A startup called Atomarine, backed by Y Combinator's S26 batch, wants to build data centers on floating barges. Each barge would carry its own power generators and use seawater for cooling, so it would not need to connect to the regular electricity grid.

According to the company's website, published July 30, 2026, each "compute barge" would carry 75–100 megawatts of capacity — roughly the power use of a small city. The barge is built in a shipyard, towed to a location, and anchored in place. To add more capacity, you tow another barge next to the first one and link them up. The company claims this approach is 4–5 times faster than building a data center on land, with a goal of deploying 1.5 gigawatts of capacity per year.

The problem Atomarine is trying to solve is straightforward. AI applications require massive computing power, and the specialized chips that run them can be installed in months. But getting enough electricity to power a new data center — connecting it to the grid — now takes four to seven years in the United States. The company also states that the U.S. is projected to fall 35–49 gigawatts short of the power needed by 2030. That gap between how fast chips arrive and how slow power infrastructure gets built is the reason Atomarine wants to put both the power plant and the data center on a barge.

The design is modular. One barge starts at 75–100 megawatts, and you can add up to six barges to reach 450 megawatts without building anything on land. The company says seawater cooling will keep the facilities efficient — a metric called PUE, which compares total power used by the building to the power that actually reaches the computers, would be about 1.1, meaning very little power is wasted on cooling and other overhead. In many land-based data centers, that number is above 1.3. The deep seawater absorbs the heat the computers produce, replacing the energy-hungry air-conditioning systems that land data centers rely on.

The first barges will not run on nuclear power. They will use natural-gas turbines — a mature, widely available technology. The nuclear piece is a future plan: when small nuclear reactors designed for ships become commercially available, a nuclear-powered barge can replace a gas-powered one in the same spot, and the whole operation switches to carbon-free energy. The company's co-founder holds a PhD in Nuclear Engineering (Y Combinator), which gives the nuclear plan some credibility, though the reactors it depends on do not yet exist in commercial form.

Atomarine is looking for AI companies, shipbuilders, energy partners, and investors to work with, and can be reached at info@atomarine.co.

This is not the first time someone has put a data center in the ocean. Microsoft's Project Natick tested small underwater data-center pods off the coast of Scotland. But those pods were small and drew power from shore. Atomarine's barges are much larger, sit on the surface, and generate their own power — that is the key difference.

The regulatory question is more complicated than it might seem. A barge anchored in coastal waters is governed by maritime law, Coast Guard rules, and environmental reviews that apply to the ocean — not the rules that govern buildings on land. Whether that path is actually faster than getting permits on land, or just a different set of hurdles, is something the company has not yet demonstrated. The 4–5× speed claim is a projection, not a proven result.

The efficiency number deserves a closer look, too. Seawater cooling can make a data center very efficient by one measure, but that measure does not account for the emissions from burning natural gas on the barge. A data center running on natural gas produces carbon emissions that a data center running on grid power in an area with lots of wind and solar might not. The eventual switch to nuclear would address this, but on a timeline the company does not specify.

The goal of deploying 1.5 gigawatts per year is ambitious. That would mean building and launching 15–20 barges annually, each the size of a small power plant. Shipyards do have experience building offshore platforms for the oil and gas industry, but producing this many specialized barges each year is a significant challenge.

Starting with natural gas is a practical choice. The technology exists, the fuel is available everywhere, and it lets the company start operating before marine nuclear reactors are ready. When reactors do become available, swapping one barge for another is simpler than rebuilding from scratch. The bet is that the floating model works well enough on gas to attract customers who need computing power now, and that nuclear reactors arrive before the gas-powered barges wear out.

Whether that bet pays off depends on execution the company has not yet shown. The claims are projections, not operational results. But the problem is real — the gap between how fast AI chips arrive and how slowly power infrastructure gets built is holding back computing growth. The shipping industry has spent decades building heavy, self-contained platforms for the ocean. The idea is unconventional, but the pieces it relies on are things the world already knows how to build.