Three months ago Panthalassa was valued at roughly $1 billion. Now investors are reportedly pricing it at $2 billion before a single commercial node has floated in the open ocean. The Information reported on August 6 that the company is raising $225 million at a post-money valuation approaching $2 billion, roughly doubling its mark from a $140 million round led by Peter Thiel in May.
The pitch: generate electricity at sea, cool the chips with seawater, and beam results back via satellite, skipping grid queues, cooling water, and land permitting entirely.
The counter: wave energy has failed commercially for 15 years, node costs remain company projections, and corrosion, storm survival, and satellite bandwidth are unproven at scale.
Panthalassa builds 85-meter steel nodes that generate about one megawatt each from wave motion. Water forced through an internal turbine powers GPUs inside a sealed, nitrogen-purged enclosure. The surrounding ocean cools the hardware. Results travel to shore through low-Earth-orbit satellite links, so the node never needs a cable back to land. The company tested the energy mechanics with Ocean-1 in 2021 and Ocean-2 in 2024, and plans to deploy its first Ocean-3 pilot series in the northern Pacific this year, with commercial systems targeted for 2027.
The timing matters. A single NVIDIA GB200 rack draws 120 to 132 kilowatts, next-generation Rubin racks are projected to pass 250 kilowatts, and the US grid interconnection queue stretches into 2030. Northern Virginia, the largest data-center market on the planet, paused new approvals while its grid catches up. That is the bottleneck Panthalassa is selling against: not a better data center, but a data center that does not need the grid at all.
The case for moving compute to sea
Start with the physics. Ocean water below 100 meters sits between 4 and 15 degrees Celsius regardless of latitude or season. That is a free, year-round cooling reservoir, and cooling can account for up to 40% of a data center's electricity bill. A sealed vessel submerged in that reservoir can run server inlet temperatures within the limits set by the American Society of Heating, Refrigerating and Air-Conditioning Engineers, without evaporating a single liter of freshwater. That solves two land constraints at once: power scarcity and water scarcity.
No grid interconnection queue — the node generates its own power and consumes it onboard, so there is no multi-year wait for transmission access.
No land or permitting fight — deployment zones are remote waters, outside the community-approval battles stalling terrestrial campuses.
The wave is the fuel — the company claims power costs near two cents per kilowatt-hour and capacity factors above 90%, though both figures remain company projections.
The architecture also dodges the single failure that killed earlier wave-energy companies: the subsea export cable. Getting electricity from the open ocean to shore requires undersea cables that, according to Bryson Robertson of the University of Victoria, can cost hundreds of millions of dollars for a few miles. Panthalassa never transmits power. It transmits tokens. The compute stays where the energy is, and the customer buys capacity rather than kilowatt-hours. That inverts the economics that defeated the wave-energy startups of the past decade, including AW-Energy, which filed for bankruptcy last year, and AquaHarmonics, which shut down in 2025.
Backers with infrastructure scale behind them are reading the same logic. Mike Schroepfer, former chief technology officer of Meta and founder of Gigascale Capital, a returning investor, put it plainly: "An additive energy source co-located with effectively unlimited cold seawater, that doesn't require waiting five to seven years for a grid hookup, is uniquely fit for this moment." That is not a retail-investor endorsement. It is a person who scaled gigawatts of Meta data centers saying the bottleneck is real and this is a plausible way around it.
We built millions of ships that survive the worst ocean conditions. So can you build the structure that can survive in the ocean wherever they want to deploy them? 100%.— Bryson Robertson, director, Institute for Integrated Energy Systems, University of Victoria
Why the ocean has beaten better ideas
Then consider the history. Wave energy has been "about to work" since the 1970s. Fifteen years ago a generation of well-funded startups promised mainstream wave power; none succeeded, and the technology remains a rounding error in global renewables. Panthalassa is asking investors to believe not that wave energy works, which is the question that bankrupted its predecessors, but that wave energy works while carrying delicate GPUs through hurricanes, salt spray, and biofouling, with nobody on board to fix a failed drive.
Every number that matters is a projection — the two-cent kilowatt-hour, the 90% capacity factor, the million-dollar node. None has been demonstrated at commercial scale at sea.
The ocean is a hostile environment — corrosion, biofouling, and storm loads on a sealed compute platform are qualitatively different from a ship, which gets regular dry-dock maintenance.
Satellite data links are a constraint, not a feature — the bandwidth and latency of low-Earth-orbit links will not support training-grade or latency-sensitive inference workloads at hyperscale.
Climate-tech analyst Michael Barnard has been the sharpest skeptic. His core argument, made in CleanTechnica, is that putting data centers on wave platforms does not make wave energy easier; it layers a second unproven technology on top of a first one. A ship can be towed to port for repair; an unmoored Ocean-3 node drifts in remote Pacific waters, and the company's own cost estimate of one to one-and-a-half million dollars per node explicitly excludes logistics and maintenance. Those are exactly the line items that grow without warning in a marine environment.
There is also the precedent the company cannot escape. Microsoft ran Project Natick, testing sealed underwater servers cooled by seawater, and reported a lower failure rate than comparable land systems. It still retired the project. The physics worked; the business case did not. Panthalassa's answer is that Natick fed a land grid while Panthalassa sells compute, but the burden of proof now sits with a company whose valuation has doubled to $2 billion on the back of a reported term sheet and no commercial deployment.
The numbers that decide
Set the narrative aside. The valuation question reduces to a handful of figures, and they are all on the table.
| Parameter | Panthalassa node | Land campus |
|---|---|---|
| Build cost per MW | ✔ $1–1.5M (est., excludes logistics) | ✗ $10.7M average (JLL, 2025) |
| Grid interconnection | ✔ None required — self-powered | ✗ Queue into 2030, multi-year wait |
| Cooling water | ✔ Seawater, zero freshwater draw | ◐ Cooling towers consume ~2M liters/MW/year |
| Deployment time | ◐ Weeks-to-months per node (claimed) | ✔ 3–5 years for permitted campus |
| Proven at commercial scale | ✗ No node deployed commercially | ✔ Thousands of live facilities |
New round, doubled mark
Reported $225 million raise at a post-money valuation near $2 billion, up from roughly $1 billion after the $140 million Series B in May. · The Information, 2026
Onboard power output
Each 85-meter node aims to generate roughly one megawatt continuously from wave motion, consumed onboard by AI inference chips. · Data Center Dynamics, 2026
Cost contrast
Average global data center construction cost rose from $7.7M to $10.7M per megawatt between 2020 and 2025, versus Panthalassa's claimed $1–1.5M per node. · JLL via Latitude Media, 2026
The $10.7 million figure is the strongest part of the story and the weakest in the same breath. Land campuses are expensive, slow, and water-hungry. The comparison is real. But Panthalassa's own estimate excludes the exact costs that tend to explode at sea, and a per-megawatt comparison against an average hides the fact that hyperscale campuses now routinely exceed that average. A fair test is not whether a node beats the median land campus on paper. It is whether a fleet of nodes, with logistics and maintenance included, can run at a cost that a hyperscaler will pay for capacity it can count on.
The honest read is that this is a bet with a new structure, and both sides of the argument are more credible than they look. The bull case has a real cost asymmetry and real infrastructure veterans behind it. The bear case has fifteen years of wave-energy history and a valuation that has outrun the engineering. Between them, one question settles it: does the Ocean-3 pilot in the northern Pacific this year produce sustained compute uptime at sea? Everything else is pricing the answer in advance.
