Technical Equipment Requirements
Operations
March 26, 2026
Documents all planned equipment for 1for.ai 1 MW GPU facility: 9 categories, $45M CAPEX, NVIDIA GR200 NVL72 core
288
GPUs (Rubin Ultra)
10.4
ExaFLOPS FP8
909 kW
IT Load
1.10
PUE Target
CAPEX Breakdown by Category
Planned Equipment Stack
GPU Compute
$26.5M
4x NVIDIA GR200 NVL72 racks
288 GPU + 144 Grace CPU
288 GPU + 144 Grace CPU
Order Q1 2027 → Delivery Q2 2028
InfiniBand Network
$4.8M
NDR800, fat-tree topology
Full fabric for 4 racks
Full fabric for 4 racks
Order Q1 2027
NVMe Storage
$2.1M
All-NVMe array
~1.5 PB usable
~1.5 PB usable
Q1–Q2 2028
DLC Cooling
$1.6M
Direct Liquid Cooling
+ mountain river water loop
+ mountain river water loop
Construction Q1–Q2 2028
Power Infrastructure
$1.06M
UPS 200 kW critical load
GPU graceful shutdown, no BESS
GPU graceful shutdown, no BESS
Q1–Q2 2028
Prefab Containers
$1.6M
Seismic-rated, modular
Data center grade
Data center grade
Construction Q1–Q2 2028
Connectivity
$0.5M
Fiber build (primary)
Starlink backup
Starlink backup
Q2 2028
Software & Licensing
$0.5M
CUDA, orchestration, DCIM
Monitoring stack
Monitoring stack
Q2–Q3 2028 (annual)
Integration & Launch
$1.0M
System integration, burn-in
Commissioning & testing
Commissioning & testing
Q3 2028
Acquisition Timeline
Q1 2027
GPU & IB Order
Q3 2027
Series A Close $45M
Q1-Q2 2028
Construction & Delivery
Q3 2028
Integration & Beta
Q4 2028
Commercial Launch
Technical Specifications
288 GPU + 144 CPU
NVIDIA Rubin Ultra GR200 NVL72
10.4 ExaFLOPS
Peak Compute (FP8)
909 kW
IT Load (1 MW / PUE 1.10)
640 kW
GPU Compute Load
NDR800
InfiniBand Fat-Tree
~1.5 PB
NVMe Storage Usable
4–12°C
River Water Temp (Year-Round)
No BESS
Flat Inference Load Profile
