Floating infrastructure for AI compute
mareflux
Where compute meets current.
Purpose-built data centers, floating offshore. Seawater-cooled and built for the AI tenants land cannot make room for in time.
The problem
Three bottlenecks. One direction.
$2M+/acre
Coastal land near power infrastructure. Communities block approvals. Even secured sites wait years for a grid connection.
7-year queue
Average grid connection wait in major markets. 2,600 GW of data center projects are queued for connection in the US alone.
5-year permitting
Environmental reviews and zoning hearings, on a timeline that was never built for AI's pace of demand.
$750B
planned global DC spend, 2026
97%
current global DC occupancy
77%
new capacity pre-leased before completion
The ocean covers 70% of Earth’s surface. Free cooling. No permits. No grid queue. No community opposition. The ocean does not say no.
The asset
A purpose-built floating data center barge.
Not a converted vessel — a hull designed from the keel up for one job. Every system on board exists to keep GPUs powered, cooled, and online.
- Capacity
- 5 MW IT load
- Racks
- 100–166 GPU racks
- Cooling
- Seawater — free, zero freshwater
- PUE
- 1.15 (vs 1.5 land average)
- Power
- Low-cost supply, upgrading to nuclear (Phase 2)
- Connectivity
- 2×100Gbps fiber + Starlink
- Latency
- <5ms to nearest IXP
- Classification
- ABS or Lloyd's Register
Why the ocean
Four advantages no land data center can match.
Infinite cooling
Seawater absorbs heat without limit. No cooling towers, no freshwater, no chemical treatment. Nautilus proved this for 5 commercial years.
No permitting
A port mooring agreement is a commercial negotiation, not an environmental review. No community opposition to a facility offshore.
Mobility
Geopolitical risk, or a better market opportunity — the barge can move. No land data center on earth can relocate.
Nuclear ready
Conventional power in Phase 1. A floating SMR moors alongside in Phase 2 — the same model already operating on Russia's Akademik Lomonosov since 2020.
Proof, not theory
Nautilus Data Technologies ran a floating data center for 5 years. Here’s the data.
Cerebras ran its CS-2 — the most powerful AI chip in the world at the time — on the Nautilus barge. This is not a speculative category. It is a proven one, with an operator seat that Nautilus is now vacating as they pivot to licensing their cooling technology.
7 MW
Operational capacity, Port of Stockton
1.15
PUE sustained over 5 years
86%
Occupancy — paying tenants
$0
Freshwater consumed
400K+
Liquid-cooled AI unit-hours
0
Hardware failures from the marine environment
Source: Nautilus Data Technologies, Port of Stockton, CA, 2020–2026 commercial operations.
The roadmap
Three phases. One direction.
2026 – 2029
First Barge
Purpose-built 5 MW barge, seawater-cooled and moored offshore. First revenue, first tenants, first proof.
2029 – 2033
Nuclear Independence
A floating SMR moors alongside the barge. The platform becomes power-independent — able to operate anywhere, no port required.
2033+
Self-Propelled Flotilla
Nuclear-powered, self-propelled vessels operating in international waters — a distributed compute network that moves with demand.
The team
The hardest part isn’t the data center. It’s making it work at sea.
Paranjay Menaria
Founder & CEO
IIM Ahmedabad · Volkswagen Group — Global Technology Leader
- —Platform strategy across North America, Europe, and China at Volkswagen Group (CARIAD)
- —Built 0→X India operations
Adnan Hussain
Co-Founder & COO
15 years Merchant Navy · Senior Officer
- —Vessel procurement, port authority relationships across multiple countries
- —Classification society network — ABS, Lloyd's, Bureau Veritas
Actively recruiting a DC Infrastructure Architect for Phase 1 — power distribution, liquid cooling, high-density GPU deployments.