Prefab containerized data center co-located BTM at hydroelectric power station, mountain Georgia. Single 1 MW facility with 288 NVIDIA Rubin Ultra GPUs, $45M CAPEX, $1M annual OPEX, $0 electricity.
⛳
Primary Facility: HPP Co-located DC
Mountain Georgia · Elevation >1,000m · Behind-the-Meter at Hydroelectric Power Station
Prefab modular containerized data center, seismic-rated. 4 GPU racks housing 288 NVIDIA Rubin Ultra GPUs with InfiniBand NDR800 networking, 1.5 PB NVMe storage, and Direct Liquid Cooling powered by cold mountain river water (4–12°C year-round). Greenfield new-build, commissioning Q3 2028.
1 MW
BTM Power
288
GPUs
<1.11
PUE Target
$0
Electricity Cost
$1.0M
Annual OPEX
$45M
Total CAPEX
Infrastructure Stack
Compute
4× GR200 NVL72 racks @ 155 kW each
288 Rubin Ultra GPUs + 144 Grace CPUs
Peak: ~10.4 ExaFLOPS (FP8)
Network
InfiniBand NDR800 fat-tree topology
Dedicated fiber: $500K build + Starlink backup
Storage
All-NVMe · ~1.5 PB usable
Cooling
DLC + mountain river water
Winter: 4–6°C · Summer: 10–12°C
Passive free-cooling, gravity-fed intake
Power
1 MW BTM from HPP · No grid dependency
UPS: 200 kW critical load + GPU graceful shutdown
No transformer CAPEX required
CAPEX Breakdown — $45M Total
Compute (4× NVL72)
$26.5M
Networking (IB)
$4.8M
Storage (NVMe)
$2.1M
Cooling (DLC)
$1.6M
Civil / Prefab
$1.6M
Power Infra
$1.06M
Integration & Launch
$1.0M
Connectivity
$500K
Software & Licensing
$500K
Facility-specific infra (civil + fiber + power + cooling): $4.76M · Compute + IT: $33.9M
Annual OPEX — $1.0M / Year 1
$1.0M
per year
NVIDIA Hardware Maintenance
$400K
Insurance + Connectivity + Visits
$185K
Software Licenses
$150K
Remote NOC + Managed Ops
$120K
Fiber + Starlink Connectivity
~$60K
Electricity (BTM)
$0
Land & Facility Access — JV Structure (No Lease)
⛽
HPP Owner
Contributes: Land access, BTM power, electrical interconnection
→
⚖
JV Entity
Separate legal entity per site. Georgian jurisdiction. Ring-fenced risk.