14 September 2026
Heard In AI

Why Ramez Naam changed his mind about computers on ocean buoys

Ramez Naam passed on Panthalassa’s early Bitcoin-mining pitch, then invested twice in 2026 at much higher valuations. The company now proposes wave-powered AI computing, cooled by seawater and connected by satellite. Its $140 million Series B is intended to support an Oregon pilot factory and northern-Pacific pilots; cheap electricity and longer-lived chips remain prospective benefits.

A briefing reports one development when it happens. We correct or clarify it later; a new development gets a new briefing. How our formats work

Ramez Naam liked the founders’ idea enough to regret turning it down. Their early pitch was to put computers on wave-powered ocean buoys, initially for Bitcoin mining. He was not sure he cared enough about that use to invest.

Five years later, he had bought in twice at much higher valuations. “I'm an idiot because I said no to these guys five years ago,” he said on Moonshots with Peter Diamandis.

The company is Panthalassa, and Naam’s enthusiasm comes with a financial interest: it is now a portfolio company of his. Its proposal combines three things available far offshore—wave energy, cold seawater and satellite communications—to produce AI computing rather than send electricity back to land.

Send answers ashore, not electricity

Offshore power has a delivery problem: electricity generated far from customers needs a way to reach them. Panthalassa’s answer, described in its May 4, 2026 funding announcement, is to use the power where it is made.

Each proposed autonomous steel unit generates electricity from waves and consumes it locally for AI inference—the work of producing answers from an already-trained model. Results travel through low-Earth-orbit satellites rather than electricity travelling through a subsea cable. What goes ashore is data, not current.

That changes where the company can look for energy. It does not need to choose a site primarily for its proximity to electricity customers or a cable connection to the grid.

Naam sketched the machine itself. A sphere sits at the surface, while a cone about 80 metres long extends into the sea, open at the bottom. As the structure bobs downward on the swell, water moves up inside it and drives a turbine. Shaped channels, he explained, are intended to turn that repeated wave motion into nearly continuous generation.

Following the waves

Naam’s geographical vision reaches well beyond the coast. He pointed to the Southern Ocean around Antarctica as the location of the strongest waves, and described the team’s starting question: what could they build if they went where the waves were bigger?

Stronger, more consistent waves could let a generator produce more electricity relative to its maximum possible output—a measure called its capacity factor. More output from the same hardware would help bring down the cost of each unit of electricity.

Naam put the eventual target at roughly two cents per kilowatt-hour. “That's their target. It will take some scaling to get there,” he said. He expects the route to lower costs to run through factories: modular units, manufactured repeatedly at volume, with improvements carried into subsequent production.

The figure is a prospective electricity cost, not a measured commercial result or a price for delivering AI answers. The computing hardware and satellite connection still have work to do after the waves have generated power.

The ocean as a heat sink

The second attraction is cooling. AI processors turn much of the electricity they consume into heat, which must be removed to keep them operating.

Naam drew a brief contrast with computers in orbit. SpaceX’s June 2026 prospectus describes an orbital computing design using radiators, vapour chambers and active cooling loops. Heat must be carried away from the chips and radiated into space.

At sea, Naam described a more direct path: a heat sink conducts heat from a GPU, the processor doing the AI work, to the device’s steel walls. Cold seawater outside absorbs it. In the conditions he described, that water is around 40°F, or a little above 4°C.

He also sees the possibility of fewer GPU failures and longer operating lives with this cooling arrangement. That is part of his investment case, rather than an established commercial reliability advantage. The devices have “their own set of technical challenges,” he acknowledged, before returning to what attracts him: modular hardware that can be built in factories.

Northern-Pacific pilots come first

The Southern Ocean is Naam’s vision of where abundant waves could take the technology. Panthalassa’s announced next deployments are elsewhere.

In its May 4 announcement, the company reported a $140 million Series B led by Peter Thiel, with Planetary VC among the participants. It said the money would help complete an Oregon pilot factory and accelerate deployments. It also described Ocean-1, Ocean-2 and Wavehopper prototype deployments in 2021 and 2024.

On the podcast, Naam said three units were in the ocean and a fourth would launch soon. The company’s published timetable called for Ocean-3 pilots in the northern Pacific during 2026, ahead of planned commercial deployments in 2027. Those pilots—not a commercial fleet around Antarctica—are the next announced step toward the factory-built ocean computers he once declined to back.

Share this article

Go to the original

Sources & further reading

  1. 01
  2. 02

From the conversation

Podcast episodes

Article history

Updates to this article

Tags

Graylin: cheaper AI could undermine the debt funding data centers

Alvin Graylin argues that AI can become more useful while earning less for the companies financing its infrastructure. His warning centers on cheaper models and local computing weakening cloud revenues, just as NVIDIA proposes financing platforms intended to mobilize more than $500 billion of outside capital.

5 min read

Why AI in orbit needs a launch industry, not just a cheaper rocket

Ramez Naam puts himself between those who dismiss orbital data centers and those expecting an imminent boom. He estimates that launch prices need to fall to roughly a quarter to a tenth of current levels for space-based AI to compete on cost. But cheaper flights leave a separate hurdle: building and launching enough hardware, with permission and reliability to keep flying.

5 min read

Why flexible AI data centers could get power sooner

Expensive AI chips can sit idle while data centers wait for grid connections. On Moonshots, energy investor Ramez Naam argued that accepting less grid power during peak demand could shorten that wait. Workload scheduling and batteries offer two ways to do it, but national estimates of spare capacity are not promises of power at a particular site.

7 min read

NVIDIA’s $500 billion financing plan faces the problem of aging GPUs

NVIDIA has signed memorandums with six financial institutions aiming to mobilize more than $500 billion in outside capital for customers’ AI infrastructure. On Moonshots, the panel debated whether rapidly changing chips can support long-term investments: Salim Ismail warned of stranded assets, Alex argued for financial hedges, and Emad Mostaque explained why older, paid-off GPUs can keep earning.

5 min read