DRIFT
Compute Galleon

AI capacity.
Fuelled and cooled
by the Ocean

Self-powered, ocean-cooled compute vessels, built in series. A data centre that makes its own power from the wind, cools itself in the sea, and needs no land, no permit and no grid connection.

Reserve Capacity →

A DRIFT catamaran under wingsails leading a fleet of larger wingsail vessels across open ocean NVIDIA Inception Program
219GW
global data-centre demand by 2030, up from 82 GW in 2025
Forecast · McKinsey
850MW
DRIFT deployment from early production run
Target · linked to funding
62GW
of location independent AI demand by 2030
Forecast
The problem

Every frontier model, every token, is constrained by the same thing. Power.

AI and cloud demand is set to grow 2.7× by 2030. All of it waiting for land, grid connections, water and permits. Years of delay and billions of cost.

  1. The grid queue
    4.4 yrs

    To power, on the global average.

    Cushman & Wakefield
  2. Long-lead kit
    ~200 wks

    For a large power transformer.

    IEA
  3. The race for land
    51%

    Yearly increase in cost of powered land.

    Cushman & Wakefield
  4. Public friction
    71%

    Of Americans opposed to an AI data centre in their area, and local moratoria.

    Gallup
The way out

Sail around the constrains. Beyond the horizon lies a vast and free untapped resource. No plan consents. No grid bottlenecks. Near unlimited scale.

The solution

The ocean is compute’s next frontier.

62 GW of AI training demand by 2030 can tolerate distance from its users. That is the demand that can go to sea.

  • Power without a queue — wind, waves and sun at sea
  • Cooling without freshwater — the ocean is the heat sink
  • Space without the intrusion – no land consents, no social backlash
Frontier models meet frontier power.

Every one of these premises has been proven separately. Nobody has yet combined them in a self-powered fleet that moves between regions and is built in series. That gap is what the Compute Galleon is built to fill.

A wingsail catamaran seen from above, alone on open blue ocean
The vessel

A data centre that makes its own free power.

Built on DRIFT’s energy-harvesting platform, with racks in the payload bay. The energy is spent onboard instead of being stored and distributed.

Render of a DRIFT vessel with a compute payload under way
PLAY

Concept render · DRIFT vessel with a compute payload

  • Wingsails capture the wind
  • Water turbines under the hull regenerate megawatts of power while under sail
  • An ocean of cooling for the racks, with no freshwater drawn
  • Redundant communication links, with two ways out from every vessel
Design targets Target

20 MW of firm IT load per full-size vessel, at a PUE of 1.1. That compares with an industry average of 1.54, and it uses zero freshwater.

The pillars

Four pillars of the Compute Galleon, and how we have validated them.

Three are the same on every DRIFT ship. The fourth is what the energy becomes: on a Compute Galleon, it powers tokens.

Core modules

Find the wind, capture it, turn it into energy. The same on every DRIFT ship.

  1. Map of the North Atlantic overlaid with several thousand candidate vessel tracks fanning out from a dense core off the Irish coast
    GOLDILOCKS
    developed with Faculty

    Several moves ahead, not one.

    Faculty on the routing model →
  2. CAD profile of a DRIFT vessel: two tall aerofoil rigs over a long slender hull, with the foil and the shaft-mounted turbine below the waterline
    The rig
    validated with Cape Horn

    Large sail area, small platform.

    The CFD the rigs are modelled in →
  3. Computational fluid dynamics render of a three-bladed DRIFT hydro-turbine on its shaft, showing the pressure field around the blades
    The turbine
    validated with RISE

    Open flow, on a moving hull.

    The turbine collaboration →
Interchangeable module

What the energy becomes. Here, compute.

  1. A DRIFT catamaran under four wingsails at sea, solar panels across its deck and the DRIFT logo on the hull
    The payload
    design target

    Racks in the bay, cooled by the sea.

    How a fleet becomes one data centre →
Diagram of wingsail vessels linked mast to mast in a mesh, with a hub vessel connected to a satellite
The fleet mesh · schematic
The fleet

A virtual compute cluster, on their own network.

Vessels are meshed into a single virtual data centre. Customers deal with one fleet, not a scatter of ships.

  • Microwave mesh network – mast to mast at 80-100km.
  • A hub vessel carries the fleet’s satellite gateway
  • Elastic: vessels join or move between fleets
One contract Target

100+ MW of firm IT per fleet, under one contract, one SLA and one interface.

How to get there

From one hull to full-size fleets.

Each stage builds the capability, the data and the customers the next one needs.

  1. MVY Mk 1
    In build Q1 2027
    1.6 MW – 700 GPU’s
    RINA AiP
    1x Commercial Pilot
  2. MVY Fleet
    H2 2028
    3 MW – 1300 GPU’s
    €500m SPV finance signed
    50+ Commercial ships
  3. GIGASHIP
    ~2030
    20 MW - 8,250 GPU’s1
    Concept drafted
    100+ Global Flotilla

Timeline linked to funding milestone, subject to change.

1 20MW supply at PUE 1.05 (19 MW IT), sized at ~8,250 Vera Rubin GPUs (~115 NVL72 racks) using NVIDIA’s illustrative DSX MaxLPS inference model. Data points scaled to regen power.

Reserve capacity →