Ocean Compute: The Emerging Landscape of Subsea, Floating, and Offshore AI Infrastructure

← All Articles

Ocean compute is moving from isolated experiments into a recognizable infrastructure category.

Companies are placing computing equipment beneath the ocean, aboard floating platforms, alongside offshore power generation, and inside autonomous systems designed to operate far from land. These projects are often grouped together under terms such as underwater data centers, floating data centers, subsea cloud, offshore AI infrastructure, and compute at sea.

The terminology can make the market appear more uniform than it is.

A floating data center inside a protected river is not solving the same problem as an autonomous wave-powered compute node in the open ocean. A sealed underwater capsule is not the same infrastructure model as a shared seabed platform designed to support independently retrievable compute modules.

The companies in this category differ across several fundamental dimensions:

• Whether infrastructure is above or below the surface
• Whether it is positioned near demand or near energy
• Whether technicians can access equipment during operation
• Whether individual compute units can be retrieved
• Whether hardware can be refreshed independently of the marine structure
• Whether networking depends on terrestrial fiber or satellite systems
• Whether capacity begins at kilowatt, megawatt, or campus scale
• Whether the project is a prototype, trial, operating facility, or commercial fleet

This article maps the major ocean-compute architectures and the companies publicly pursuing them as of 2026.

It is not an investment ranking. Public evidence varies significantly between companies, and many of the projects remain at an early stage. The objective is to clarify what each company is building, what has been publicly demonstrated, and which technical and commercial questions remain unresolved.

What Counts as Ocean Compute?

Ocean compute includes physical computing infrastructure that uses a marine environment as part of its deployment, cooling, power, networking, or operating model.

The category can be separated into six major approaches.

Modular Subsea Infrastructure

Shared, long-life seabed infrastructure supports multiple compute modules that can be installed, monitored, retrieved, serviced, and replaced independently.

Primary company: Seabase.

Submerged Capsules and Monolithic Units

Servers, networking, cooling systems, and supporting equipment are sealed inside a single vessel or container that is deployed underwater.

Companies and projects: NetworkOcean, Subsea Cloud, Microsoft Project Natick, and several Chinese underwater projects.

Floating River and Port Data Centers

Compute remains above the waterline aboard a barge or floating facility, usually inside a protected industrial waterway.

Companies: DENV-R and Nautilus Data Technologies.

Floating Generation-Integrated Compute

Large floating data halls are paired with offshore wind, gas turbines, battery systems, or proposed marine nuclear generation.

Companies: Aikido Technologies and Atomarine.

Autonomous Offshore Compute

Smaller self-powered nodes operate farther offshore, typically using wave energy and satellite communications.

Companies: Panthalassa and Mocean Energy's proposed Blue Core platform.

State-Backed Commercial Underwater Clusters

Underwater computing systems are developed with direct participation from large industrial, telecom, energy, and government organizations.

Examples: Hainan and Shanghai Lingang projects in China.

Companies and Projects Shaping Ocean Compute

The following table summarizes public positioning as of 2026. It is a map of architectures and open questions, not a ranked list.

Seabase: Modular Subsea Infrastructure Near Coastal Demand

Seabase is a demand-first, power-source-flexible subsea compute platform using terrestrial fiber.

Seabase is developing modular subsea infrastructure for AI capacity positioned near coastal demand. The platform is power-source flexible and connected to terrestrial fiber rather than depending on a proprietary offshore generation system or satellite-first networking.

Its architecture separates the long-life subsea platform from the shorter-lived compute equipment it supports. That design logic is developed further in Beyond Project Natick and Floating vs. Subsea Data Centers.

Independently retrievable modules are intended to allow capacity to be:

• Added in stages
• Isolated at the module level
• Retrieved for inspection or service
• Reconfigured for customer requirements
• Refreshed across hardware generations
• Distributed across multiple foundations and regions

Seabase is therefore pursuing a different operating model from both sealed monolithic capsules and floating data-center vessels.

The company is also developing Nori as the capacity, reservation, monitoring, and regional-placement layer above the physical infrastructure. The objective is to make distributed hardware understandable and usable through a common customer interface.

Seabase remains in engineering and validation. The platform must still demonstrate integrated thermal performance, marine handling, power and fiber interfaces, module retrieval, environmental monitoring, manufacturing, and long-term operating economics.

Its stated differentiation is the combination of four attributes:

1. Modular subsea infrastructure
2. Near coastal demand
3. Power-source flexible
4. Connected to terrestrial fiber

NetworkOcean: Mixed Floating and Submerged Positioning

NetworkOcean entered the category as a Y Combinator-backed company proposing submerged data-center capsules.

Its announced San Francisco Bay test created early regulatory scrutiny. Reporting found that the company had not sought permits from key Bay regulators before publicly discussing the proposed test. Regulators subsequently indicated that authorization could be required.

Public evidence does not clearly establish that the announced full capsule became a permitted commercial underwater deployment.

The company's current public positioning includes both submerged capsules and floating data-center platforms. This broadens its possible market but also creates questions about which architecture is its primary product and which systems have completed physical qualification.

Important questions include:

• What marine hardware has been built and deployed?
• Which tests were completed under appropriate permits?
• Is advertised GPU capacity physically installed in ocean infrastructure?
• Is the long-term product floating, submerged, or both?
• How are hardware failures and refreshes handled?
• What environmental and marine-operating evidence is available?

Subsea Cloud: Subsea Facility and Trials

Subsea Cloud markets standardized underwater compute units and presents its platform as suitable for AI and high-performance computing.

Its clearest public program is Project OTTO in Norway, which offers customers time-limited trials involving underwater space, power, cooling, and supplied or customer-provided hardware.

The company states that CPU or GPU systems may be supported. Public materials do not clearly identify a named production AI customer, an operating commercial GPU cluster, or a long-duration production AI workload.

Subsea Cloud is therefore best understood publicly as a subsea facility and trial platform with AI positioning.

Important questions include:

• What exact hardware has operated underwater?
• Which workloads were trials and which were commercial production?
• What customers have used the system?
• How are server-level failures handled?
• How is hardware refreshed?
• What independent evidence supports the company's performance and deployment claims?

DENV-R: Floating Regional Cloud Infrastructure

DENV-R operates a floating data center on the Loire River in Nantes.

Its architecture preserves direct access to computing equipment while using a river or port environment to reduce land requirements and support water-based heat rejection.

This model has clear advantages in protected industrial waterways:

• Direct technician access
• Familiar data-center operations
• Shipyard or port fabrication
• Reduced terrestrial land use
• Access to local cloud customers
• Integration with regional power and networks

Its central expansion question is whether a protected-waterway model can be replicated across jurisdictions and expanded into higher-density AI infrastructure.

Floating systems remain exposed to surface weather, motion, humidity, corrosion, navigation constraints, community visibility, and mooring requirements. These disadvantages become more significant as projects move away from protected rivers and harbors.

Nautilus Data Technologies: Operating Floating Precedent

Nautilus provides one of the clearest Western examples of commercial water-cooled computing infrastructure.

Its Stockton facility demonstrates that floating or water-adjacent data-center systems can operate commercially. The company has also accumulated significant experience with direct liquid cooling and high-density facility design.

Nautilus is increasingly positioned around facility-scale cooling technology, prefabricated data halls, and liquid-cooling systems.

This makes it an important operating precedent, though not necessarily a direct architectural comparison with every subsea startup.

The primary question is whether Nautilus's future centers on expanding a fleet of floating data centers or licensing and deploying its cooling systems across conventional facilities.

Aikido Technologies: Compute Integrated with Floating Wind

Aikido is developing floating wind platforms that incorporate data halls, batteries, grid connections, and offshore computing.

Its model attempts to solve generation and compute together.

The potential advantage is direct access to offshore wind resources and the reuse of floating-wind project sites, industrial capabilities, and supply chains.

The primary challenge is integration.

The system combines:

• A large floating turbine
• Variable wind generation
• Battery storage
• Grid support
• Data-center cooling
• Marine motion
• Offshore personnel access
• Fiber connectivity
• Vessel-based maintenance

Each subsystem can affect the availability and economics of the others.

Atomarine: Floating Compute with Dedicated Power Vessels

Atomarine proposes standardized floating compute platforms supplied by separate power vessels.

Its stated initial pathway uses gas turbines, with marine nuclear reactors proposed as a future generation source. This makes Atomarine a floating generation-integrated compute company rather than a subsea platform.

The architecture could avoid some terrestrial interconnection and zoning constraints. It also begins at a very large proposed scale.

Important questions include:

• How will early compute and power vessels be financed?
• Where will high-capacity fiber be installed?
• How will gas be supplied offshore?
• How will marine maintenance and weather affect availability?
• What is the timeline for commercially available marine reactors?
• How will nuclear licensing and insurance be handled?
• Can the system begin at a smaller scale before reaching a full floating campus?

Panthalassa: Autonomous Wave-Powered Compute

Panthalassa is developing autonomous floating platforms that convert ocean-wave energy into electricity and use that power for onboard AI inference.

Its thesis is energy first. Compute moves to the offshore energy resource rather than requiring the energy to be transmitted to land.

This could create a valuable platform for workloads suited to remote inference and autonomous operation.

Networking remains a central constraint. Satellite connections may be sufficient for selected inference outputs, telemetry, and control. They are less naturally suited to moving large datasets, model checkpoints, video streams, storage traffic, and other high-bandwidth workloads associated with general-purpose AI infrastructure. The latency tax framing remains relevant here.

The platform also introduces wave-energy conversion, autonomous operation, marine recovery, and remote-maintenance requirements.

Mocean Energy: Offshore Power Moving Toward Compute

Mocean Energy has developed offshore wave-power systems and is extending that experience into a floating compute concept combining wave energy, solar power, batteries, satellite communications, and server equipment.

The company brings relevant offshore power and controls experience.

Its central challenge is scale. Existing offshore power systems measured in kilowatts are materially different from persistent AI infrastructure measured in megawatts.

The planned demonstrations will help establish which computing workloads can be supported economically by the platform.

China: Commercial Evidence at a Different Institutional Scale

Chinese projects provide the strongest public evidence that underwater data centers can move beyond experiments.

The Hainan cluster has expanded from an initial commercial unit into a broader intelligent-computing platform. Shanghai has also announced underwater infrastructure integrated with offshore wind.

These projects provide technical and strategic validation for the category.

They do not automatically establish the commercial model for Western companies.

Western deployments must operate within different systems involving:

• Private project finance
• Environmental review
• Insurance
• Customer procurement
• Hardware vendor requirements
• Local permitting
• Public transparency
• Service and retrieval obligations
• Cybersecurity and data governance

China demonstrates that underwater computing can operate at meaningful scale. It does not determine which architecture will succeed in Western markets.

Project Natick: Foundational Research Rather Than a Company

Microsoft's Project Natick remains the best-known Western underwater-compute program.

It demonstrated that servers could operate reliably inside a sealed subsea vessel and that the controlled internal environment could reduce certain hardware failure modes. (Microsoft)

Natick should not be categorized as an investable ocean-compute company.

Its role is historical and technical.

The program helped answer whether computing equipment could operate underwater. Current companies are addressing the next set of questions:

• How should capacity be serviced?
• How should hardware be refreshed?
• How should projects expand?
• How should customers reserve capacity?
• How should environmental performance be monitored?
• How should the infrastructure be financed?

Demand-First and Energy-First Ocean Compute

One of the clearest ways to understand the category is to separate companies based on what determines their location.

Demand-First Systems

Demand-first companies position compute near users, enterprises, fiber, data sources, and regional applications.

Potential advantages include:

• Lower network distance
• Regional data control
• Access to metropolitan demand
• Terrestrial fiber connectivity
• Reserved customer capacity
• Reduced data movement
• Better support for persistent applications

Their primary challenge is securing suitable coastal sites, power, permits, marine access, and environmental approval.

Seabase is principally a demand-first system.

Energy-First Systems

Energy-first companies position compute near offshore generation or energy resources.

Potential advantages include:

• Access to stranded energy
• Reduced dependence on terrestrial grids
• Avoidance of land and interconnection constraints
• Integration with wind, wave, gas, or nuclear systems

Their main challenge is transferring data economically and reliably between remote compute and customers.

Aikido, Atomarine, Panthalassa, and Mocean are primarily energy-first systems.

Neither category is universally superior. They serve different workloads and solve different infrastructure constraints.

Energy-first ocean-compute companies move compute to a proprietary power system. Seabase brings a power-flexible compute platform to the coastal regions where energy, terrestrial fiber, and customer demand already converge.

Power-Source Flexibility Lets Seabase Focus on Compute

Several ocean-compute companies combine the data center with a specific new source of offshore power.

Aikido integrates computing with floating wind turbines. Panthalassa and Mocean Energy are developing wave-powered systems. Atomarine proposes dedicated gas-generation vessels followed eventually by marine nuclear reactors.

These approaches may create value where the power resource itself is the central opportunity. They also require the company to solve two major infrastructure problems at the same time: producing reliable offshore electricity and operating customer-grade AI compute.

Seabase takes a different approach.

The Seabase platform is designed to integrate with the most appropriate qualified power source available in each region. Depending on the jurisdiction, that could include:

• Existing utility power
• Hydroelectric generation
• Wind
• Solar
• Natural gas
• Industrial or behind-the-meter generation
• Wave or tidal power where commercially qualified
• Conventional nuclear power
• Small modular reactors as they become licensed and commercially available
• Battery energy storage supporting any of these systems

This flexibility allows Seabase to work with utilities, power producers, ports, industrial facilities, renewable developers, and nuclear operators rather than becoming the primary developer of every generation asset.

Seabase can therefore concentrate its internal engineering and commercial effort on its core business:

• High-density AI infrastructure
• Modular subsea deployment
• Thermal management
• Compute hardware integration
• Module retrieval and servicing
• Regional capacity expansion
• Customer reservations
• Environmental monitoring
• Nori placement and infrastructure management

The distinction reduces coupled technology risk.

A Seabase deployment does not need to wait for a new wave-energy converter, floating wind platform, marine reactor, or offshore fuel system to become commercially mature. It can begin wherever qualified power, fiber, marine access, and customer demand already align. For how that integration works across regional energy systems, see power-flexible AI infrastructure.

Power Flexibility Preserves Geographic Flexibility

A platform tied to one generation technology must generally follow that energy resource.

Wave-powered compute must operate where wave conditions are suitable. Floating wind-integrated compute must follow offshore wind sites. Gas-powered vessels require an offshore fuel pathway. Nuclear-powered platforms depend on reactor availability, licensing, specialized operators, security, and host-jurisdiction approval.

Those locations may be far from the users, enterprises, data sources, and terrestrial networks consuming the compute.

Seabase reverses the relationship.

Rather than locating compute wherever a proprietary energy system can operate, Seabase identifies coastal markets where demand, power, fiber, ports, and marine conditions intersect. It can then integrate with the most appropriate regional energy system.

That allows Seabase to remain close to terrestrial fiber infrastructure, including:

• Submarine cable landing ecosystems
• Internet exchanges
• Carrier networks
• Cloud on-ramps
• Metropolitan fiber
• Enterprise networks
• Regional data sources

Terrestrial fiber can provide substantially greater, more predictable, and more economical bandwidth than systems built primarily around satellite connectivity. It also allows Seabase to support broader workloads, including video, storage, model distribution, retrieval systems, persistent agents, and other applications that move much more data than a simple text response.

Battery Storage as the Repeatable Layer

Primary generation may vary by region, but battery energy storage can provide a repeatable supporting layer across Seabase deployments.

Depending on the site design, BESS may support:

• Short-duration ride-through
• Power-quality management
• Renewable smoothing
• Switching between power sources
• Controlled shutdown
• Grid support
• Generator optimization
• Microgrid stability

Batteries do not replace the need for sufficient firm generation. They help make different regional power systems more stable and compatible with high-density compute.

A Focused Infrastructure Company

Seabase is open to nuclear, including future SMR integrations, but it does not need to become a reactor company.

It is open to wind and wave power, but it does not need to develop turbines or wave-energy converters.

It is open to natural gas and industrial generation, but it does not need to operate a fleet of fuel vessels.

This lets Seabase remain focused on building a modular AI infrastructure platform that can use whichever qualified power system best fits the host jurisdiction.

That focus is a meaningful difference within the ocean-compute landscape. Many competing approaches are simultaneously betting on a new compute architecture, a new marine platform, and a specific generation technology. Seabase is building the compute and subsea infrastructure layer while preserving access to multiple power pathways and high-capacity terrestrial networks.

How the Main Architectures Compare

Rather than ranking companies, the useful comparison is architectural.

• Floating river and port systems preserve continuous human access and can work well in protected industrial waterways.
• Submerged capsules simplify deployment of a sealed payload but can tie service and hardware refresh to recovery of a complete unit.
• Generation-integrated platforms can access offshore energy but combine compute with difficult power and marine-engineering challenges.
• Autonomous offshore nodes can operate far from land but face networking and maintenance constraints.
• Modular subsea platforms aim to preserve underwater placement while creating a serviceable and expandable infrastructure layer near demand.

No single architecture has become the industry standard. The relevant question is which model matches which geography, workload, service requirement, and financing path.

Questions Shaping the Category

The questions below will determine which architectures become commercial.

What Is Actually Operating?

The category must distinguish between:

• Concept
• Component test
• Marine prototype
• Powered compute demonstration
• Customer trial
• Production workload
• Commercial infrastructure
• Repeatable deployment

How Is Hardware Serviced?

Systems may depend on:

• Continuous human access
• Complete vessel recovery
• Individual module retrieval
• ROV intervention
• Planned no-service periods
• Autonomous maintenance

The service model affects downtime, cost, refresh cycles, and customer confidence.

Can the System Support Multiple Hardware Generations?

AI hardware changes much faster than most marine assets.

Infrastructure that is permanently tied to one accelerator or rack design may become economically obsolete before the marine structure reaches the end of its useful life.

What Is the Complete Network Path?

Investors, customers, and regulators should evaluate:

• Fiber availability
• Cable routes
• Backhaul
• Cross-connects
• Satellite links
• Bandwidth
• Tail latency
• Network redundancy
• Data-transfer cost

Cheap remote power does not automatically create inexpensive delivered AI service.

What Is the Unit of Failure?

A failure may affect:

• One server
• One rack
• One module
• One capsule
• One barge
• One turbine platform
• One offshore campus

Architecture determines how widely failures propagate.

How Are Environmental Effects Measured?

Credible projects should address:

• Baseline studies
• Thermal conditions
• Underwater acoustics
• Materials and corrosion
• Biofouling
• Seabed effects
• Retrieval
• Monitoring
• Data sharing
• End-of-life removal

Can the Model Move Beyond Venture Financing?

Ocean infrastructure will eventually require capital beyond conventional startup equity.

Possible pathways include:

• Reserved-capacity contracts
• Strategic investment
• Equipment financing
• Project finance
• Infrastructure funds
• Government support
• Regional project companies
• Customer prepayments

Where Seabase Fits

Seabase is a demand-first, power-source-flexible subsea compute platform using terrestrial fiber.

Seabase is developing modular subsea infrastructure near coastal demand. The platform is power-source flexible and connected to terrestrial fiber.

Its intended position within the landscape is defined by:

• Modular subsea infrastructure with independently retrievable compute modules
• Near coastal demand, ports, and marine operations
• Power-source flexibility: Seabase can integrate with grids, industrial generation, hydroelectricity, wind, solar, gas, nuclear, and emerging marine energy systems without depending on any one of them
• Connected to terrestrial fiber: nearshore placement allows Seabase to use established metropolitan and international fiber rather than relying primarily on bandwidth-constrained remote communications
• Staged capacity expansion and hardware refresh across generations
• Module-level failure isolation
• Reserved capacity with Nori-based placement and monitoring
• Site-specific environmental accountability

This does not remove the company's execution risk.

It defines the particular risks Seabase has chosen to solve.

The company must demonstrate that marine infrastructure, AI racks, thermal systems, power, fiber, retrieval, environmental monitoring, and customer operations can function together as a repeatable platform.

The architecture avoids several dependencies present elsewhere in the market. It does not require a new reactor, wave-energy converter, floating wind system, satellite network, or permanently occupied marine facility.

Seabase's central thesis is that ocean compute becomes most valuable when modular subsea infrastructure is placed near coastal demand, remains power-source flexible, and stays connected to terrestrial fiber.

A Category Still Being Defined

Ocean compute is becoming a legitimate infrastructure category, but no single architecture has yet become the standard.

Floating systems preserve human access but remain exposed surface assets.

Submerged capsules simplify deployment but can tie service and hardware refresh to a complete sealed unit.

Generation-integrated platforms can access offshore energy but combine compute with difficult power and marine-engineering challenges.

Autonomous nodes can operate far offshore but face networking and maintenance constraints.

Modular subsea platforms aim to preserve the benefits of underwater placement while creating a serviceable and expandable infrastructure layer near coastal demand, with power-source flexibility and terrestrial fiber adjacency.

The next phase will not be decided only by whether servers can operate near or beneath the ocean. That question has largely been answered.

It will be decided by which systems can be permitted, financed, monitored, serviced, refreshed, connected to customers, and repeated across regions.

That is the environment in which Seabase is building: modular subsea infrastructure, near coastal demand, power-source flexible, and connected to terrestrial fiber.

For Seabase's architecture comparison with floating systems, see Floating vs. Subsea Data Centers. For modular subsea infrastructure and hardware refresh, see Beyond Project Natick. For power-source flexible regional energy integration, see power-flexible AI infrastructure. To discuss capacity near coastal demand on terrestrial fiber, contact Seabase.