Somewhere in the journey from an AI idea to a running workload, somebody needs to provide electricity.
NetworkOcean has taken that requirement in a fairly unusual direction.
The company, founded by Sam Mendel and Eric Kim, is developing floating solar data centres. In its latest public launch, the team says it ran a customer AI workload on an H100 powered by solar on the water, going from paper to deployment in 53 days.
That is the company’s account of an early demonstration. It does not establish the economics of a large data centre.
It does give us a concrete reason to look closer.
The proposed advantage is getting compute and power working together faster. It is an infrastructure bet about what has to happen before the model gets to run at all.
The awkward questions arrive immediately. What happens overnight? How do maintenance and connectivity work? What does the installation cost once you include the whole system? Which water sites actually make sense?
Good. A strange idea should survive specific questions.
We’re interested in whether the team can turn an initial workload into a repeatable installation customers want to use. The useful comparison is the full cost and time required to deliver dependable compute.
We also like that the public story has changed from an earlier underwater data centre concept to the current floating solar approach. The details of the implementation matter more than loyalty to the first pitch.
There is plenty here left to prove.
But a small team putting a real workload on an unusual source of infrastructure is exactly the sort of thing we want to notice while it still looks a little odd.
Budera Worth Watching. The companies we think you’ll wish you found earlier.
