Progress

Building in the open.

Every milestone, test and setback, logged as it happens. Once the pilot pod is in the water, its readings will stream here live.

Last updated
Current phase
Phase 0
Bench and dock trial
Milestones complete
3 of 18
Next milestone
Pod sealed
Target: December 2026
First chips underwater
Feb 2027
Dock trial, Miami
Live telemetry

Pod DR-1, from the seabed.

Streams live from Feb 2027
Internal temp
-- °C
Humidity inside
-- %
Power draw
-- kW
Days without fault
--

We'll publish every reading, including the bad days.

Roadmap

Five phases to a megawatt.

  1. Phase 0 · In progress

    Bench and dock trial

    Nov 2026 - Feb 2027 · Miami
    • System architecture locked: seabed pod plus separate ocean-current generators
    • Pilot pressure vessel designed from standard steel pipe and flanges
    • Resource model built on 40 years of NOAA current measurements
    • Workshop, boat and berth secured
    • Federal test-site and research grant applications submitted
    • Pod sealed, nitrogen-filled and leak-tested in the workshop
    • Four GPUs running AI jobs underwater at the dock for 30 days
  2. Phase 1 · Next

    Pilot pod on the seabed

    Mar - Jun 2027 · Off Fort Lauderdale
    • Pod lowered to the seabed in the Gulf Stream
    • Surface buoy with Starlink sends results to shore
    • 90 days of continuous operation, telemetry published
    • Pod recovered and opened for inspection
  3. Phase 2 · Planned

    On ocean-current power

    Jul - Sep 2027
    • Pod connected to a small ocean-current generator
    • Chips run entirely on Gulf Stream power
    • Measured output compared against the NOAA-based forecast
  4. Phase 3 · Planned

    DR-100 demo node

    2028
    • 100 kW-class generator powering around 30 GPUs
    • First paying customers for seabed compute
    • Federal and state permits for a commercial site
  5. Phase 4 · Planned

    First DR-3500 node

    2030
    • Two 1.75 MW-class generators, 1.28 MW average
    • About 288 current-generation AI chips on the seabed
    • Template for a fleet along the Florida coast
Update log

What changed, and when.

  1. Milestone

    Architecture locked: chips on the seabed, power from the current

    We split compute from power. Chips go in a sealed, nitrogen-filled pod resting on the seabed; separate tethered generators fly in the current above it. The pod never moves, which makes it simpler, cheaper and longer-lived.

  2. Engineering

    Pilot vessel designed from off-the-shelf steel

    The DR-1 pod uses standard schedule pipe and flanges rated well beyond our test depth, keeping the first vessel to a few thousand dollars instead of a custom build.

  3. Research

    Resource model built on measured current data

    We rebuilt our output model on measured velocities off Fort Lauderdale (mean about 1.5 m/s at 50 m) and 40 years of NOAA cable data, rather than textbook peak speeds.

  4. Company

    Deeprun founded in Miami

    Adam Todd and Alex Abalionov start Deeprun to put AI compute where the power already is: in the Gulf Stream.