The Merits of Maritime Data Centers

In May 2026, a startup called Panthalassa raised $140m led by Peter Thiel at a reported $1bn valuation (GeekWire). Its product, in development, is a self-propelled platform that drifts in the open ocean, generates its own electricity from the motion of the waves, runs racks of GPUs on board, and beams the results back to shore over Starlink. It is a data center that floats in the Pacific.

A few years ago, this might have read as science fiction, but today it is a response to the single biggest constraint on AI: power. In our earlier piece, Satellites + Data Centers, we looked at orbit as one answer, due to abundant unobstructed solar power and free cooling from the vacuum of space. The ocean offers similar advantages, its own power from wind and waves and free cooling from seawater, and deployment does not require fighting gravity with rockets for initial deployment. In fact, groups are already running compute at sea at commercial scale.

The label "maritime data center" turns out to cover several different bets. Below, we walk through why the ocean is suddenly attractive, the companies chasing it, the design decisions that separate them, and the problems that no amount of capital can make disappear.

Why the Ocean, Why Now

Electricity is now the biggest bottleneck for AI. Connecting a large new facility to the power grid in the United States takes roughly 5-8 years, because every project has to wait in the interconnection queue, the line that grid operators use to study and approve new connections one at a time (Lawrence Berkeley National Laboratory, LBNL). The demand on the other end is enormous: data centers could consume up to 12% of all U.S. electricity by 2028, according to the Department of Energy and LBNL, with the Electric Power Research Institute (EPRI) projecting a range of 9% to 17% by 2030. Our existing grid was not built for the demands of AI.

So developers are increasingly going around it. More than half of gigawatt-scale data center projects now plan to use on-site or hybrid power rather than wait in the grid queue (Morgan Lewis). The second pressure point is water usage. Traditional data centers consume large volumes of water for cooling, which is an increasingly contentious issue in dry regions; albeit top-tier data centers typically leverage reclaimed or non-potable water and are closed systems that do not need continuous supplies of water.

The Ocean offers a solution for both of these at once. Power can be generated on-site from offshore wind or wave energy. Cooling comes free from an effectively infinite cold seawater heat sink with no fresh water consumed. Sitting offshore sidesteps land acquisition and the grid queue entirely.

One Label, Several Different Bets

Every maritime data center has to solve the same two problems: powering the servers and cooling them. The ocean helps with both, but not equally, and that gap is what actually separates these companies.

Cooling is the easy part, and nearly everyone claims it. Seawater is a near infinite cold heat sink, so any facility in or on the ocean can pull heat off its racks with little or no fresh water. Orca Connex, a Norwegian floating data center team, is leaning into seawater’s role as a heat sink, citing a 1.06 Power Usage Effectiveness (PUE, the ratio of total facility power to the power delivered to the IT equipment, where 1.0 is perfect) across 40-120 MW facilities with zero fresh water use. NetworkOcean, a YC company, makes a similar pitch claiming seawater cooling cuts power use by up to 40%.

Power is where the field actually splits, because it is the harder problem. Some players bring their own generation to sea, which makes them as much energy companies as data center operators. Panthalassa, for example, generates wave power and runs AI inference on top of it, cooling its sealed servers with cold deep sea water; the energy source is the real innovation. Aikido Technologies repurposes distressed and stranded floating offshore wind sites, pairing a 15 MW class turbine with roughly 10 to 12 MW of compute per platform (New Atlas). These companies are building the power plant and the data center as a single object.

From Scrappy Startups to State-Scale Reality

What makes this space interesting is not any single startup, but the spectrum of players now in it, which runs from venture seed rounds all the way to national infrastructure.

At one end are the relatively early startups: NetworkOcean, Orca Connex, and Aikido, for example. A step up in funding is Panthalassa, whose $140m round makes it one of the best funded startups in the space currently. Further along, some existing players are already generating revenue with water surface data centers: Nautilus Data Technologies (a data center cooling company) has operated a commercial floating data center on a barge in Stockton, California since 2021.

Mitsui O.S.K. Lines, one of the world's largest shipping companies, has signed two separate floating data center deals, one with the power ship operator Karpowership and another with Hitachi to convert second-hand vessels into data centers (Data Center Dynamics). Samsung Heavy Industries, the South Korean shipbuilder, has won class certification for a 50 MW floating data center design (Smart Maritime Network).

At the far end is China, which has moved significantly past pilots. Beijing-based Highlander launched the world's first commercial underwater data center off Hainan in 2023, with China Telecom and Tencent as customers. Its affiliate, HiCloud, now runs a 24 MW offshore wind-powered underwater facility off Shanghai that reached full commercial operation in 2026, housing nearly 2,000 servers about 35 meters below the surface and cooled passively by seawater (Tom’s Hardware). This is a working facility at meaningful scale, built with state backing.

Floating or Submerged: The Critical Choice

The single most important decision in this space may not be the power generation method, but rather, whether the data center floats on the surface or sits sealed beneath it.

The case for going underwater is strong on paper. Submerged pods get the best cooling, the best physical security, and remarkable reliability. Microsoft proved this with Project Natick, a research effort that ran 864 servers on the seafloor off Orkney, Scotland, for two years and found them eight times more reliable than an identical land-based control group. A result it attributed to the sealed, oxygen-free environment (Microsoft).

And yet Microsoft shut Natick down in 2024 and confirmed it would build no further subsea data centers. Microsoft did not officially blame the technology, but analysts point to the obvious flaw: you cannot service or upgrade hardware inside a pod welded shut on the ocean floor (Tom’s Hardware). For AI workloads, where the GPU generation turns over every 2-3 years and the chips are the whole point, an asset you cannot refresh is a depreciating liability. Reliability does not help if the hardware is obsolete before it fails. It is the same headwind we flagged for data centers in orbit, where maintenance is the unresolved cost: the harder a facility is to reach, the harder it is to keep current.

This is why many implementations are drifting toward the surface. A floating data center can be serviced and upgraded like any land facility while still capturing seawater cooling and offshore power. It trades the durability of the deep ocean for the practicality of accessibility. China's bet runs the other way, that state capital can simply absorb the refresh problem at national scale. Most private players cannot, which is why floating, rather than submerged, data centers look like the most viable commercial path.

The Problems Capital Does Not Solve

Even the floating thesis carries real friction. Marine permitting is slow, fragmented, and politically charged: a single project can answer to multiple agencies covering coastal development, water quality, and wildlife, and discharging waste heat into a bay or estuary raises genuine environmental questions, including whether thermal discharge could disturb marine life or trigger algae blooms.

Beyond regulation sit the engineering realities: salt corrosion, clogging from mineral scaling and biofouling (accumulation of microorganism and other marine life), the cost and difficulty of repairs at sea, insurance, latency back to shore, and physical and cyber security for assets floating in open water. And the underlying economics remain unproven. Even Panthalassa's backers face open questions about whether wave energy can be made cheap enough (Latitude Media).

The same uncertainty hangs over the power source many see as the eventual answer: Ocean Thermal Energy Conversion (OTEC). OTEC generates electricity from the temperature gap between warm surface seawater and cold water pumped up from the deep, using the warm water to boil a low-boiling-point fluid that spins a turbine and the cold water to condense it again. Its appeal for compute is twofold. Unlike wind, waves, or solar, the ocean's thermal gradient is constant, so OTEC could in theory supply steady, around-the-clock baseload power, and the same cold deep water that drives the turbine can also cool the servers. But it is unproven at scale. The temperature difference is small, which makes the process inefficient and forces plants to move enormous volumes of water through kilometer-long pipes, and it only works in deep tropical waters. No one has built a commercial OTEC plant, let alone one paired with a data center.

Takeaways: The ocean is drawing capital because it eases the two things holding back AI data centers: it provides nearly free cooling from seawater and a way to generate power on site, without the multi-year wait to connect to the grid. But cooling is the easy half, and seawater gives it to almost everyone. The harder and still unproven question is whether generating your own power offshore can be made cheap and reliable enough to beat building on land.

The second dividing line is serviceability. Submerged facilities cool well and run reliably, but hardware sealed on the seafloor cannot be upgraded, and AI chips are replaced every few years. We expect the lasting commercial businesses to be floating and sited near shore, serviceable like any other data center, while submerged designs are relegated for workloads that never need new chips.

Whether the ocean becomes a permanent tier of compute or only a bridge until the grid catches up is the same question we asked of data centers in space. With grid connections still measured in years, the demand for alternatives is not going away soon.

‍