The race to power artificial intelligence has already led tech companies to restart shuttered nuclear reactors, drill for novel geothermal energy, and seriously consider putting data centres in orbit. Now, backed by $140 million from some of Silicon Valley’s biggest names, a startup called Panthalassa wants to take the servers to sea — and let the ocean do the rest.
The Problem They’re Trying to Solve
Tech companies are on course to spend around $750 billion on data centres in 2026 alone. The bottleneck isn’t money or land — it’s power. Electrical grids around the world are already straining under existing data centre demand, and the pace of AI infrastructure buildout is far outstripping the ability to bring new energy capacity online.
Permits take years. Grid connections take longer. Communities are increasingly pushing back.
The consequence is a scramble for alternatives, and the solutions being seriously considered are getting increasingly unconventional.
Floating Servers, Wave-Powered
Oregon-based Panthalassa’s answer is a nearly 300-foot-long ocean-going node — part data centre, part wave energy generator. A buoyant sphere sits on the surface while a tube-like housing beneath holds the computer servers. As the node bobs with the waves, the motion forces seawater up into a pressurised reservoir, driving a turbine that generates electricity for the chips. The surrounding ocean water simultaneously cools the hardware — eliminating two of the most resource-intensive problems facing land-based facilities at once.
The nodes connect to the internet via SpaceX’s Starlink satellite network, removing the need for any undersea cabling for either power or data. They are self-propelling, able to navigate to their deployment location, and designed to stay on station and operate autonomously for extended periods with minimal moving parts — a necessity when maintenance crews are a long way away.
The crucial insight behind the design, according to the company’s CEO, is that ocean wave energy projects have historically failed because getting electricity back to shore is enormously complex and expensive. Panthalassa sidesteps the problem entirely by using the power exactly where it’s generated.
The concept has already been tested in the real world — a three-week trial of their second-generation node off the Washington state coast in 2024 validated the core technology. Backers for the latest funding round include Palantir co-founder Peter Thiel and Salesforce’s Marc Benioff, with $140 million raised to complete a pilot manufacturing facility near Portland and build out the next-generation Ocean-3 nodes, scheduled for Pacific trials later this year. Commercial deployment is targeted for 2027.
The Challenges Ahead
The ocean is not a forgiving operating environment, and the concept faces real hurdles that its backers are candid about.
The most significant is bandwidth. Starlink satellite links offer far lower data throughput than the optical fibre connections that typically network large data centres. Combined with the potential for signal latency, this could meaningfully limit how well ocean nodes handle the heavy, latency-sensitive workloads at the heart of AI computing. Whether satellite connectivity will prove adequate for serious AI infrastructure — rather than lighter workloads — remains an open question.
Then there’s the practical challenge of deploying and maintaining a fleet of sophisticated hardware in one of the harshest environments on Earth. Storms, salt corrosion, marine growth, and the sheer remoteness of open-ocean operation all present engineering problems that no amount of clever design can entirely eliminate.
A More Grounded Alternative to Space
The idea has an obvious point of comparison in orbital data centres, another concept attracting serious investment, which would use abundant solar energy and the cold of deep space to power and cool their hardware. Panthalassa’s approach shares the same basic logic of escaping land-based constraints by moving computing infrastructure somewhere radically different. But going to sea is orders of magnitude cheaper and simpler than going to orbit, making it a potentially more near-term viable path.
Neither approach is a silver bullet. But both reflect the same underlying reality: the AI industry’s appetite for power has grown large enough that the constraints of conventional land-based infrastructure are starting to genuinely bind, and the solutions being reached for are no longer theoretical. They’re being built.