Architecture
Pre‑launch preview. Projected launch: 2026.
How we tie low-visual renewables to modular, liquid-cooled GPU halls with a resilient data fabric.
Power architecture
Generation blend
Tidal and in-stream river provide high predictability; compact wind and BESS fill gaps. Where viable, DC coupling reduces conversion losses.
Distribution
Medium-voltage feeders into hall PDUs with per-pod metering. Liquid loops recover heat at 35–45°C for waste-heat pilots.
Compute & cooling
Pods
8× GPU pods per rack with liquid-cooled manifolds, 60–80 kW per rack. Short air paths; front-to-rear to keep delta-T tight.
Target efficiency
PUE ≈ 1.20 in cold climates via free air and rear-door HX. We monitor hourly energy mix for matching certificates.
Network & data
Fabric
Redundant leaf-spine (Ethernet or IB). Edge gateways per hall; segmentation for tenant isolation and sovereign controls.
Backhaul
Northern fibre loops with satellite or microwave as tertiary failover. Multi-region replication options for storage/DR.