Panthalassa Opens First Ocean Wave Factory for AI Nodes

Three years from now, the AI inference job that answers your coding question might be processed on a floating steel orb in the North Pacific, powered entirely by ocean waves and cooled by seawater. That is not a speculative pitch deck slide. On November 6, Panthalassa cut the ribbon on its first industrial factory in Vancouver, Washington, designed specifically to mass-produce those nodes at scale. The company's argument is blunt: the land-based data center industry is running out of land, power, water, and political tolerance all at once. The ocean, by contrast, has an inexhaustible supply of all four.

What the Vancouver Ribbon-Cutting Actually Signals

The factory occupies a large open warehouse near the Columbia River, in what was once part of the Kaiser Shipyard that built Liberty Ships during World War II. According to OPB, which attended the November 6 ribbon-cutting, Panthalassa began prototype fabrication at the site in 2022. What changed on November 6 is not the address. It is the production model: the facility is now transitioning from custom, one-at-a-time assembly to a repeatable assembly line. Workers will roll plate steel and use robotic welding systems to build the company's floating compute nodes. When fully staffed, the facility will employ about 50 people, and completed nodes can be loaded onto barges and shipped down the Columbia River for open-ocean deployment, according to Tech Insider's factory coverage.

CEO Garth Sheldon-Coulson described Vancouver as "the first step" toward a network of node factories worldwide. That framing matters. Panthalassa is not selling a single facility or a research demonstration. It is positioning itself as a new category of infrastructure manufacturer, one whose product happens to operate in international waters rather than on campus land parcels. Each factory site feeds the same assembly blueprint into a coastal manufacturing chain, allowing the company to scale production without negotiating land deals in constrained markets. The 2027 commercial availability date hangs entirely on the upcoming Ocean-3 Pacific pilot proving that the production design can survive real ocean conditions and deliver competitive compute. The Vancouver factory is the moment when that claim becomes testable at scale.

How a Wave-Powered Orb Actually Runs AI Workloads

Panthalassa's nodes are described as white orbs larger than a sailboat, with a hollow tube running down into the water column beneath them. As the platform rises and falls with wave motion, water moves through internal turbines, generating electricity. That power stays on-site, running AI hardware directly. No undersea cable brings shore power in. No connection to a utility grid is required. The surrounding ocean acts as a passive heat sink, cooling chips to operating temperature without drawing a single gallon from a municipal water system.

AI inference results travel by low-Earth-orbit satellite network, according to Interesting Engineering's coverage of the Ocean-3 platform. Commands enter the floating node through a satellite downlink; completed AI outputs return through the same uplink. Panthalassa targets batch and asynchronous workloads where real-time latency is not the binding constraint: large-scale training runs, retrieval-augmented generation pipelines, overnight inference jobs, and similar tasks that can tolerate a few hundred milliseconds of additional round-trip delay in exchange for virtually unlimited power supply at competitive cost.

Earlier field tests of the Ocean-1, Ocean-2, and Wavehopper designs ran in 2021 and 2024, verifying the basic energy generation mechanism under real Pacific conditions. The Ocean-3 is the first variant designed for commercial deployment: standardized enough for factory assembly, robust enough for extended deepwater operation, and densely packaged enough to run real AI workloads per unit. At roughly six stories of total height, mostly below the waterline, each deployed node is a self-contained power plant and data center in one structure. The company says it plans to release Apache 2.0 node specifications after the Pacific pilot completes, though it has set no firm date for that disclosure. A full assembly stands roughly six stories tall, with most of the structure submerged, keeping the visible profile low enough to avoid maritime navigation hazards in shipping lanes.

Why 40 Million Came In Before a Single Commercial Node Is Earning Revenue

Panthalassa closed a $140 million Series B in mid-2026, led by Peter Thiel, with other technology investors participating in a round that Fortune reported valued the company near $1 billion at that point. Industry press now cites a figure closer to $2 billion following further financing in the second half of the year. Total capital raised in 2026 is reported to exceed $300 million across all instruments, though the precise sum has not been disclosed publicly. The company has not published revenue figures, a commercial customer count, or a deployed-node count. Every dollar raised to date is a bet on a future that has not yet demonstrated commercial viability.

The investor logic becomes straightforward once you understand the land-side constraint. New grid connections for large-load customers in the most congested US regions now carry queue times of five to seven years. Nuclear recommissioning projects, the fastest credible new power supply, will not deliver usable new capacity before 2028 at the earliest. Permitting for onshore data centers has become entangled in ballot measures, utility rate cases, environmental review, and community opposition at a pace that slows construction timelines in ways that were not true in 2022. For investors who believe AI compute demand will continue compounding at 40 to 60 percent annually, an asset that sidesteps the land-side bottleneck permanently is worth paying a valuation premium to own before its technical proof-points arrive. What Thiel and his co-investors are funding is essentially an option on a new compute geography.

The Grid and Water Crisis That Made This Pitch Land

The land-based bottleneck is not theoretical. PJM Interconnection, the largest power grid region in the United States, recently drew federal watchdog scrutiny for a 76 percent electricity price spike driven in part by AI data center load additions, according to Tom's Hardware coverage of the regulator's report. The International Energy Agency documented that global data center electricity demand jumped 17 percent in 2025 and projects AI server power density to roughly quadruple again by 2027, testing supply chains for power electronics and thermal management equipment simultaneously.

November 2026's midterm elections made the political limits concrete. Voters across roughly 30 counties in eight states backed ballot measures limiting or blocking new data center approvals. Democrats took control of Georgia's utility commission, shifting the approval calculus in a state that holds a large share of the nation's planned AI data center pipeline. Texas elected a governor who pledged an approval freeze on new large-load customers. These outcomes arrived in the same week that OPB's first-look profile of Panthalassa ran, and not by coincidence: political appetite for land-based data center expansion contracted at the moment AI industry demand for new megawatts accelerated.

Freshwater compounds the problem. Evaporative cooling systems in conventional data centers draw millions of gallons from local water supplies annually. In drought-stressed western US watersheds, water already competes directly between agriculture, municipal use, and new industrial load. Panthalassa's ocean-cooled nodes avoid that conflict entirely, eliminating freshwater draw as a permitting issue in every jurisdiction where land-based competitors face it. CEO Sheldon-Coulson has stated publicly: "There are three sources of energy on the planet with tens of terawatts of new capacity potential: solar, nuclear, and the open ocean." He is positioning the ocean not as a niche alternative but as the only remaining unconstrained energy frontier at the scale AI demands.

The Case Against Betting AI Infrastructure on Ocean Waves

The bear case, however, is straightforward and the ocean compute industry has not yet answered it. Panthalassa has not shown that wave-powered floating compute can deliver AI inference at a cost per token that competes with land-based hyperscale. Wave energy is variable. The Pacific is not a uniform resource: storms, calm periods, and geographic variation in swell height all affect power output from individual nodes. A node that cannot guarantee uptime service-level agreements will not win enterprise contracts away from a hyperscaler offering five-nines availability from a grid-connected campus with diesel backup and redundant fiber.

Skeptics point out that every earlier wave energy company that reached the factory stage has struggled to cross the gap from demonstration to commercial scale. The economics of steel fabrication, deepwater operations, maintenance at sea, and satellite connectivity stack costs in ways that onshore construction does not face. Panthalassa has disclosed no power output figure per node, no compute density specification, and no cost-per-kilowatt-hour target. Without those numbers, the $2 billion valuation rests on a projected market size, a credible CEO, and Thiel's pattern of funding infrastructure contrarianism early. That is a thesis, not a proven business.

The latency constraint is structural, not solvable through software updates. The fastest-growing category of AI spend in 2026 is real-time inference: coding assistants, voice agents, enterprise search, agentic pipelines. Those use cases require sub-100-millisecond round trips that LEO satellite relays cannot reliably deliver. Panthalassa's ocean compute architecture is structurally excluded from the most latency-sensitive and highest-margin tier of AI demand, limiting it to training and batch inference, two segments that are also the most price-competitive. Building a defensible business in those segments against Google, Amazon, and Microsoft requires a cost advantage that the company has not yet demonstrated exists.

The Deployment Milestones That Will Determine Whether This Works

The Pacific pilot is the single most important indicator to watch in the next 12 months. If Panthalassa can demonstrate continuous operation through a full Pacific winter, it will have answered the most credible objection to the technology: storm survivability and power consistency in adverse conditions. A pilot that completes without node loss or extended downtime removes the existential technical risk from the investment thesis and opens the path to the first commercial customer commitments, probably with enterprises running training workloads or bulk inference that can be scheduled around satellite windows. A pilot that loses nodes, requires emergency recovery, or goes dark for weeks of storm weather resets the company's timeline by at least two years and raises questions about whether the design is ready for the manufacturing ramp the Vancouver factory implies.

Watch also for a data release on power output and inference throughput per node. When those numbers appear, compare them against the cost structure of the two most direct competitors for the same workload segment: Google's Jupiter network-connected TPU pods (roughly $1.20 per TPU-hour at current cloud pricing) and Amazon's Trainium2 instances at comparable rates. If Panthalassa can reach comparable or lower effective compute cost when amortizing node capital over a ten-year service life, the ocean-compute thesis becomes commercially viable. If the numbers show a two-to-three-times premium, the company needs a differentiation story beyond power sourcing and freshwater avoidance.

The regulatory surface deserves monitoring. Ocean deployments in international waters avoid domestic land-use permitting, but they will eventually intersect with maritime environmental review under UNCLOS and MARPOL frameworks, and with spectrum licensing for satellite uplinks. No ocean compute company has navigated that surface at commercial scale. Data Center Dynamics has tracked at least a dozen companies pursuing different ocean energy and offshore compute approaches globally. Oregon State University operates an $80 million wave energy research hub near Newport that could incubate a competing entrant with academic infrastructure behind it. Meanwhile, land-side alternatives are not standing still: Microsoft and Google have committed to nuclear restart capacity projected to deliver 2.25 GW before 2028, the xAI Colossus Memphis facility is running converted jet-engine turbines for immediate power, and the SpaceX Arizona campus proposal, as covered by Tesla North, would add up to 6 GW of AI compute capacity in a single location if regulators and utilities cooperate. Panthalassa's window to establish itself before land-side alternatives close the gap is probably three to four years. The Vancouver factory is the opening move. The Pacific pilot, beginning later this year, is the first real answer.

The ocean has three times the AI energy potential of every nuclear plant humanity has ever built. The question is whether anyone can collect it cheaply enough to matter.

---