Somewhere off the northern tip of Scotland, full-scale wave energy converters bob in the North Atlantic, feeding electricity into the local grid from one of the world’s most punishing stretches of open ocean. The machines at the European Marine Energy Centre’s Billia Croo test site were designed to prove that waves can generate usable power. Now a growing number of technologists and energy startups want to use that same power source for something far more ambitious: running the data centers behind artificial intelligence.
The pitch goes like this. Instead of fighting for space on an already strained electrical grid, put your servers on the water. Let ocean waves generate the electricity. Let seawater handle the cooling. Skip the land acquisition battles, the transmission bottlenecks, and the years-long interconnection queues that have turned new data center construction into a logistical nightmare across much of the United States.
It is a compelling vision. It also remains, as of spring 2026, almost entirely unproven at the scale AI workloads demand.
Federal dollars target marine energy for offshore work
The strongest institutional endorsement of the concept so far comes from the U.S. Department of Energy, which launched a $1.7 million prize competition called Power at Sea. The program challenges engineers to design systems that harvest wave, tidal, or ocean current energy to run equipment far from shore, where the usual alternatives are expensive diesel generators or long undersea cables.
The DOE competition does not single out data centers by name. But the agency’s broader marine energy research portfolio, managed partly through the National Renewable Energy Laboratory, explicitly studies how ocean power can serve remote, off-grid loads, a description that fits offshore computing infrastructure neatly. By framing marine energy as a serious candidate for powering economic activity at sea, the DOE has given the floating data center concept something it badly needs: legitimacy with investors and regulators who might otherwise dismiss it as science fiction.
That legitimacy matters because AI-driven electricity demand is surging. The International Energy Agency projected in early 2025 that global data center power consumption could more than double by 2030, with AI training and inference driving much of the increase. In the U.S., utilities and grid operators have flagged data center load growth as a major planning challenge, particularly in Northern Virginia, central Texas, and the Pacific Northwest. Siting servers offshore and pairing them with co-located wave energy converters would, in theory, sidestep that grid competition entirely.
Prize funding of $1.7 million is modest compared with what it costs to build even a small data center. But federal competitions serve a different purpose: they de-risk early-stage technologies, attract private co-investment, and signal to permitting agencies that the government considers the underlying science credible. For floating data center advocates, that signal is essential. No server farm operates without dependable upstream power, and no new power technology scales without credible customers willing to bet on it.
Scotland’s wave test site offers real hardware data
Any offshore computing concept that depends on wave energy needs hardware that works reliably in open ocean, not just in a wave tank. The most advanced proving ground for that hardware is Billia Croo, operated by the European Marine Energy Centre off Orkney, Scotland. The site is documented by the Pacific Northwest National Laboratory through its Tethys research platform, which catalogs marine energy projects under the U.S. Department of Energy. Billia Croo allows developers to deploy full-scale wave energy converters in real Atlantic swells and feed electricity into the grid, generating performance data that no simulation can replicate.
Critically, Billia Croo operates under a formal consenting and permitting framework. Device developers must clear environmental impact reviews, cable routing approvals, and grid interconnection standards before they plug in. That regulatory track record is directly relevant to anyone proposing a wave-powered floating data center, because such a project would face the same categories of questions: effects on marine life, navigational safety, electromagnetic interference from subsea cables, and thermal discharge from cooling systems.
What Billia Croo has not yet demonstrated is the kind of sustained, uninterrupted output that data centers require. Training a large language model or running real-time AI inference demands steady, high-wattage electricity around the clock, with tight tolerances for voltage and frequency. Wave energy output fluctuates with tide cycles, storm intensity, and seasonal conditions. A grid-connected test site can smooth those swings by drawing backup power from shore. A standalone floating data center would have to solve that variability problem on its own, likely through onboard battery storage or a hybrid system combining wave power with wind, solar, or another source. No publicly documented project has demonstrated such a hybrid at data center scale.
The companies circling the concept
The floating data center idea did not emerge from nowhere. Microsoft’s Project Natick, which ran a sealed underwater data center pod on the seafloor off Orkney from 2018 to 2020, proved that servers could operate reliably in a marine environment with dramatically lower failure rates than comparable land-based facilities. That experiment used grid power delivered by cable, not wave energy, but it established that the ocean’s natural cooling properties could reduce one of the biggest cost drivers in data center operations.
Other companies have pushed the concept closer to the surface. Nautilus Data Technologies has operated a barge-mounted, water-cooled data center in Stockton, California, using river water for cooling. Subsea Cloud, a Texas-based startup, has proposed seafloor-mounted computing pods for military and commercial clients. Neither company has publicly committed to wave energy as a primary power source, but both have demonstrated that moving compute off dry land is technically feasible.
What none of these efforts have done is close the loop between ocean-based computing and ocean-generated power. The integration step, pairing a working wave energy converter with a working offshore server installation and running real workloads on the combined system, remains the missing piece. Until someone builds and operates that pairing, the wave-powered floating data center exists as a logical extension of proven components rather than a proven system.
The gap between pitch and deployment
Several hard problems stand between the current state of the technology and a functioning commercial facility.
Power density. Today’s most advanced wave energy converters produce power in the hundreds of kilowatts. A single AI training cluster can draw tens of megawatts. Bridging that gap requires either dramatic improvements in converter output or arrays of dozens or hundreds of devices working in concert, raising questions about cost, maintenance, and ocean space allocation that no developer has publicly answered.
Hardware durability. Marine environments are brutal on equipment. Saltwater corrosion, biofouling from barnacles and algae, and storm damage can drive maintenance costs far above onshore equivalents. Server hardware, which is sensitive to vibration, humidity, and temperature swings, adds another layer of vulnerability. Proponents argue that sealed, pressure-controlled pods can protect electronics, and Project Natick’s results support that claim. But wave energy converters, which must interact directly with the ocean surface, face exposure that sealed pods do not.
Regulatory uncertainty. The United States has no established permitting framework tailored to permanent floating data centers in federal waters. Offshore energy projects already face review processes stretching years, involving NOAA, the Bureau of Ocean Energy Management, the Army Corps of Engineers, and state coastal agencies. Adding a data center, with its own cybersecurity, data sovereignty, and reliability considerations, would likely trigger additional scrutiny. NOAA’s ongoing monitoring of ocean acidification and marine chemistry changes also raises questions about how thermal discharge and electromagnetic fields from offshore computing infrastructure might interact with already stressed ocean ecosystems. No published environmental impact assessment has addressed these factors for a wave-powered data center scenario.
Cost transparency. Without published cost models from an operator or research institution, the economic case for wave-powered floating data centers rests on projection rather than evidence. Proponents highlight savings from natural cooling, avoided land costs, and reduced transmission losses. Skeptics counter that marine operations carry insurance, logistics, and maintenance premiums that could erase those savings. Until a pilot project publishes real numbers, investors are working from assumptions.
Where the current leaves the concept
The verified record, as of May 2026, supports two concrete facts. The DOE has committed real funding to powering offshore economic activity with marine energy. And the EMEC Billia Croo test site is an operational, grid-connected wave energy facility whose data is cataloged by a U.S. government-affiliated research lab. Both confirm that institutional money and testing infrastructure exist to support the broader idea of ocean-powered offshore work.
The connection to AI computing specifically is logical but indirect. Cloud providers and AI companies are under genuine pressure to decarbonize and expand capacity quickly, and the grid constraints they face onshore are real and worsening. Moving mission-critical compute to floating platforms powered by a still-maturing energy source represents a significant operational gamble, but it is the kind of gamble that looks more rational every time a utility tells a hyperscaler that grid capacity will not be available for another five years.
What the concept needs next is not another prize competition or another test-site deployment of a wave energy converter in isolation. It needs someone to build the integrated system: wave power feeding servers on the water, running real workloads, publishing real performance and cost data. Until that happens, ocean-powered floating data centers will remain one of the more plausible answers to AI’s energy problem that nobody has actually tried.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
