Ocean thermal energy conversion has been the power technology of the near future for more than 140 years. The physics has never been in doubt. A heat engine running between warm surface water and cold water pulled from a kilometre down can spin a turbine day and night, with no fuel and no weather window. The problem was always the same: nobody had a load that could sit on top of the resource. Drogue, a Washington, D.C. startup founded in 2025, thinks AI data centers are that load. On 13 November it told TechCrunch it had raised $20 million from XYZ Venture Capital, Toyota Ventures, Tamarack Global, Garuda Ventures, Slow Ventures and others, and that it will put a multi-megawatt prototype on a conventional ship in early 2028.
The premise is sound. The timetable is still a promise. Floating a heat engine above its cold reservoir is a marine-engineering problem with a long history of cost overruns, and Drogue has not published a net power figure, a cost per megawatt-hour, or a single named customer. What it has is the first plausible answer to the question that killed every previous OTEC project: who buys the electricity?
A technology that has been almost ready since the 1880s
OTEC was first proposed in the 1880s, built in small pilots in the 1950s, and tested again in the 1980s before being shelved. The cycle is simple. Warm surface water boils a low-pressure fluid such as ammonia, the vapour spins a turbine, and cold water from depth condenses it back into a liquid inside a sealed loop. TechCrunch describes the arrangement as a heat pump run in reverse. The deeper the cold pipe, the larger the temperature gap, and the more work the cycle can extract from the same seawater.
The engineering has never been the weak link in theory. It has been the weak link in practice. A commercial plant has to move enormous volumes of seawater through a pipe long enough to reach the thermocline, then through heat exchangers the size of buildings. Salt water corrodes. Biofouling coats every surface. Storms stress the cold-water riser, which is the largest single structure on the platform. Each problem has a solution. None of them is cheap at sea.
The historical failure mode was capital, not thermodynamics. The plants were small, the output was modest, and the customers were islands paying diesel prices. Those customers are real, but they are few and slow to sign. In October 2026 the International Energy Agency's ocean-energy programme and the Ocean Thermal Energy Association signed a cooperation agreement, describing OTEC as promising while noting that technical and economic challenges remain. That is a polite way of saying the industry still has not crossed the line.
Concrete progress exists. Global OTEC, a rival developer, said on 2 November that its floating structural prototype had completed a year of continuous testing off Gran Canaria. That is a structural milestone, not a power milestone. Drogue's own "approval in principle" from the American Bureau of Shipping is the same kind of signal: the classification society does not see an obvious reason the vessel will sink. Neither announcement shows a kilowatt-hour sold.
The customer that finally showed up
Data centers change the customer problem in two ways. They are power-hungry at a scale islands never were. A projection cited by Garuda Ventures puts data-center consumption above 10% of total US electricity by 2030. They can also float. Servers are compact relative to a ship, and moving data off a platform is far cheaper than moving fuel or finished goods onto one.
The land-based alternative keeps getting harder. Grid interconnection queues run for years. The largest gas-turbine makers are reported to be sold out past 2030. Hyperscalers have started building their own generation. Amazon is reported to be building 7.7 GW in Texas, xAI stood up a 1 GW gas microgrid in Memphis, and OpenAI's Abilene campus runs on about 1.2 GW of dedicated turbines. Each of those is a way of paying to leave the grid. Drogue's version is to leave the land as well.
There is a second saving that matters more than the headline power number. Sea water can cool the servers directly. Garuda estimates that land-based facilities spend 30–40% of their power on cooling; a platform sitting in cold deep water removes most of that overhead. More GPUs per megawatt is a better sales pitch than more megawatts.
That is the believer's case, and it is coherent. The ocean is the largest solar collector on the planet, it never sends an interconnection bill, and the heat it absorbs — about 5.8 million terawatt-hours a year, against roughly 180,000 terawatt-hours of annual human energy use — dwarfs anything the AI build-out can consume. The resource is not the constraint.
What the thermodynamics actually permit
The constraint is geography and capital. OTEC needs a year-round temperature difference of at least 20°C between the surface and the cold intake. That limits deployment to tropical waters. Deep water sits near 4°C at about 1,000 metres, so the resource is real, but only in a band around the equator and only where a cold pipe can be installed and serviced.
The temperature gap is also small, which means the cycle converts only a thin slice of the heat it moves into electricity. A practical plant is therefore a water-moving business: large pipes, large pumps, large heat exchangers. Capital cost per installed megawatt is high, and the first-of-a-kind vessel carries every learning cost at once. Drogue has not said what its prototype will cost, what it will produce net of parasitic loads, or what it expects to charge. Dealroom notes that no offshore data center has yet shown it can run reliably or economically at scale.
✔ The case for
+ Co-located cooling removes 30–40% of the power land facilities spend on chillers.
+ AI compute is already paying premiums to bypass grid queues, so it can be sited offshore.
Criteria for confirmation: a named customer PPA and a measured net-megawatt figure from the 2028 prototype.
✗ The case against
− Cold-water pipes, corrosion and biofouling are unresolved at vessel scale.
− $20 million is small against the capital cost of ships and heat exchangers.
Criteria for refutation: the prototype slips past 2028, or runs below its stated net output for a full season.
This is where the argument should be decided, and it will not be decided by press releases. The relevant comparisons are already in the water. Fervo Energy reported $11.9 billion of contracted revenue in its third-quarter 2026 results, built on geothermal wells that sell firm power to hyperscalers on land. As we wrote in October, Cowboy Space's Big Iron turns a rocket upper stage into a 1 MW orbital data center — a different answer to the same question of where to put compute next to cheap energy. Drogue's platform has to beat both on delivered cost, not on elegance.
The $20 million question
The round is small, and that is informative. Twenty million dollars does not build a commercial OTEC plant. It buys a ship, a prototype, and the right to raise the next round on data instead of a deck. Drogue's investors are buying an option on a specific bottleneck: the moment when AI demand for power outruns the grid's ability to deliver it without a decade of paperwork.
What would settle the argument is unglamorous. A named customer with a signed power purchase agreement. A measured net output from the 2028 vessel, after pumps and hotel loads. A survivability record through a storm season. A cost per megawatt-hour that can be compared with Fervo's geothermal power or a gas plant behind the meter. Until those exist, OTEC remains a technology with a better customer story than it has ever had, and the same old engineering risk.
The first ship will not prove that the ocean can power the AI economy. It will answer a narrower question: whether a heat engine can run for a year at sea, unattended, at the output its builder promised. That number is the whole bet.