Supply
Operations
March 26, 2026
Procurement architecture, supplier dependencies, and supply chain risk profile for 1for.ai GPU cloud infrastructure
$45M
Total CAPEX
$1.0M
Annual OPEX
~70%
NVIDIA Share
40%
GPU Delay Risk
7.7%
OPEX / Revenue
CAPEX Breakdown — $45M
Primary Supply Sources
| Supplier | Product | Origin | Value |
|---|---|---|---|
| NVIDIA | GR200 NVL72 GPU, InfiniBand NDR800, AI Enterprise SW, Maintenance | USA / Taiwan | ~$31.3M |
| ODM/OEM (TBD) | Server chassis, rack assembly | USA / Taiwan | ~$3.5M |
| DLC Vendor (TBD) | Cooling CDUs, manifolds, per-rack DLC | USA / Germany | ~$1.6M |
| Power Vendor (TBD) | UPS, PDU, switchgear | Germany / France | ~$1.06M |
| Local Contractor | Prefab containers, civil works | Georgia | ~$1.6M |
| Georgian Telecom | Dark fiber build to IX | Georgia | ~$0.5M |
| Starlink | Backup satellite connectivity | USA | Subscription |
Supplier Dependency Profile
NVIDIA
~70%
Critical · Single-source
ODM / OEM
~8%
Moderate · Substitutable
DLC Cooling
~4%
Low · Multiple vendors
Power / Civil / Connectivity
~9%
Low · Commodity markets
Alternative Source Availability
NVIDIA GPU
No Alternative
Sole source at GR200 NVL72 performance tier; AMD MI-series incompatible with NVL72 architecture
InfiniBand
Partial
Ethernet alternatives emerging (Ultra Ethernet Consortium); requires full architecture redesign
ODM / OEM
Fully Substitutable
Dell, HPE, Wistron, Quanta — all NVIDIA-certified
DLC Cooling
Fully Substitutable
Rittal, Asetek, CoolIT, GRC — competitive market, modular
Power / Civil / Fiber
Fully Substitutable
Eaton, ABB, Legrand; Georgian construction market; Starlink backup Day 1
Supply Disruption Risk Matrix
GPU Allocation Delay
40% High
6–18 month lead times; allocation-based distribution favors hyperscalers. Mitigation: early ordering Q1 2027, NVIDIA partnership.
InfiniBand Constraint
25% Moderate
Bundled with GPU procurement; same NVIDIA supply chain. Mitigation: order bundled, phased fat-tree deployment.
Fiber Construction
20% Moderate
Mountain site fiber build 6–12 months, weather-dependent. Mitigation: Starlink backup from Day 1, parallel construction.
DLC Cooling Delay
15% Low
Competitive market with multiple vendors. Mitigation: modular per-rack design, 12-month pre-order window.
TSMC Fab Disruption
10% Low
Earthquake or geopolitical Taiwan Strait scenario. Industry-wide risk; no company-level mitigation possible.
BIS Export Restriction
20% Moderate
Georgia status under US BIS may change. Compliance documentation required; NVIDIA may impose additional due diligence.
NVIDIA GPU Procurement Flow
30%
Deposit at Order
~$9.4M
~$9.4M
→
40%
At Manufacture
~$12.5M
~$12.5M
→
30%
At Delivery
~$9.4M
~$9.4M
→
✓
Deploy
Q3 2028
Q3 2028
GPU + Networking ~$31.3M · 18-month procurement cycle · Series A equity funded · All values [Assumption]
Supplier Selection Criteria — Priority Rank
1
NVIDIA Ecosystem Compatibility
Non-negotiable for GPU, networking, and cooling integration
2
Delivery Timeline Reliability
Must meet Q1–Q2 2028 construction window
3
NVIDIA-Certified ODM Status
Required for GR200 NVL72 server assembly
4
Remote Technical Support
Essential for mountain Georgia location; on-site SLA for critical hardware
5
Total Cost of Ownership (7yr)
Upfront + maintenance + replacement lifecycle cost
6
BIS Export Compliance
End-use documentation for all US-origin equipment
7
Mountain Site Logistics
Heavy equipment transport to remote Georgian hydroelectric site
