Startup Panthalassa has raised a total of $210 million to build ocean-based AI computing nodes powered by wave energy. Giant 85-meter spheres are set to be tested in the Pacific Ocean by end of 2026, cooled by seawater and transmitting AI model outputs via satellite. The project, backed by Palantir co-founder Peter Thiel, is a response to growing challenges in building AI data centers on land.
Key takeaways
- Latest round: $140M — total raised: $210M
- Investors: Peter Thiel (Palantir) and Founders Fund
- Ocean-3 node is 85 meters long; Pacific test planned for 2026
- Wave energy directly powers AI chips; seawater cools the servers
- Data transmitted via satellite — experts cite this as the main bottleneck
How does a floating data center work?
A Panthalassa node looks like a giant steel sphere floating on the water with a vertical tube descending into the depths. Wave motion pushes water up through the tube into a pressurized reservoir, which releases it to spin a turbine generator producing power. The electricity goes directly to AI chips mounted on the unit.
Cooling is the second key advantage. Traditional data centers consume enormous amounts of electricity and fresh water for server cooling. Ocean water temperature is significantly lower than air in typical land-based locations. "Ocean-based compute might offer a massive cooling advantage because the ambient temperature is so low," says Benjamin Lee, an engineer at the University of Pennsylvania.
AI model outputs (inference tokens) are transmitted to customers worldwide via satellite link. Panthalassa has already tested earlier prototypes: Ocean-1 (2021) and Ocean-2 (a three-week sea trial off the coast of Washington state in February 2024). The new Ocean-3 — nearly as tall as London's Big Ben — will be the first test of a full-scale unit.
Challenges: satellites and maintenance in the middle of the ocean
Experts point to several serious limitations. Satellites transmit data at speeds ranging from tens to hundreds of megabits per second — sufficient for responding to individual queries, but insufficient for coordinating multiple nodes.
Frequent communication and coordination between nodes may be challenging. And transferring larger volumes may require physically transporting storage disks to data center nodes by ship — but this should be done only periodically.
Benjamin Lee, University of Pennsylvania engineer
The maintenance question is equally serious. The company is looking for engineers capable of ensuring units "survive for more than a decade in the harshest ocean conditions" without human intervention. Those are demanding engineering requirements for hardware that must operate in an environment where a failure means a week-long journey by service vessel.
Context: why oceans are becoming attractive
Panthalassa's decision is not accidental. The tech industry has planned to spend $765 billion on AI data centers in 2026. At the same time, it is facing growing local community resistance to new land-based investments and construction delays due to power supply constraints and labor shortages.
The idea of ocean data centers is not new. Microsoft experimented with underwater servers in Project Natick (2015, 2018) and confirmed that sealed, seawater-cooled systems achieve lower failure rates than land-based systems. However, Microsoft chose not to commercialize the vision. China went further — deploying underwater data centers near Hainan Island and off the coast of Shanghai.
Panthalassa is more ambitious than its predecessors. Its nodes are designed to operate in the open ocean, be self-propelled and autonomous — not submerged near the coastal seabed. That is a fundamental engineering difference.
Why it matters
Floating AI data centers are not just an engineering concept — they are a symptom of deeper infrastructure pressure. When an industry planning to spend hundreds of billions of dollars encounters barriers in the form of community opposition, land acquisition difficulties, and grid constraints, it begins looking for alternatives radically outside established paths.
Oceans cover 71% of Earth's surface and have no municipal councils blocking building permits. If Panthalassa solves the technical challenges — satellite bottleneck, autonomous maintenance, hardware durability — it could create an entirely new category of AI infrastructure. If not, it will be another warning that Silicon Valley too easily funds technically difficult projects when land-based alternatives are simply too politically complicated.
What's next?
- The Ocean-3 test in the Pacific in 2026 is the critical technical feasibility checkpoint
- Maritime regulations and ocean use rights may become the next barrier — ocean areas have complex jurisdiction
- Success could attract further investors to the "ocean computing" category; failure could close the topic for years
Sources
- Ars Technica — Silicon Valley bets $200M on AI data centers floating in the ocean
- BusinessWire — Panthalassa Raises $140 Million to Power AI at Sea
