Silicon Valley investe centenas de milhões de dólares em centros de dados de IA marítimos: geração de energia por ondas + resfriamento, com investimentos de grandes nomes como Peter Thiel

Portland, a startup Panthalassa, completes a $140 million Series B funding round led by Peter Thiel, with follow-on investments from heavyweights like Marc Benioff, Max Levchin, and others. The company plans to deploy floating AI computing nodes powered by wave energy in the North Pacific.
(Background: Elon Musk said space AI data centers are “bound to happen,” while SpaceX’s IPO warns it may not go forward)
(Additional context: First in the US: Maine proposes banning large data centers, as AI’s high energy consumption sparks public outrage)

Table of Contents

Toggle

• Wave-driven computing power
• A decade of prototypes, three generations of testing, commercial start in 2027
• Offshore vs. onshore: grid bottlenecks pushing computing to the edge

A giant steel sphere floating on the sea, with a vertical pipe extending from its bottom, waves crash against it, pushing water into a pressurized reservoir, which then releases to drive turbines, generating electricity fed directly into the AI chips inside the sphere. This is the technical approach Panthalassa has been testing in the Pacific for ten years.

This week, the Oregon-based public interest company announced it completed a $140 million Series B funding round, led by Peter Thiel, with investors including John Doerr, TIME Ventures (Marc Benioff’s firm), SciFi Ventures (Max Levchin), Susquehanna, server manufacturer Super Micro Computer, Sozo Ventures, and others.

Capital currently backing the “offshore AI data center” concept in Silicon Valley has already reached hundreds of millions of dollars.

Wave-driven computing power

The traditional approach is to send electricity from power plants to data centers; Panthalassa’s approach is to move the computing power next to the energy source, then send the results back via satellite. Instead of long-distance electrical transmission, data travels long distances, and the marginal cost of data transfer is much cheaper than cross-sea power lines.

Each “wave-driven node” is a self-sufficient unit: waves generate power, seawater provides cooling, satellites provide communication channels. After transmitting the AI model from land to the node, inference requests are processed at sea, and results are sent back via satellite to clients. The entire process does not rely on land power grids or require building substations or laying new transmission lines onshore.

Cooling is also a key structural advantage of offshore computing. Land data centers need significant electricity and freshwater to keep servers at safe temperatures; ocean environments are naturally cooler, making this aspect almost cost-free in design.

Given that current large-scale data centers use about 30% of their power for cooling, this structural difference could lead to a fundamental cost reorganization.

A decade of prototypes, three generations of testing, commercial start in 2027

Panthalassa is not a startup just starting out. Since its founding in 2016, the team has tested three generations of prototypes at sea, each validating different engineering hypotheses: the stability of wave power generation, durability of nodes in harsh sea conditions, and satellite communication latency and bandwidth limits.

The specific use of the current Series B funds, according to official press releases, is divided into two directions: one is to complete the construction of a pilot factory near Portland for scaled hardware production; the other is to accelerate deployment of the third-generation commercial node, Ocean-3. According to plans, Ocean-3 will be deployed in the North Pacific in 2026 and officially enter commercial operation in 2027.

However, experts also warn that executing AI inference offshore means models must be transmitted to the sea nodes first, then continuously respond to requests from land. Larger models incur higher initial transmission costs; if the number of nodes is large and distributed widely, model synchronization and version management become more complex.

These are not insurmountable problems, but they are additional engineering layers that land-based data centers do not face.

Offshore vs. onshore: grid bottlenecks pushing computing to the edge

Panthalassa’s focus on the ocean is based on the logic that political and infrastructural pressures on land site selection are systematically increasing.

Recently, Maine enacted the first US ban on large data center construction, citing the visible impact of AI training and inference on local power grids and residents’ lives. Similar public discontent is brewing in other US states and parts of Europe, prompting legislative responses.

The investor list itself is a signal. Thiel, Doerr, Levchin, Field—these names in Silicon Valley are not just about funding but also serve as market endorsements for a long-term bet on this technological direction.

Wave energy itself is not a new technology, but coupling it directly with AI inference power, bypassing the grid, is what Panthalassa is truly doing. Whether they can deliver measurable commercial data by 2027 will determine if this route is the next-generation computing infrastructure or just a carefully crafted financing narrative.