Despite Obvious Challenge

The latest round of underwater data centers is taking shape. Highlander Digital Technology has a 24-megawatt facility running off the coast of China, powered by an offshore wind farm. A Portland startup called Panthalassa has Peter Thiel money and a plan to put autonomous server pods in international waters connected via Starlink. Capital Clean Energy Carriers signed an MoU with Samsung and Lloyd's Register earlier this month on a data-center-on-ship deployment. Nautilus actually shipped its Marseille facility, and the Stockton deployment is still up at six and a half megawatts. The Register wrote it up this week under the headline that did most of the editorial work for the show: datacenters dip a toe back into waterborne computing despite obvious challenges.

The word that matters in that headline is "back." Microsoft ran Project Natick from 2015 to 2020. Eight hundred sixty-four servers in a sealed steel cylinder off the Orkney Islands, two years on the seabed, a failure rate one-eighth what the same hardware showed in a land-based control cluster. Peer-reviewed results. Successful experiment by Microsoft's own published account. And then in 2024, Microsoft quietly confirmed there were no longer any underwater data centers in operation, never fully explained why, and built thirty conventional land-based data centers in the same window. The decision is the artifact. The next data centers the world's most invested operator built after running the most rigorous experiment on the question were all conventional. The current cycle isn't a return. The previous cycle never stopped — it ran the experiment, measured the result, and made a deployment decision. The current operators are running the same experiment.

The editorial center is the gap between two things both being true. The technology works — Microsoft proved that, nobody has to re-prove it. The deployment economics don't — Microsoft proved that too, by what they built next. The gap closes when both sides converge. The technology side has been ready since 2020. The economics side hasn't moved. The version of waterborne that actually has a chance, the panel lands on, is the pod-as-tenant model where the marine operating costs are amortized against a different revenue stream — a wind farm operator with the boats and divers and grid interconnect already, a shipping company with vessel operations already running, somebody else owning the marine layer and the data operator being a customer on the platform. Highlander, collocated with an offshore wind farm, is that model. Panthalassa, with autonomous wave-powered pods four days' steam from the nearest human, is the moonshot version that runs the same experiment again with the same results pending.

Source Article

Datacenters dip a toe back into waterborne computing despite obvious challenges — The Register, June 23, 2026.

Panel

• The Legacy Sysadmin
• The DBA
• The Startup Founder
• The Goat Farmer's Counsel