Panthalassa is building ocean-based energy platforms to power distributed AI compute and synthetic fuels

Summary

Panthalassa, an 80-person Portland-based deep tech company, is building ocean-deployed wave energy platforms designed to power distributed AI compute and synthetic fuel production. The company draws heavily on aerospace talent from SpaceX, Blue Origin, and Virgin Orbit, and has been in development for nine years, with the founder previously doing energy research at MIT and investment work at Bridgewater Associates.

The Technology

Each platform, called a node, spans 10 to 20 meters across at the surface and extends 80 to 100 meters into the water column. The design is intentionally minimal: no moving parts except a single water turbine. Wave motion drives water into an onboard reservoir, which then powers the turbine in a configuration analogous to a hydroelectric dam. The axisymmetric steel shell construction uses single-curvature surfaces, enabling fabrication via standard plate rolls and automated welding heads — the same manufacturing approach used for offshore wind turbine monopiles.

Energy Output and Compute Capacity

Each node generates between 200 kilowatts and 1 megawatt, with a nominal output around 400 kilowatts. That is sufficient to run a few hundred to roughly 1,000 GPU equivalents per unit. Projected energy cost sits at 2 to 3 cents per kilowatt hour, with further reductions anticipated at scale. At gigafactory-equivalent capex, the company estimates it could produce 20 gigawatts per year of capacity — a figure it positions against the West's structural difficulty in adding tens of gigawatts of new terrestrial energy annually.

Self-Propulsion and Maintenance

The nodes are self-propelled without propellers, generating thrust from hull geometry alone. Panthalassa has demonstrated figure-eight navigation in open water trials. The operational model calls for units to deploy near shore, transit autonomously to deep-water energy resources, operate for three to five years, then return independently for payload swaps. This removes the need for continuous crewed servicing and allows GPU or inference chip payloads to be cycled based on mean-time-between-failure schedules.

AI and Market Positioning

Panthalassa identified computing as its primary platform application as early as 2017, with patents referencing AI and brain simulation from that period. The current thesis centers on the demand for cheap token throughput at scale, targeting long-running inference workloads and reinforcement learning training as the dominant AI compute paradigms going forward. The company frames ocean-based distributed energy as a faster-scaling alternative to terrestrial data center buildout, which faces permitting, grid interconnection, and land constraints.

Bitcoin mining at 2 to 3 cents per kilowatt hour remains a stated fallback revenue source. A pilot manufacturing line in the Pacific Northwest is partially operational in January, with full operation expected in spring. Once validated, Panthalassa intends to replicate the factory model in other locations.