Ocean Compute — Subsea GPU Infrastructure
OCEANCOMPUTE
OCEANCOMPUTE

The bottleneck isn't chips. It's power.

Ocean Compute builds flotillas — floating structures carrying their own solar power and storage, with sealed GPU pods hanging in cold water underneath. Everything a datacenter needs, built as one machine and towed out to sea.

On land, cooling is a tax: chillers, fans, cooling towers, and the fresh water they evaporate. The sea is cold, dark and always moving. It takes the heat away and never sends a bill.

5 yr
Waiting for a grid connection

Median time from interconnection request to operation, US projects built in 2023. Berkeley Lab, Queued Up.

0.7%
Servers lost underwater

Microsoft sealed 855 servers on the seabed for two years and lost six. An identical group on land lost 8 of 135.

0 L
Fresh water

No cooling towers, so nothing evaporates.

Fig. 01 — The pod

Everything on land is waiting for something. A flotilla is just built.

Pod — sealed GPU container Drag to turn
What a flotilla is
An array on top

Solar panels and batteries on a floating deck. Power is made where it's used, so there's nothing to request and no queue to join.

A pod underneath

A sealed container of NVIDIA GPUs, closed on land under an inert atmosphere and lowered into cold water. It isn't opened again until the hardware is due for replacement.

Only data goes ashore

Data goes to land; power doesn't come from it. That's the whole difference.

01

We build it in a yard

Deck, solar, batteries and pod are assembled on land, where building things is cheap.

02

We seal the pod

GPUs go in, the air comes out, and it closes. Nothing opens it at sea.

03

We tow it out and moor it

The flotilla is anchored, the pod is lowered, and the sea starts taking the heat.

04

We build the next one

The second flotilla is the same drawing. Capacity grows by repeating a build, not by finding another site.

Land datacenters are built once, in place, and never again the same way. A flotilla is built the same way every time. Get the first one right and the rest is a production line.

Where we are

Now

Design and engineering

Pod architecture, thermal modeling, and the structural design of the flotilla.

Next

Dry-run prototype

One sealed rack run on land against the thermal model.

Then

First submersion

One pod under a small moored rig near shore. Real workloads, and our first measured numbers.

After

The first flotilla

Solar, storage and pods as one machine, with capacity available to AI teams.