A plan to generate electricity from ocean temperature differences and use it to run offshore AI data centers is starting to take shape. According to TechCrunch's report of November 13, 2026, US startup Drogue has raised $20 million and plans to put a prototype to sea in early 2028. The company aims to generate power from ocean temperature differences, cool servers with seawater, and run AI inference without relying on an onshore power grid. However, not all of the electricity generated can go to computing. Even after subtracting the power needed to pump seawater and run cooling equipment, enough must remain to keep AI servers running reliably. That question is the key to whether the idea becomes practical.
Moving power and cooling out to sea
Drogue founder Sam Sipe was introduced as an exhibitor at "The Ocean Awakens," an event held on September 21, 2026. According to the organizer, Plastic Odyssey Fund, the company aims to use ocean thermal energy to build an offshore power source independent of the grid, supporting the next generation of AI inference. Investor Toyota Ventures' portfolio page also describes a concept that combines ocean-heat power generation with seawater cooling, significantly reducing the burden of the auxiliary equipment needed to operate a data center. The idea is to place power-hungry data centers next to the generating equipment, avoiding the wait for connection to onshore grids. That said, what has been presented so far is a future plan, and actual performance has not been confirmed.
The technology Drogue aims to use is called ocean thermal energy conversion (OTEC), which generates electricity from the temperature difference between warm surface water and cold deep water. Seawater serves as both the heat source and the cold source for power generation, and is also used to cool the data center. Concentrating power generation and computing equipment offshore is the defining feature of the concept.
How a temperature difference of about 20°C produces electricity
According to the US Energy Information Administration (EIA), a typical OTEC system uses heat from warm surface water to vaporize a working fluid, and the vapor spins a turbine and generator. After passing through the turbine, the vapor is cooled back into a liquid with cold seawater pumped up from the deep ocean, and the cycle repeats. Ammonia and similar substances are used as working fluids, and an explanation from Hong Kong's Electrical and Mechanical Services Department describes a method of drawing cold seawater from a depth of about 1,000 meters. Rather than burning fossil fuels, the system generates power from the temperature difference in seawater.
According to the EIA, OTEC generally requires a temperature difference of at least about 20°C. The US National Oceanic and Atmospheric Administration (NOAA) likewise identifies tropical waters, where the difference between surface and deep water exceeds 20°C year-round, as promising sites. Unlike wind or wave power, whose output tends to fluctuate with the weather, using seawater temperature differences may make it possible to obtain relatively stable power. However, this is a characteristic of OTEC technology in general, and does not indicate where Drogue plans to deploy its equipment or what its actual uptime will be.
OTEC itself is not a new technology, and demonstrations already exist. According to the EIA, a 105-kilowatt demonstration plant began operating in Hawaii in 2015 and supplied power to the local grid. What sets Drogue's concept apart is that it combines this generation method directly with an offshore AI data center, consuming the electricity where it is produced.
Net power available to run AI servers is the key to commercialization
In OTEC, not all the electricity a generator produces can be supplied externally. Power is also needed to run the pumps that draw cold water from the deep ocean and circulate the working fluid. Hong Kong's Electrical and Mechanical Services Department also notes that OTEC efficiency is generally low because pumping seawater and transmitting power to land require a great deal of energy. In Drogue's case, the need to send power to shore may be reduced, but the power consumed by the generating equipment itself, as well as the power needed to cool the AI servers, must also be taken into account.
Evaluating Drogue's concept therefore requires looking not just at the generators' gross output but at net output after subtracting the power needed to run the facility. How much the integration of generation and cooling described by Toyota Ventures actually helps will also need to be verified with measured data. In the upcoming prototype, the following items will be especially important for judging the concept:
| Planned technology | What the real system needs to show |
|---|---|
| Power generation from ocean temperature differences | Net output after subtracting the power needed for seawater pumping and facility operation |
| Server cooling with seawater | Power consumption of the cooling equipment, and the reduction in auxiliary power compared with conventional cooling methods |
| Offshore AI inference | Processing performance, power consumption, and a track record of stable operation when generation and AI processing run simultaneously |
Neither the Plastic Odyssey Fund's event page nor Toyota Ventures' portfolio page presents measured values for these items, or any operating record of running power generation and AI inference together. Information is also lacking on Drogue's specific deployment waters, the facility's uptime, and the cost of AI processing. The prototype will need to show how much computing load can be sustained, and in what temperature-difference conditions.
The offshore data center Drogue is pursuing is an attempt to complete both power supply and cooling for AI computing infrastructure within the ocean environment. If the prototype planned for early 2028 can verify net power output and AI processing performance at the same time, it will become possible to evaluate concretely the potential of a new kind of data center that does not depend on onshore grids. What matters is not whether electricity can be generated from ocean temperature differences, but whether that power can run AI servers continuously and economically.