Preface:

The data center industry has traditionally focused on scale, but AI has disrupted this rhythm, with large model training and inference steepening the power consumption curve.

Competition in AI data centers no longer hinges solely on securing GPUs but also on accessing stable, affordable, low-carbon, and rapidly deployable power sources.

Author | Fang Wensan

Image Source | Internet

Satisfying the insatiable demand for computing power offshore

Last month, Panthalassa announced the completion of a $140 million Series B funding round led by Peter Thiel. The funds will be used for pilot manufacturing facilities near Portland, Oregon, and to accelerate the deployment of Ocean-3 offshore nodes.

The company plans to deploy Ocean-3 pilot nodes in the North Pacific by 2026 to validate offshore AI inference capabilities and prepare manufacturing processes for commercial deployment in 2027.

Panthalassa is targeting this niche: since offshore waves hold untapped energy potential and seawater naturally offers cooling advantages, why not relocate AI nodes offshore to perform computations directly where energy is generated?

It proposes a novel infrastructure combination—wave power generators, offshore floats, sealed server modules, autonomous navigation systems, satellite communication links, and distributed AI inference tasks—all compressed into a manufacturable, deployable, and networkable offshore node.

Traditional offshore renewable energy projects typically focus on transmitting electricity back to land. Panthalassa's technological logic bypasses this step; it does not intend to transmit large-scale electricity back to the grid but instead consumes it on-site offshore.

Electricity is no longer the final product—inference results are.

Panthalassa's truly radical move is not placing nodes offshore but reframing the power transmission problem as a data transmission problem.

This reframing is crucial. Power transmission is a capital-intensive system constrained by cables, grid connections, losses, approvals, and power market mechanisms.

Offshore nodes do not need to serve every latency-sensitive interaction scenario or handle all peak loads. They can start by targeting price-sensitive, latency-tolerant, batch-processable inference tasks as commercial entry points.

This defines Panthalassa's commercial positioning: it resembles an "off-grid AI inference power plant" more than a traditional hyperscale cloud data center.

It also explains why Panthalassa chose AI inference as the early validation direction for Ocean-3 rather than directly targeting large-scale training clusters.

Ocean-3 transmits only inference results, not electricity, back to shore

The Ocean series represents autonomous floating computing nodes. The latest generation, Ocean-3, features a spherical module approximately 50 meters in diameter atop a 60-70 meter steel tubular tower, totaling nearly 85 meters. Most of the structure remains submerged, with only the top sphere visible.

It employs an overtopping wave energy conversion approach, akin to a floating micro-hydroelectric plant.

Waves drive the node's vertical motion, drawing seawater through a bottom neck into a top pressurized chamber, creating stable water pressure to drive turbines for electricity generation.

A single Ocean-3 can reach a peak power output of 1 megawatt with a theoretical capacity factor of around 90%, significantly higher than onshore solar and wind power.

Seawater cooling further amplifies computing efficiency. Land-based data centers consume substantial electricity for cooling, with a typical PUE (Power Usage Effectiveness) of 1.2 to 1.5.

Ocean-3 leverages cold deep-sea water to dissipate server heat, achieving a theoretical PUE close to 1.0 and unlocking more usable computing power from the same electricity generation.

The most groundbreaking feature is its anchorless design. It does not rely on seabed cables, submarine cables, or pipelines. Instead, its hydrodynamic shape enables autonomous movement at a cruising speed of approximately 50 kilometers per day, allowing it to navigate toward high-wave-energy areas or avoid typhoons and massive waves.

Deployment requires only tugboat placement, minimizing construction costs, ecological disruption, and maritime approval pressures.

The internal design adheres to minimalist engineering principles: the power generation system's core moving part is the turbine, and the main structure uses mature steel materials for mass production.

The spherical module houses AI server clusters, paired with immersion cooling and external seawater cooling. Communication relies on low-Earth-orbit satellite constellations like Starlink for unmanned operation and remote maintenance, with maintenance vessels dispatched only for faults.

Seawater, waves, and sealed modules: Offshore nodes transform natural environments into system components

Placing servers offshore adds complexity but also grants access to conditions unattainable for land-based data centers.

As AI chip power density rises, cooling has become a central constraint in data center design. Land facilities rely on air cooling, liquid cooling, cooling towers, chilled water systems, immersion cooling, or hybrid solutions to manage temperatures.

Each cooling upgrade entails higher engineering complexity and stricter operational requirements. The offshore environment provides a natural cooling medium: seawater.

For offshore AI nodes, seawater acts as a heat sink, and waves serve as a power station.

Panthalassa integrates wave power generation, surface floats, autonomous systems, and computing loads into a single node—an autonomous offshore energy-computing unit. This system design carries three layers of technological significance.

① Wave energy serves as the primary power source. Derived from wind energy, waves exhibit greater temporal continuity than ordinary intermittent renewables. Especially in high-energy offshore zones, wave resources are more stable and do not compete with land for space.

② Seawater reduces cooling system burdens. Its high thermal capacity efficiently dissipates chip waste heat, while sealed modules mitigate humidity, oxygen, and dust impacts on server lifespan. The external environment actively participates in cooling, theoretically lowering non-IT energy consumption.

③ Autonomous systems are essential. Remote nodes cannot rely on frequent human inspections. The "unmanned" challenge for offshore data centers far exceeds land-based automation. Land-based facilities can dispatch engineers for repairs; offshore nodes involve ships, weather windows, maintenance costs, and asset recovery risks.

This represents Panthalassa's most formidable technical barrier: integrating marine engineering, renewable energy, AI computing, satellite communications, and unmanned systems into an industrial product exposed to high-salt, high-humidity, and strong-wave environments for extended periods.

Its differentiated value lies in forming a new commercial closed loop by combining off-grid energy, natural cooling, and schedulable inference tasks.

Selling computing power, not electricity: A reverse commercial logic

Humanity has studied wave power generation for decades. The global technically exploitable wave energy potential exceeds 50 terawatts, with the International Energy Agency estimating annual output in the tens of thousands of terawatt-hours.

Yet the wave energy industry has failed to scale, primarily due to transmission challenges: wave-rich areas often lie far from onshore electricity demand centers, and the costs and losses of laying submarine cables undermine economic viability.

Panthalassa's solution is to avoid transmission altogether—not selling electricity but computing power.

When computing power becomes the final product, Panthalassa's competitors shift from energy giants like NextEra or EDF to cloud providers like AWS, Azure, and Google Cloud.

From this perspective, Panthalassa is fundamentally an AI computing power company, not a marine energy company.

By anchoring computing power offshore, Panthalassa aims to carve out a blue ocean in the red sea of cloud services.

Commercial viability boundaries: Scenarios and cost calculations for offshore computing power

Offshore computing power is transitioning from concept to a new data center track (Chinese term for "track" or "sector"). By 2025, China's floating data center market will reach RMB 4.78 billion, growing 32.6% year-on-year, significantly outpacing the national data center sector's overall growth.

The global underwater and offshore data center market stands at $1.11 billion, projected to reach $2.78 billion by 2030.

However, most domestic floating data centers currently deploy nearshore with shore-based power, essentially relocating containerized data centers onto offshore platforms to alleviate land constraints.

Panthalassa pursues a different path: fully offshore and energy-self-sufficient.

The appeal of offshore computing power lies in long-term marginal costs. Once operational, nodes occupy no land, consume no freshwater, and face near-zero electricity costs, making them ideal for long-duration, large-scale computing demands.

Yet they are not inherently cheap. Marine engineering, satellite communications, and remote operations raise upfront investments.

Only under full-load operation and lifespans exceeding a decade can unit computing costs fall below those of coastal, high-electricity-price land-based data centers.

Their ideal scenarios are also clear: non-real-time batch AI inference, scientific computing and engineering simulations, and offshore local computing for wind farms and ocean-going vessels.

Large model pre-training, real-time recommendations, cloud gaming, high-frequency trading, and data businesses with strong compliance requirements are ill-suited for offshore nodes.

Thus, offshore computing power fills cost gaps and scenario voids that land-based infrastructure cannot cover. Future computing supply will evolve into a three-dimensional network spanning land, ocean, and space.

Conclusion:

Computing power will ultimately flow to locations with the lowest combined energy density, cooling efficiency, and regulatory friction. Land will not lose its central role, but unconventional scenarios like oceans, deserts, polar regions, underground spaces, and space will undergo continuous reevaluation.

As AI scales toward generalized inference, computing infrastructure will shift from "centralized data centers" to "multi-modal energy-computing networks."

Partial references: FORTUNE: "Peter Thiel Bets on a 'Subsea Data Center'," Wall Street See: "Peter Thiel Leads Investment in U.S. Startup Exploring 'Ocean Data Centers'," Forbes: "Elon Musk Wants to Go to Space, Peter Thiel 'Throws' Data Centers into the Sea," Ars Technica: "Silicon Valley Pours $200 Million into Building Floating Offshore AI Data Centers."