The problem of where to put AI infrastructure has a new proposed solution: the Pacific Ocean.
A startup called Panthalassa is developing floating data centers that sit mostly underwater, powered by ocean waves and cooled by seawater. This week it emerged that Peter Thiel, the venture capitalist who co-founded PayPal and Palantir Technologies, has invested $140 million in the company, pushing its valuation toward $1 billion.
The company's pitch is essentially a response to the same pressures behind this week's North American grid emergency alert: land-based data centers are straining electricity grids, they require expensive long-distance transmission infrastructure, and they generate enormous heat that demands energy-intensive cooling. Panthalassa's answer is to take data centers off the grid entirely and move them to an environment that provides both power and cooling for free.
The basic unit is an 85-metre-long "node," roughly the length of a large building, that sits mostly submerged. Inside the sealed structure, AI servers run on power generated by the device itself. As ocean waves cause the structure to rise and fall, that movement drives water through internal pipes that spin turbines and generate electricity. The surrounding ocean simultaneously serves as a constant cooling medium, eliminating the need for the power-hungry cooling systems that typically consume 30 to 40 percent of a conventional data center's energy budget.
The nodes are designed to move autonomously using ocean currents rather than engines, and to relay data back to shore via satellite connections. Panthalassa's engineering team reportedly draws on alumni from SpaceX, Tesla, and NASA.
The concept of underwater computing is not entirely new. Microsoft ran a sealed underwater data center pilot called Project Natick between 2018 and 2020, deploying a pressurized capsule on the seabed off the Orkney Islands in Scotland. The experiment found that the sealed submarine environment actually reduced hardware failure rates: eliminating oxygen and humidity removes the primary causes of corrosion in conventional data centers. Microsoft eventually wound the project down, citing the logistical complexity of maintenance and recovery. Panthalassa's bet is that the engineering challenges of deepwater deployment are now more tractable than the political and regulatory challenges of securing land, permits, and grid connections on shore.
Wave energy is where that bet gets most speculative. Generating electricity from ocean waves has been technically demonstrated for decades, but has not become commercially competitive with other renewable sources at meaningful scale. The engineering demands of maintaining turbines and electronics in a corrosive marine environment, recovering nodes for servicing, and managing the latency implications of satellite-only connectivity for latency-sensitive AI workloads all represent real challenges that a pilot deployment will need to resolve at a cost the business model can absorb.
At the same time, the urgency of the demand problem is genuine. North America's grid watchdog issued its rarest-ever reliability warning this week specifically about AI data centers, and European facilities are running at half capacity because their electricity grids cannot feed them. Grid constraints are projected to delay around one in five planned data center builds globally by 2030. In an environment where the grid is becoming a hard limit on AI expansion, the appetite for unconventional alternatives is real, and Thiel's $140 million signals that at least some investors believe the ocean's unconventional constraints are more manageable than land's conventional ones.