What if AI data centres didn’t need land?

Neil L. Rideout

9/15/20265 min read

What if AI data centres didn’t need land?

We usually imagine AI data centres as massive buildings packed with servers, power systems and cooling infrastructure sprawling across industrial parks or desert expanses. But a new idea is challenging that assumption. What if computing infrastructure could float on water — or operate beneath it?

AI is pushing data-centre power density higher, with modern GPU-heavy racks reaching 30–50+ kW per rack. And all that computing produces heat. This is where water-based data centres become interesting. Floating data centres place computing infrastructure on vessels or platforms. Submerged data centres seal computing modules to operate underwater. The potential is intriguing: more efficient heat rejection, potentially lower freshwater use for cooling, modular deployment, less dependence on land, and new possibilities for locating computing closer to coastal demand.

But these benefits come with a completely different set of engineering challenges — from corrosion and maintenance to power, connectivity and environmental considerations. So, is this the future of data-centre infrastructure? Maybe. Maybe not. But it’s definitely a technology worth understanding. I’m exploring it in a three-part series: the concept (what are floating and submerged data centres?), the engineering (how do they actually work?), and the reality (benefits, challenges and scalability).

The Concept — What Are Floating and Submerged Data Centres?

Traditional land-based facilities face mounting constraints. Power grids in many regions struggle to deliver the multi-hundred-megawatt loads AI training and inference demand. Land acquisition triggers community opposition, permitting delays stretch for years, and freshwater for evaporative cooling is increasingly scarce. Water-based approaches sidestep several of these bottlenecks by treating the ocean itself as both a cooling medium and, in some designs, a power source.

Floating data centres take several forms. Some are converted barges or purpose-built platforms moored near shore, drawing seawater for cooling loops while connecting to onshore power and fibre. Others are fully autonomous nodes that generate their own electricity from waves or wind and communicate via satellite. Submerged designs go further: sealed pressure vessels or pods rest on the seabed or hover in the mid-water column, using the surrounding seawater for passive or semi-passive cooling.

Microsoft’s Project Natick offered the most public proof of concept for the submerged approach. In 2018 the company deployed a shipping-container-sized vessel containing 864 servers off Scotland’s Orkney Islands. Retrieved after more than two years, the underwater cluster showed a failure rate roughly one-eighth that of a comparable land-based control group. The stable, cool temperature and nitrogen-filled atmosphere were credited for the reliability gains. Although Microsoft later paused commercial development of the concept, the data validated the core physics.

China has moved further toward operational reality. Off Shanghai’s Lingang coast, an underwater data centre powered by a nearby offshore wind farm reached commercial operation with roughly 24 MW capacity. Developers report a power usage effectiveness (PUE) around 1.15, zero freshwater consumption for cooling, and land-use reductions exceeding 90 percent. Similar commercial efforts by companies such as Highlander and Subsea Cloud continue to refine pressure-tolerant, immersion-cooled pods.

Floating concepts have proliferated more recently. Nautilus Data Technologies operated a barge-based facility in Stockton, California, using river water for cooling. Shipbuilders including Seatrium, Hanwha Ocean and Samsung Heavy Industries have released modular floating data-centre designs in the 30–200 MW range that combine onboard generation, seawater cooling and relocatable modules. Start-ups such as Panthalassa are pursuing free-floating, wave-powered nodes that operate far offshore without anchors or grid connections, transmitting AI inference results via satellite.

The common thread is modularity and proximity to the cooling resource. Whether floating or submerged, these systems treat the ocean as infrastructure rather than merely a location.

The Engineering — How Do They Actually Work?

Cooling is the most obvious advantage. Seawater provides a vast, relatively stable heat sink. In submerged designs, heat exchangers transfer server heat directly into the surrounding water; closed-loop systems prevent contamination while maintaining low operating temperatures. Floating platforms often use open or closed seawater loops that eliminate the need for energy-intensive chillers and cooling towers. Reported PUEs frequently fall below 1.2, sometimes approaching 1.05 under ideal conditions — a meaningful improvement over many land facilities that still operate above 1.4.

Power is the harder problem. Near-shore floating centres can simply connect to the terrestrial grid or to offshore wind. Fully autonomous systems must generate their own electricity. Wave energy is particularly attractive for floating platforms. Devices such as Panthalassa’s nodes use the vertical motion of waves to drive water through turbines, producing electricity that powers onboard GPUs. Hybrid designs add floating solar or integrate data halls into the ballast tanks of floating wind platforms, as proposed by companies like Aikido.

Tidal energy opens another pathway, especially for submerged or near-shore installations. Swedish company Minesto has developed a distinctive technology based on underwater “kites.” The Dragon 12 is a 12-metre-wingspan, roughly 28-tonne device rated at 1.2 MW. Anchored to the seabed by a tether, it “flies” in a figure-eight pattern through tidal currents, accelerating relative to the flow and driving a rear-mounted turbine. Because the kite moves faster than the ambient current, it can extract useful power from slower tidal streams that conventional fixed turbines cannot economically exploit. Power travels up the tether and through a seabed umbilical to shore.

Minesto has operated Dragon 12 grid-connected in the Faroe Islands’ Vestmannasund, validating performance and refining maintenance procedures. An upgraded version targeting 1.75 MW generator capacity, optimised wing profiles and longer tethers is planned for a first commercial “Dragon farm” of approximately 10 MW in Hestfjord. The technology’s relatively light weight per megawatt and ability to operate in moderate flows make it a candidate for co-location with coastal or submerged compute loads. In principle, a cluster of tidal kites could supply baseload renewable power to nearby floating or underwater data modules, creating tightly integrated ocean energy–compute systems.

Connectivity remains a challenge. Near-shore installations use conventional submarine fibre. Far-offshore autonomous nodes rely on low-Earth-orbit satellite constellations. Latency is higher than terrestrial fibre, so these systems are better suited to inference workloads or batch processing than to real-time interactive applications. Corrosion, biofouling, and the logistics of retrieving sealed modules for maintenance demand specialised materials, coatings and remote-operation strategies. Pressure vessels must withstand depth-related forces, while floating structures must survive storms and extreme sea states.

Environmental considerations include the local thermal plume from heat rejection, electromagnetic fields from cables, and the physical footprint of foundations or anchors. Careful site selection and monitoring can mitigate impacts, but regulators will scrutinise large-scale deployments.

The Reality — Benefits, Challenges and Scalability

The benefits are real. Water-based centres can dramatically reduce freshwater consumption, lower cooling energy overhead, and free land for other uses. Modular designs allow capacity to be added incrementally and, in some cases, relocated. Locating compute near coastal population centres or renewable generation sources can shorten data-transport distances and improve overall system efficiency. For AI workloads that tolerate higher latency, autonomous wave- or tidal-powered platforms offer a route to carbon-free capacity unconstrained by terrestrial grids.

Yet challenges are equally substantial. Capital costs for marine-grade construction, specialised vessels and retrieval systems are high. Maintenance windows are weather-dependent and expensive. Regulatory frameworks for ocean use, environmental impact assessment and maritime safety remain underdeveloped in many jurisdictions. Supply chains for pressure-tolerant electronics and long-life materials are still maturing. Scalability to the multi-gigawatt levels AI demand may eventually require has not been demonstrated; most operational or near-term projects remain in the tens of megawatts.

Hybrid approaches may prove most practical in the medium term. Floating platforms moored near offshore wind farms, or submerged pods co-located with tidal arrays such as Minesto’s Dragon systems, could combine reliable renewable generation with efficient cooling while remaining accessible for maintenance. Ship-based or barge concepts that can be towed into port for upgrades offer another pragmatic path.

Whether water-based data centres become a mainstream solution or remain a specialised niche will depend on continued cost reduction, proven long-term reliability, and the relative severity of land and power constraints onshore. The physics of ocean cooling and the growing maturity of marine renewables — including tidal kites like the Dragon 12 and advanced wave-energy converters — make the concept technically credible. Economic and operational realities will decide its ultimate scale.

For now, the idea that AI infrastructure need not be tied to land is no longer purely speculative. It is an active engineering frontier, one that forces us to rethink where — and how — the computational substrate of the digital age can live. Understanding the trade-offs is the first step toward deciding whether the ocean will host a meaningful share of tomorrow’s data centres.

Contact

Head Office

Green Life Enterprises LLC
8 Biltmore Est
Suite 310,
Phoenix, AZ 85016

greenlifedatacenters@gmail.com

+1-813-220-0001

Canadian Office

Green Life Enterprises LLC
3142 Nicholson Ave
Suite 10
New Waterford, Nova Scotia B1H 1N8