They Sunk the Cloud. Now It Floats on Atoms.

In partnership with

1,000+ Proven ChatGPT Prompts That Help You Work 10X Faster

ChatGPT is insanely powerful.

But most people waste 90% of its potential by using it like Google.

These 1,000+ proven ChatGPT prompts fix that and help you work 10X faster.

Sign up for Superhuman AI and get:

• 1,000+ ready-to-use prompts to solve problems in minutes instead of hours—tested & used by 1M+ professionals

• Superhuman AI newsletter (3 min daily) so you keep learning new AI tools & tutorials to stay ahead in your career—the prompts are just the beginning

Claim your free prompts

They sunk the cloud. Now it floats on atoms.

Fifty gigawatts

That is the size of the hole in the world's power supply by 2030, according to a number Atomarine put on its own homepage and dared anyone to argue with.

Ten years. That is how long the electric grid will make an AI company wait for an interconnection right now. A decade. You could raise the money, build the whole thing, and be obsolete before the utility returns your call.

More than half of every planned data center project has been killed since 2025. Not by hackers. Not by regulators. By neighbors. People with signs at zoning meetings, holding photographs of their own water bill.

So the smartest people in this industry are doing what smart people do when a fight gets expensive. They are leaving the room.

They are taking the computers to sea.

Two men, four MIT degrees, one reactor

Dimitris Koutentakis has an engineering degree from MIT. Four of them, actually — EECS, mechanical engineering, an MEng, and an MBA from Sloan, which is the kind of resume a man builds when he is not yet sure what he is.

Then he went to Arrowstreet Capital, a $200 billion hedge fund, and ran quantitative research on trade execution across $2 billion of daily trades and broker selection across $180 billion a year. Then Amazon, where he designed the optimization math for electric delivery fleet charging and put a $193 million dollar figure on the capital expense he avoided.

He is Greek. Which means when he says the word shipping, he does not mean it the way you and I say the word shipping.

Emile Germonpré has a PhD in nuclear engineering and a master's in chemical engineering. Fulbright scholar. Valedictorian of Ghent University's engineering school. He is the one who actually knows what happens when you bolt a reactor to something that moves.

That is the whole company. Two people. Y Combinator Summer 2026. The partner who signed them is Tyler Bosmeny.

Build it like a car. Float it like a barge.

Atomarine builds data centers the way you build a car: on one line, in a shipyard, over and over. Standardized, plug-and-play barges. 75 to 100 megawatts of compute each. Towed to site. Moored beside the last one. The campus grows from a single barge to 450 megawatts and nobody breaks ground.

The claims on the site are specific, which I respect: four times faster to deploy than building on land. 1.5 gigawatts of capacity per shipyard per year. A power usage effectiveness under 1.1, because the cooling is seawater and the ocean does not send an invoice. Twenty to forty year infrastructure life.

And here is the trick that makes the whole thing more than a stunt. The power is a separate vessel. It moors alongside the compute. Gas turbines today, marine reactors tomorrow. You swap the engine without ever opening the data hall. Nuclear arrives the way a software update arrives.

Now the part the press release deletes

Small modular reactors for commercial shipping do not exist yet. The United States Navy has them. The Navy also does not care what they cost and can burn bomb-grade uranium. A startup cannot do either of those things.

So the first pilot, promised for 2028, runs on natural gas. Which means the clean nuclear story is a promise, and the pilot is a gas plant on the ocean. Every data center developer on earth pulls this move: announce the clean system, run the dirty one for a few years, call it a transition.

Then there are pirates. Real ones. Atomarine's pitch is that it can deploy anywhere on Earth, and its site does not mention security once.

Then there is the small matter of saltwater, which eats metal for a living. Microsoft sank an entire data center off Orkney to test exactly this idea and hauled it back up — the cooling worked beautifully, the maintenance was hell. Data centers at the bottom of the sea have been tried. They did not work.

And this is a barge. It needs a serious internet link. Starlink will not cut it.

The harshest review in existence sits on a blog called Pivot to AI, and its best line is that the plan reads like move fast and irradiate things.

I love it anyway. Not because it will be easy. Because it is the only pitch on the board that behaves like the problem is real. AI's bottleneck was never intelligence. It is a wall outlet, and there are not enough of them.

"If we were doing everything we do, would we choose to do it again from zero?"

— Garry Tan, via TheFoundersPack

Because that is the actual question Atomarine is asking. Not can we add a wing to the data center. Whether you would build the whole thing again from nothing, on water, if you had the choice. They looked at the queue, the zoning board, and the water bill — and answered from zero.

The pack is the point

Here is what gets lost when this reads as a nuclear-at-sea story. It is not one company. It is a pack.

Atomarine is not building the reactor. It is not welding the barge hull. It is wiring together established barge designers and builders, partnerships with multiple SMR vendors, and a standardized platform it can sell to AI operators and hyperscalers. Everyone holds a piece. Everyone's risk goes down. Nobody has to own the whole animal.

That is a keiretsu — the same architecture a pack of Japanese trading houses assembled after the war, and the exact structure in Form Your Own Pack. The shared asset is the shipyard line. The cross-shareholding is a licensing process you only have to run once. And the coordination protocol is the mooring: the physical socket where power meets compute. Pillar three. Amplify with technology. Atomarine's version of technology is a plug on the side of a barge.

Same shape as any pack

Two people, a shipyard, and a bet that the rules can be rewritten. That is the same shape as every pack that ever worked, and it is the same shape as Genflick — an AI filmmaking agent that lets a two-person team shoot a feature instead of renting a crew of sixty. Different ocean. Same math. Fewer people, more leverage, one shared machine doing the heavy lifting. If you are building something the old way and it is killing you, go try Genflick free and see what a pack of one plus an agent can actually finish.

Meet the founders

Dimitris Koutentakis — Founder/CEO. Four MIT degrees (EECS BS/MEng, MS Mechanical Engineering, MBA Sloan). Previously quantitative research at Arrowstreet Capital ($200B AUM) and fleet-charging optimization at Amazon. Greek shipping connections. @_dkout · LinkedIn

Emile Germonpré — Founder/CTO. PhD in Nuclear Engineering, MSc in Chemical Engineering. Fulbright Scholar. Valedictorian of Ghent University's engineering school. LinkedIn

Links: YC Profile | Website | @Atomarine_co | Company LinkedIn | Launch Post | Launch Video | Partner: Tyler Bosmeny