (got to make building ships cheaper than a warehouse) and about a billion other things. I think we'll just try to think of the smallest numbers we can, no one will check. Oh, and since that nuclear thing is so hard, we'll actually just get tons of fossil fuels instead and pollute away in international waters, we promise we'll switch to that miracle nuclear mega reactor we're building! Now all that's left is to AI generate us some text, AI generate some images, AI generate us a website and that's it! YC26 baby!

Did I get that right? This is like investing into a perpetual motion machine on Kickstarter level crazy. Am I still living on the same planet?

I think you left off figuring out how to connect a floating data center to the internet without making that aspect as expensive as powering the thing in the first place.

I think that's actually the easiest part of this entire thing? Assuming you're anchored you just dangle a subsea cable directly underneath the platform.

Note that there are already some pilot ocean based datacenter projects. The most recent one I'm aware of is attached to nearshore wind. I have to imagine that makes far more sense than floating a civilian nuclear reactor on the high seas.

What about (extreme) weather? The nice pictures looks like being in open sea, and there big storms may have an expensive impact if not everything is taken into account.

What's the benefit from using an off shore platform against building them on the beach and using pipes for cooling?

Zoning laws and jurisdiction, perhaps.

Accident waiting to happen... who's gonna insurance these facilities?

Privatize the earnings, socialize the risk. The neoliberal dream.

Translation: they won't let us build datcenters on land because we screw things up too much there. And we can't afford space. So we'll do it at sea where we can run all the gas turbines we want and wreck the Earth from there.

See panthalassa as an alternative

How do they handle rough seas?

who's ready for gpu piracy?

The question now becomes: Will the AI bubble pop before we get sea datacenters?

YCombinator S26?

Guys, guys, guys... so hear me out - remember last month when we were all giddy about datacenters in space? Well, apparently investors weren't too keen on that idea so we've come up with something much better: Nuclear datacenters at sea!

Now this is truly a visionary proposition that will certainly become another unicorn and have no issues whatsoever.

Why not fusion powered data centers on the moon?

We just need to solve the issue of fine moon dust in those computer fan bearings /s

Up next: nuclear data center Zeppelins in the cloud!

Innovative new air cooling: instead of using lots of tiny fans to move air through the servers, we simply use 2 big fans to move the servers through the air!

And while we're at it, we'll completely disrupt the existing Zeppelin market that depends on expensive helium for some reason. We use daring, innovative hydrogen technology, enabling a massive cost reduction!

the term zeppelins needs to be used more often.

love this idea, where do i mail my money to invest!!!!!

This is future technology my friend. They only accept payments in crypto, tulips or SpaceX stock.

The current major roadblocks for any datacenter are, planning permissions, power connections, and cooling. The on a barge approach solves all of the above, unlike space which makes all of these problems worse.

Its power connections which is the major roadblock and as such obviously planning permissions get tricky because as we see in so many cases, so much grid electricity is used that that the planning permissions become tricky and rightfully so.

As many countries are now trying to suggest, bring your own power and we are happy.

So in that sense, Nuclear? sure, already being tried but not fast or big enough right now.

Sea? I don't think so. You literally get into murky waters on the future of the project and this part feels really hype based, atleast to me.

Perhaps some people feel that Sea could cool things off but in my understanding, if you already have nuclear, it would be better building Datacenters in colder stranded regions then to have really effective temperature/natural cooling.

Power is the core issue, Everything else is bolted on top of it in my opinion.

Reply on news.ycombinator.com