Panthalassa wants to put AI computing where the electricity is: out at sea, aboard floating systems powered by waves. Somewhere, a server technician has just added “seasickness” to the list of things to consider before accepting a job.
The Portland company announced a $140 million Series B on May 4, led by Peter Thiel and joined by investors including John Doerr and Marc Benioff’s TIME Ventures. That is the size of the round, not a statement that Thiel personally supplied every dollar.
Use the Power Where It Is Generated
According to Panthalassa, its autonomous nodes use wave energy to run AI chips onboard, send inference output to land through satellites and use the surrounding ocean for cooling. The funding is intended to finish a pilot manufacturing facility near Portland and support the next deployments.
The company reports at-sea work with its Ocean-1, Ocean-2 and Wavehopper prototypes in 2021 and 2024. Its stated plan is to deploy the Ocean-3 pilot series in the northern Pacific during 2026, ahead of commercial deployments targeted for 2027. Those are plans, not completed commercial results.
The interesting design choice is using energy locally instead of first transporting it to an onshore computer. Inference means running an already-trained model to produce an answer. That workload is different from building a tightly connected cluster to train a large model from scratch.
Why Investors Are Looking Beyond the Grid
The pressure behind experiments like this is real. In its 2025 Energy and AI report, the International Energy Agency estimated that data centres consumed about 415 terawatt-hours in 2024 and projected around 945 TWh by 2030 in its base case. That covers data centres generally, not AI alone, and the later figure is a projection.
Available generating capacity and the ability to connect a particular site are different things. A computer cannot run on the reassuring knowledge that someone is building a power plant somewhere. Locating computing beside an energy source is therefore an understandable idea to investigate.
It does not settle the cost comparison. You would still want to know the useful computing delivered per dollar after manufacturing, communications, maintenance and replacement. The nearest wave does not send an electricity bill, but the machinery that turns it into useful work certainly has a budget.
Microsoft Tried Underwater Computing, With a Different Setup
There is a useful precedent in Microsoft’s Project Natick. Its Northern Isles data centre was placed on the seabed off Orkney in 2018 and recovered in 2020. Microsoft reported encouraging reliability results from that experiment.
Natick was connected by cable to the Orkney electricity grid. It was not a free-floating machine generating its own wave power. The distinction matters: success at keeping servers in a sealed underwater environment does not automatically validate a different system’s power generation, positioning or communications.
The Questions a Pilot Needs to Answer
Here is what would make the next stage informative, beyond a convincing photograph of hardware in water:
- Useful availability. How much computing can customers actually obtain through changing sea conditions and equipment outages? Peak generating capacity is only part of that answer.
- Repair and recovery. What happens when a component needs attention, and what does retrieving or servicing it cost? A design has to account for maintenance as well as successful operation.
- Communications. Which workloads fit the available satellite bandwidth and response times? Satellite links are neither a universal solution nor an automatic disqualification; the application’s requirements matter.
- Environmental effects. What do measurements show about heat discharge, noise, marine life and the footprint of a larger fleet? Access to cooling water is not itself an environmental-impact assessment.
- Operating arrangements. Which permissions, navigation safeguards, responsibilities and end-of-life plans apply? Distance from shore is not a substitute for answering those questions.
These are questions for evidence, not a prediction that the project will fail. Nor does a distinguished investor list establish that every engineering question has already been resolved.
Our look at AI’s pressure on memory supply explores another part of the physical infrastructure behind apparently weightless software. Chips, power, cooling and connections still have to exist somewhere.
Panthalassa’s somewhere is unusually wet. If the pilots demonstrate dependable computing at a competitive cost, that would be a meaningful result. For now, the proposal is intriguing enough without declaring the Pacific a finished data centre. Cats would understand the attraction of warm, humming equipment. The commute might be a harder sell.
