Floating Data Centers Harness Ocean Energy to Power AI

The latest:

A Washington, D.C. startup that has operated quietly since its founding six months ago has raised $20 million to build floating data centers powered by the temperature gap between warm surface water and the cold deep ocean. Drogue told TechCrunch its first multi-megawatt prototype should be at sea in early 2028, mounted on a conventional oceangoing ship.

Details:

• The investors: Drogue told TechCrunch exclusively that its $20 million came from XYZ Venture Capital, Toyota Ventures, Tamarack Global, Dream, FDVC, Garuda Ventures, Flexcap, Boost VC and Slow Ventures, among others. The company did not break out the round structure or valuation.
• The technology: The system is ocean thermal energy conversion, or OTEC, first developed in the 1950s. Drogue describes it as a heat pump run in reverse: instead of spending electricity to move heat, it converts a temperature difference into power.
• How it runs: Warm surface water first cools the servers, then boils ammonia, which turns to gas at -28°F. The gas spins a turbine, then passes through a condenser cooled by water drawn from 1,000 meters down. The ammonia recirculates in a sealed loop.
• The physics claim: Deep water averages about 39°F (4°C), and Sipe argued the differential cannot realistically be exhausted, saying there is no amount of heat the system could pull out that would collapse it. He framed the oceans as a resource that is routinely ignored.
• The scale figure: The oceans absorb an additional 5.8 million terawatt-hours each year as the planet warms, against roughly 180,000 terawatt-hours consumed annually by all of human civilization, according to figures cited in the report.
• The timing argument: Sipe said data center development is effectively a real estate business running on 8-to-10-year timelines, including interoperability with utilities and the independent system operators that manage regional grids. He said AI demand has risen 10x in three years, calling the two incompatible.
• Why data centers: Power generated at sea is hard to use: cabling it ashore is long and costly, and onboard manufacturing would require expensive shipping of inputs and goods. Servers are compact and moving data off the platform is cheap. Each vessel would stay at sea roughly a decade.
• The regulatory step: Drogue has secured approval in principle, a preliminary sign-off, from the American Bureau of Shipping, which sets and verifies ship safety standards. The company is now purchasing its first vessel to build the open-ocean prototype.
• The founder: Sipe flew E-2D Hawkeyes off U.S. Navy aircraft carriers before founding the company. He said he interviewed people who worked on earlier OTEC efforts, and that manufacturing techniques for piping and heat exchangers have changed substantially since the 1980s.

Background:

The underlying concept was tested in the 1980s and then abandoned. Sipe said the engineers he consulted told him the physics worked, and that the obstacle was raising enough capital to build at a scale large enough to hit the economic sweet spot.

Between the lines:

The report draws a parallel to Elon Musk’s orbital data center plan: both strategies abandon the grid and move compute to the power source. Drogue’s version relies on ships rather than rockets. The comparison also underlines the bet’s nature — the 2028 prototype, not the funding or the approval in principle, is the first real test of whether 1980s-era physics clears today’s engineering bar.

What’s next

The purchase of Drogue’s first ship, and whether the multi-megawatt prototype reaches open water on schedule in early 2028. Also watch whether approval in principle from the American Bureau of Shipping advances to full certification.