Among big future businesses Elon Musk is selling investors in newly public SpaceX is his plan to put data centers in space: solar-powered satellites, spread across a vast network, processing information in space and beaming it back to Earth. As pitches go, it has the clean geometry of a Musk bull case. It’s the kind of “I want to die on Mars, just not on impact” sci-fi idea the newly minted trillionaire is famous for. And it’s particularly well timed: the AI feeding frenzy is in overdrive, but the terrestrial data centers they require are becoming an unwanted menace in many communities, raising utility rates, creating noise and pollution, and generating few local economic benefits.

SpaceX hopes to begin launching orbital data centers in 2028, though its IPO filing gives no cost estimates for such a system. It does, however, include the kind of caveat that sits in a securities filing like a flare on the runway: The plan involves “significant technical complexity, unproven technologies, or technologies that do not exist or may require significant advancement, and such initiatives may not achieve commercial viability.”

SpaceX lawyers meant it as a warning. Musk could probably plaster it on the lobby wall.

But if the goal is simply to move data centers off land and run them at lower cost, there’s a far better option: the ocean. It’s far away from taxpayers, zoning fights and the sudden arrival of hyperscale neighbors. And it may be a climate-friendly source of power and a cheap way to cool massive data centers.

“What we’re doing is totally crazy”

This is where Panthalassa wants to go. The Portland, Oregon startup, backed by Peter Thiel and a raft of Silicon Valley venture firms, has spent the past decade developing floating data centers that generate their own electricity from open-ocean waves and cool themselves with cold seawater. It expects commercial units to be operational in 2027, a year before SpaceX says it may begin putting compute satellites in orbit, with all those securities filing caveats.

“What we’re doing is totally crazy,” CEO and cofounder Garth Sheldon-Coulson told Forbes. “We're the first company that’s going to the middle of the ocean to do this.”

The Ocean-2 prototype node that Panthalassa (Greek for “all sea”) has been testing off the coast of Washington state since last year looks less like a data center than a marine-industrial lollipop: a 70-meter steel tower submerged below the surface, with a bulbous head floating above the waterline. As it bobs up and down in the waves, water pumps up through the neck and into the spherical reservoir at the top, then flows through a turbine that can generate up to a megawatt of continuous electricity. The unit Panthalassa plans to deploy next year will be loaded with chips and computing hardware to run AI learning operations on board, beaming out data via satellite, just like Musk’s SpaceX concept.

“This will be the lowest cost way to do large segments of AI computing, inference reinforcement learning, without any emissions at all,” Sheldon-Coulson told Forbes.

Tapping the ocean for energy has intrigued scientists for over a century. It has also humbled them. No large-scale systems or techniques have yet proven commercially viable. The interest persists because it's a massive resource. One assessment from the International Energy Agency estimated wave power could produce thousands of terawatt-hours of electricity annually. Even capturing a fraction of that on a consistent basis would be a game-changer. The ocean, inconveniently, has had a vote in every prior business plan.

Panthalassa isn’t the first company to see the ocean as a data-center workaround. Microsoft spent years testing undersea units connected to onshore power off the coast of Scotland, before ending the research in 2024. China is also experimenting with undersea data centers powered by wind turbines. Those projects use the ocean mainly as a cooling system. Panthalassa wants it to be a power plant as well.

“We operate in the deep ocean where the wave energy is most abundant, as opposed to shallow coastal waters,” Sheldon-Coulson said. “Our nodes are self-propelled and can reposition themselves autonomously. There is no connection to the seafloor.”

He cofounded Panthalassa in 2016, after earning a master's at MIT and a Harvard law degree, with engineer Brian Moffatt, who was also researching wave energy. Chief engineer Daniel Place came from SpaceX, while other engineering staff came from tech and aerospace giants, including Google, Blue Origin, Apple, Boeing, Amazon and Tesla. In May, Panthalassa raised $140 million in a Series B round for its first commercial deployment, backed by Thiel, John Doerr, Marc Benioff’s TIME Ventures, Max Levchin’s SciFi Ventures, and tech funds including Gigascale Capital, created by Mike Shroepfer, who oversaw construction of data centers for Meta when he was its CTO.

Shroepfer sees the concept of floating data center buoys as audacious. He also thinks it is a possible answer to anti-data-center backlash and the brutal economics of trying to feed AI’s appetite for power and cooling.

“We're going to use literally 10 terawatts of untapped wave power in a part of the ocean that no shipping is in. There's nothing there,” he said.

Both space-based and sea-based data centers are attempts to claim free energy: sunlight in orbit, waves in the Southern Ocean. Schroepfer’s argument for the ocean version begins with logistics. Putting hardware at sea is hard. Putting hardware in orbit is the same problem with a massive rocket bill attached: SpaceX charges up to $90 million per launch.

“If you compare how much it costs to launch a ton into the ocean versus a ton into space, the answer is it’s a hundred times more expensive to launch it into space,” Shroepfer said. “So we've got a 100X cost advantage. … Let's say we're off by a factor of 1. We've still got a factor of 10X better in terms of cost.”

Panthalassa wants to deploy hundreds – eventually thousands – of free-floating data center buoys in the seas between the South Pole, South America and Africa, because it has the steadiest, most powerful waves and is far from shipping lanes. The energy they generate would be used on site, as transmitting electricity back to shore would be far too costly. If its data center plans work, Panthalassa’s next goal, starting in the early 2030s, is to also use its floating electricity nodes to generate fuels like carbon-free hydrogen or ammonia, using desalinated seawater and electrolyzers to split the H2O.

“We get that onto ships, and we bring it to land where it’s needed,” Sheldon-Coulson said. Making green hydrogen in this way, with no carbon emissions, would cost a fraction as much as doing so with solar energy, he said.

His case rests on the price and consistency of the power. “We have an enormously low cost of energy. Our cost of electrons is down around 2 cents per kilowatt hour, and we also have a very high capacity factor, which means we are on almost all the time, with over 90% capacity factor,” he said. “You can envision that what we're trying to build is this whole new energy ecosystem out there using super-abundant energy in the middle of the ocean, far from land, far from conflicting uses, to supply these two goods that humans really need: lots of computing and clean fuel.”

First, the machines have to survive the place they are meant to exploit. The Southern Ocean is particularly rough, owing to the absence of large land masses, allowing unimpeded buildup of the planet's most powerful wave system.

To help ensure that, Panthalassa’s nodes have relatively few moving parts to make energy and are built with the types of sturdy industrial materials heavy ships use: thick steel with coatings of zinc or aluminum. They should last at least 15 years, according to Sheldon-Coulson. “We plan to swap out the compute payload about every five years.”

The cooling story is simpler than the power story. And it is particularly compelling at the moment because data centers are turning cooling into a water, power, permitting and irate citizen problem on land. The average temperature in regions where Panthalassa plans to deploy its nodes is just 10 Celsius (50 Fahrenheit). At that temperature, you don’t need data center-specific chillers, cooling towers or fresh water.

 “It's a big bet, but it would be a place to put a lot of compute that no one would ever have to worry about.” 

“It’s much more efficient, much lower cost, much lower resource consumption and it provides a much better environment for the chips, which causes them to last longer as well,” Sheldon-Coulson said.

Cooling may prove to be the biggest challenge for Musk’s space-based data center concept, since satellites orbiting the earth operate in an environment in which temperatures fluctuate from as cold as -170 to 120 Celsius. And because they’re also in a vacuum, which prevents heat from being expelled through air cooling, they need sophisticated thermal systems to prevent damage to sensitive computing systems.

Panthalassa’s CEO declined to make a direct cost comparison with Musk’s orbital concept for obvious reasons, but it’s easy to extrapolate from his remarks. “We will be significantly lower cost than data centers on land. And I think that means we will also be quite a bit better than orbital concepts, at least for the foreseeable future,” he said.

There’s still a real chance Panthalassa’s plan fails. Wave energy has a long history of eating elegant machines, and the Southern Ocean can be an unfriendly, downright malicious laboratory. But the upside potential is huge.

That’s what compelled Shroepfer to invest. “It's a big bet, but it would be a place to put a lot of compute that no one would ever have to worry about.”

MORE FROM FORBES

ForbesGM Doubles Down On Energy Business To Serve Data Center Electricity DemandForbesHow The Iran War Oil Shock Is Helping Launch A Market For Electric TugboatsBy Alan OhnsmanForbesTurning Idle California Farmland Into A Clean Energy PowerhouseBy Alan Ohnsman