While Everyone Raced to Build Data Centers on Land, This Founder Built Them for the Sea. Here Is Why That Patience Just Paid Off.
By Creatives Takeover · July 1, 2026
The smartest AI infrastructure bet 🌊
In 2016, Garth Sheldon-Coulson had an MIT master's degree, a Harvard law degree, and an idea most people in energy would have called a waste of both. Put AI data centers in the ocean. Not near the coast. Hundreds of miles out, in deep water, where waves carry the most energy on the planet.
He and cofounder Brian Moffatt started building anyway.
For almost a decade, nobody paid attention. Land based data centers were cheap, fast to permit, and the obvious choice for every hyperscaler racing to add compute. Meanwhile Panthalassa tested prototypes off the coast of Washington state, refining a technology most investors saw as a science experiment with no market.
Then the market caught up to the problem Panthalassa had already solved.
The constraint nobody wanted to talk about
AI's energy appetite exploded past what land infrastructure could absorb. The International Energy Agency projects AI related energy consumption will grow 30% a year through 2030. Tech companies are on track to spend $765 billion on data centers in 2026 alone. Communities near proposed sites started fighting back over utility rate hikes, water use, and noise. Land, once the easy part, became the bottleneck.
Sheldon-Coulson had been saying for years that solar, nuclear, and the open ocean were the only three energy sources on the planet with tens of terawatts of untapped capacity. Two of the three were already crowded. The third was sitting there mostly ignored.
How the technology actually works
Panthalassa's answer is a node: an 85 meter steel structure, roughly the height of Big Ben, that floats mostly submerged in deep water. As waves move the structure, water is forced through an internal turbine, generating electricity with no engine, no hinges, no gearboxes. That power runs AI chips sealed inside a pressure vessel, cooled directly by the surrounding seawater instead of the industrial cooling systems land based centers depend on.
There's no cable to shore and no anchor to the seabed. Each node is self-propelled, towed out horizontally, then flips upright and navigates to position using the hydrodynamic shape of its own hull. Compute results transmit back to land through SpaceX's Starlink network, the only connection the system needs.
The company says the approach could bring power generation costs as low as $0.02 per kWh, undercutting land based alternatives by a wide margin.
The money finally showed up
Earlier this year, Panthalassa closed a $140 million round led by Peter Thiel, pushing the company's valuation close to $1 billion. Salesforce CEO Marc Benioff, PayPal and Affirm cofounder Max Levchin, and longtime investor John Doerr, one of the earliest backers of Google and Amazon, joined the round.
Thiel didn't mince words about why. "The future demands more compute than we can imagine," he said. "Extra-terrestrial solutions are no longer science fiction. Panthalassa has opened the ocean frontier."
Panthalassa is using the capital to finish a pilot manufacturing facility in Oregon and deploy its next generation of nodes in the northern Pacific, with commercial deployment targeted for 2027.
Why patience beat speed
Ocean based compute isn't a new idea. Microsoft ran undersea servers off Scotland for years through Project Natick before shelving the project in 2024, even though the trial showed lower failure rates than land based systems. Chinese firms have deployed underwater centers near Hainan and Shanghai. Singapore's Keppel has built floating data centers of its own.
None of them built a company around it for ten straight years while the rest of the industry looked elsewhere.
That's the part worth sitting with. Panthalassa didn't win because the idea was original. It won because Sheldon-Coulson and Moffatt kept refining a hard technical problem long after it stopped being fashionable, so that when the constraint finally hit, the solution was already built, tested, and ready to scale.
Most founders chase the trend that's obviously heating up. The rarer bet is building for a constraint that hasn't bitten yet, and being willing to look irrelevant until it does.