Wave-Powered Data Centers: Can Panthalassa’s Untethered Ocean Buoys Deliver 2¢/kWh AI Compute?

The Machine: A Floating Dam That Never Touches Shore

Picture a steel sphere sitting at the surface with a long tube hanging straight down into deep water. That’s it. Panthalassa calls the current design Ocean-3, and CEO Garth Sheldon-Coulson has described it publicly as a hydroelectric dam that floats.

Each passing wave lifts the entire structure and lets it fall. That motion drives water up through the tube into the sphere, where it passes through a turbine, spins a generator, and cycles back through to do it again. No fuel. No engine. No direct emissions.

There’s also nothing holding it in place. The node carries no mooring lines, never plugs into a power cable, and gets towed out of port lying flat before flipping itself upright in open water. It holds position using the hydrodynamic shape of its own hull. Sheldon-Coulson’s description to CBS Sunday Morning:

“It’s like a little Roomba, except it’s enormous.”

The electricity never leaves the buoy. Inference chips ride inside the node, cold seawater handles the cooling, and wave power gets spent on the spot answering AI queries. A question travels up to a Starlink satellite, the chips process it, and the answer rides the same link back down.

That one design decision quietly changes what the product actually is. Panthalassa isn’t selling electricity. It’s selling compute, with electricity as a captive ingredient.

Why Wave Energy Has Always Failed — Until Now

Wave power has been five years away since the 1970s. The physics were never the problem. The open ocean carries enormous, continuous energy that delivers day and night in a way solar can only envy.

What killed every project was the infrastructure around the generator. Mooring systems fight the sea constantly. Subsea transmission cables cost a fortune to lay, a fortune to maintain, and a fortune to repair when the ocean wins a round. You’d capture some of the cheapest raw energy on the planet and watch the budget drown in the hardware needed to move electrons home.

Panthalassa’s answer is to delete the cable entirely rather than pay for it. That single move sidesteps the problem that has bankrupted wave-energy ventures for decades.

Whether it creates new problems is a fair question. We’ll get to those.

The Money Behind the Buoy

The Series B round, announced May 4, was led by Thiel and earmarked to finish a pilot manufacturing facility near Portland. The investor list reads like a Silicon Valley reunion: John Doerr, Marc Benioff’s TIME Ventures, Max Levchin’s SciFi Ventures, and Super Micro all participated.

Thiel’s own statement ended with six words you don’t often hear in marine engineering:

“Panthalassa has opened the ocean frontier.”

Here’s the funding picture at a glance:

• Founded: 2016, as a public benefit corporation
• Ocean-1 prototype: In the water 2021
• Ocean-2: Deployed 2024
• Pre-Series B total raised: ~$78 million
• Series B (May 2026): $140 million, led by Peter Thiel
• Reported valuation: ~$1 billion
• Team size: ~120 people, drawn from SpaceX, Tesla, Blue Origin, Boeing, and NASA
• Pilot Ocean-3 deployment: North Pacific, targeting August 2026
• Commercial target: 2027
• Claimed generation cost at scale: $0.02/kWh

The company is older than the hype suggests. Nearly a decade of quiet hardware testing preceded this moment, which is either reassuring or a sign of how hard the problem actually is — probably both.

There’s also a broader pattern forming around capital like this. Two months before Panthalassa’s round, Starcloud — a startup planning orbital data centers — raised $170 million and crossed a $1.1 billion valuation. Orbit, seabed, open water: if the grid can’t feed the machines, investors will fund almost any address that can.

The 2-Cent Claim: What It Would Mean If True

The number doing the most work in this story is $0.02 per kWh. Panthalassa asserts that at scale, its platform could generate power at that cost. That figure would undercut gas, undercut solar in most markets, and undercut basically everything else on the grid.

It would also fundamentally change the economics of AI inference. Power is one of the largest operating costs in any data center. Cut it by 80% and the competitive math across the entire cloud compute market shifts.

But this is a projection from a company with zero commercial nodes in the water. Wave energy has a long history of cost models that looked elegant right up until the ocean started editing them.

The Skeptic’s Case Is Real

The asterisks here are worth taking seriously.

Satellite bandwidth caps the market. Starlink works well for inference — small questions, small answers, modest latency tolerance. It’s a poor fit for training frontier models, which require racks of GPUs wired together with fat, low-latency connections. That limits Panthalassa’s addressable market to a slice of AI compute, not the whole stack.

Stationkeeping is unproven at fleet scale. A free-floating machine with no mooring has to hold its patch of ocean through North Pacific winter storms using hull shape and software alone. Nobody has demonstrated that at commercial scale.

Maintenance is expensive and slow. A generator that breaks down sits a very long boat ride from port. Saltwater corrosion and biofouling are relentless. The cost models for offshore hardware almost always underestimate what the sea charges for access.

Panthalassa’s response to all of it is consistent: build the thing and put it in the water. Ocean-3 construction is already underway. Sheldon-Coulson expects units operating offshore around August 2026, with commercial deployments following in 2027.

That timeline is aggressive. It’s also the only way to find out if the physics and the economics actually converge.

What This Means for the AI Tools Ecosystem

For founders and operators building on top of AI infrastructure, the relevant question isn’t whether Panthalassa succeeds. It’s what the market looks like if even a fraction of this approach works.

Ultra-cheap, off-grid inference compute changes the pricing floor for AI APIs. It opens geographic markets where grid infrastructure is weak or expensive. It creates a new category of edge compute that doesn’t depend on hyperscaler buildout timelines or utility approval cycles.

The grid bottleneck has been a structural advantage for incumbents — AWS, Azure, Google Cloud — because they already have the power deals locked in. Any credible alternative supply of cheap compute erodes that moat, even partially.

Watch the August deployment closely. If Ocean-3 holds position through its first North Pacific winter and the cost numbers start to look defensible, the conversation around AI infrastructure pricing changes faster than most people expect.

The Ocean Is Having Its Moment

Japan is pulling round-the-clock electricity from the salinity gradient between fresh and salt water in Fukuoka. China parked its servers on the seabed. And this summer, if the schedule holds, a giant steel lollipop will start answering AI queries somewhere in the North Pacific with nothing beneath it but very deep water.

If the 2-cent claim survives contact with the real ocean, every grid fight in the AI buildout starts to look optional. If it doesn’t, the Pacific gets a very expensive bath toy.

Either way, the answer arrives fast — and it arrives by satellite.