Colocation & Enterprise Infrastructure
AI Hardware & Data Center Deployment Specifications
Comprehensive deployment requirements including server models, GPU quantities, power consumption, network bandwidth, and installation timeline.
Req #1, #2, #3
Server Specs & Rack Size
- Total Servers: 4 - 6 Nodes (Phase 1: 2 Nodes in Q3 2026, Phase 2: 2-4 Nodes in Q4 2026)
- Inference Node: 4U Server with 4-8x NVIDIA L40S 48GB GPUs
- Training Node: 8U Server with 8x NVIDIA H100 / H200 80GB SXM5
- Rack Space: Total 12U - 16U (Recommend 1/2 Rack or 1 Full 42U Rack)
Req #4
Power Consumption
4U L40S Server (Per Unit)
Typical: 3.8 kW |
Max: 5.5 kW
8U H100 HGX Server (Per Unit)
Typical: 7.5 kW |
Max: 10.2 kW
* Requires A+B Redundant Power Feeds (10 kW - 15 kW per Rack density).
Req #5
Network Architecture
- Public DIA Internet: Dedicated 1 Gbps (Burst up to 10 Gbps)
- IP Addressing: Public IPv4 Block /28 (16 IPs) or /27 (32 IPs)
- Private Interconnect: 100G/400G RoCE v2 / InfiniBand for GPU Cluster
- Target Latency: < 30 ms regional latency across SEA countries
Req #6
Required Managed Services
24/7 Remote Hands
Hardware swaps & power cycles
Smart Lock & CCTV
Cabinet access control
PDU Monitoring
Real-time power & temp API
Req #7
Deployment Timeline (Q3 2026)
Week 1-2
Rack preparation, A+B Power cabling & Network Drop.
Week 3
Server mounting, CUDA / Driver initialization & Staging.
Week 4
Network Stress Test, Latency validation & Go-Live.