A six-month-old startup raised $20M to put AI servers on ships powered by ocean thermal energy conversion (OTEC). No one has ever run OTEC commercially above about 1 MW. The round probably covers only part of the prototype, and the federal license a U.S.-owned OTEC “plantship” needs has never been issued.
Probably not. IRENA puts small OTEC plants at $16,400–35,400 per kW (2010 dollars, CONFIRMED). For the 2.5 MW prototype in Drogue’s own memo, that works out to about $60M–$130M in today’s dollars (EST). The ship, servers and cold-water pipe come on top of that.
Not commercially. The physics works in demonstrations: Mini-OTEC (1979), Makai’s 100 kW Hawaii plant (2015) and Japan’s Kume Island (2016). But in more than 90 years of prototypes, none has run commercially above about 1 MW (CONFIRMED, POWER).
That’s unclear. The 1980 OTEC Act says a U.S.-owned OTEC “plantship” needs a license. NOAA removed its licensing rules in 1996 and said it had never received an application. Drogue has said nothing about this publicly.
Key Finding: Drogue is aiming at a real problem: AI data centers wait years for grid power. But $20M probably covers only a fraction of a multi-megawatt OTEC prototype, and the technology has never run commercially above about 1 MW. A U.S.-owned OTEC ship would also need a federal license that has never been issued, under rules NOAA removed in 1996. This is a venture bet on rescuing OTEC, not a data center business.
OTEC runs a heat engine on the temperature gap between warm surface water and cold deep water. That gap is about 20–24°C, so even the theoretical maximum efficiency is only about 7–8% (DERIVED). Every step in the loop below costs some of that.
Surface water at about 25–28°C. A 2.5 MW unit needs roughly 12 m³/s of it (DERIVED).
Warm water cools the servers first. Server heat is low-grade, so this barely raises power output.
A closed ammonia loop evaporates in the heat exchangers. This is where Drogue says manufacturing has improved.
Real-world net efficiency has been 1–3% (CONFIRMED). Pumps take about 20–30% of gross output (EST).
Water at about 4°C, pumped from 1,000 m deep. This pipe is the industry’s main failure mode.
The cooling synergy is overstated. Running warm intake water past the servers adds heat load but does little for the turbine. CleanTechnica made the same point about Optimal Transit’s rival “Kraaken” design, which also combines OTEC with AI compute. The real advantage is location: data is cheap to send off a ship, and power is expensive to send ashore.
That is how Sipe tells it. The record is worse. The DOE’s OTEC-1 was stopped after about four months and $50M. Lockheed Martin and Reignwood announced a 10 MW plant in 2013, and we found no evidence it was built. Global OTEC’s 1.5 MW “Dominique” barge has a Lloyd’s Register approval in principle and has slipped from 2025 to 2026. OTI holds three ABS approvals in principle for its designs and has built nothing. An approval in principle is a common starting point in OTEC, not proof of progress.
Drogue’s F4 profile calls it “the world’s first offshore floating data centers.” Underwater and barge data centers already run commercially. What Drogue would be first at is OTEC-powered compute.
Wave-powered floating AI compute nodes. $140M Series B in May 2026, reportedly led by Peter Thiel, at a valuation of about $1B. About $210M raised in total (CONFIRMED as reported), roughly 10x Drogue’s capital.
The most direct rival: OTEC, desalination and AI compute on one ship. Claims 100 MW with 60 MW of compute for about $587M all-in (EST, company claim). Raising a Series A.
Underwater data centers running commercially off Hainan since 2023, plus a Shanghai project of about 24 MW powered by offshore wind. More than ¥100M raised (CONFIRMED).
Aikido pairs floating wind with submerged pods: a 100 kW demo in Norway now and a 10–12 MW U.K. site planned for 2028. Subsea Cloud runs a pod of about 1 MW off Washington State.
Makai Ocean Engineering (Navy-funded 100 kW plant), Global OTEC, OTI, Kume Island and Korea’s KRISO (338 kW achieved from a 1 MW design). Each is a likely supplier, rival or cautionary tale.
Hyperscalers want gigawatt campuses connected by fiber. A ship of a few megawatts with satellite-only links suits inference, not training. That limits the market Drogue can serve (EST).
The story keeps changing. Drogue’s memo says it is “not a data center company,” while TechCrunch presents it as one. The F4 profile describes “untethered semi-submersibles”; TechCrunch describes a converted ship. The memo says “2.5+ MW”; the announcement says only “multi-megawatt.” At six months old, some drift is normal. Investors should still ask which version the $20M is paying for.
Seven risks the $20M does not resolve. Two of them, licensing and storm survival, Drogue has never addressed in public.
At IRENA’s small-plant costs, a 2.5 MW prototype costs about $60M–$130M (EST), before the ship, GPUs and cold-water pipe. A much larger round has to close before early 2028.
No OTEC system has run commercially above about 1 MW in more than 90 years. A 1,000 m cold-water pipe hanging from a ship has never been proven. Net efficiency is around 2%.
Under 42 U.S.C. §9111, a U.S. citizen needs a license to own or operate an OTEC plantship. NOAA removed its licensing rules in 1996 and had never received an application. A foreign license brings sanctions and data-sovereignty problems.
The temperature gap OTEC needs exists mainly in the tropical cyclone belt. A ship can sail away from a storm; a 1,000 m pipe can’t. POWER notes a pipe failure “could endanger the vessel.”
Without a cable to shore, the ship relies on satellite links. That works for inference but not for training or data-heavy enterprise work. Swapping GPUs every 3–5 years on a 10-year ship means offshore maintenance and salt corrosion.
Panthalassa has about 10x the capital and is deploying first. Optimal Transit is chasing the same OTEC-plus-AI niche with a much larger design.
Only one founder is identified publicly, and the company is about six months old. No marine, thermal or ammonia-cycle engineering leaders are named. We could not find an independent profile of Sipe’s background beyond the TechCrunch report.
Drogue is a bet on rescuing OTEC, with AI servers attached. The pitch has a real logic: data is cheap to move off a ship and power is expensive to move ashore. But the hard parts haven’t been solved: a 1,000 m cold-water pipe, net efficiency around 2%, and prototype costs that likely exceed the round several times over. Before the next check, diligence should get answers on three things: the licensing path under the 1980 OTEC Act, a storm-survival plan for the pipe, and a named compute customer.
Based entirely on publicly available information, including the TechCrunch report of November 13, 2026. Numbers are labelled CONFIRMED (published source), DERIVED (calculated from sourced inputs) or EST (estimate or company claim). We found no company website, privacy policy or investor thesis post. Drogue’s Notion pages are cited from search snippets only.
We build and maintain websites, help businesses of every size put AI to work, and run growth marketing that’s held to the same standard as our research.
