H100/H200 Bare-Metal GPU Rental — Taiwan Sovereign Cloud
From training to inference, TW Compute provides complete AI computing solutions. Supporting flagship GPUs like NVIDIA H100, A100, RTX 4090, helping enterprises quickly deploy AI workloads.
100% Bare Metal Performance
No virtualization loss, 100% GPU performance released directly.
48-Hour Rapid Deployment
From subscription to launch, complete GPU cluster deployment within 48 hours.
Flexible Scaling Architecture
From single GPU to 8-GPU cluster, supporting vertical and horizontal expansion.
Taiwan Data Sovereignty
Data stays in Taiwan, meeting enterprise data sovereignty and compliance needs.
Two billing models
Cloud GPU vs Bare-Metal GPU
Roadmap
TW Compute Cloud GPU
Time-shared, billed by vCPU·GPU·RAM per minute — boots in minutes. Best for elastic spikes and short experiments.
- ·No reservation, spin up on demand
- ·Per-minute billing, off when stopped
- ·Pre-loaded vLLM / TGI / SGLang images
Available now
TW Compute Bare-Metal GPU
Full-node exclusive, 100% performance, delivered to a Taiwan datacenter within 48 hours. Built for long training runs and large inference.
- ·Whole node exclusive, no noisy neighbors
- ·Monthly / yearly / custom project terms
- ·Data stays in Taiwan, zero cross-border
GPU Computing Plans
Choose the GPU configuration that best suits your AI workload
Spec comparison
All GPU plans side by side
Node-level vCPU / RAM / Storage / VRAM specs and per-GPU pricing
| GPU plan | vCPUs | RAM | Storage | VRAM | NT$/hr·GPU | Notes |
|---|---|---|---|---|---|---|
| NVIDIA GB300 NVL72 Super Flagship | 1,920 | 30,720 GB | 57,600 GB NVMe | 13.8TB GDDR7 | 58,333 | NVL72 flagship rack |
| NVIDIA GB200 NVL72 AI Super Host | 1,728 | 23,040 GB | 57,600 GB NVMe | 13.8TB HBM3e | 48,611 | NVL72 flagship rack |
| NVIDIA HGX H200 Super Host | 192 | 2,048 GB | 8,000 GB NVMe | 1.1TB HBM3e | 42,500 | 8× flagship node |
| NVIDIA HGX H100 AI Host | 128 | 2,048 GB | 8,000 GB NVMe | 640GB HBM3 | 35,000 | 8× flagship node |
| NVIDIA RTX PRO 6000 Blackwell Professional Host | 128 | 1,024 GB | 4,000 GB NVMe | 384GB GDDR7 | 15,000 | 8× flagship node |
| NVIDIA A100 GPU Server | 96 | 1,024 GB | 4,000 GB NVMe | 640GB HBM2e | 20,000 | 8× flagship node |
| NVIDIA RTX 5090 Flagship Host | 96 | 768 GB | 2,000 GB NVMe | 256GB GDDR7 | 6,000 | 8× flagship node |
| NVIDIA RTX 4090 Professional Rendering Server | 80 | 768 GB | 2,000 GB NVMe | 192GB GDDR6X | 4,500 | 8× flagship node |
Decision matrix
Bare-Metal Rental vs Colocation
Trade-offs between performance, flexibility, and cost — same row, side-by-side comparison
| Dimension | TW Compute Bare Metal | Public Cloud / Self-built DC |
|---|---|---|
| Pricing basis | Per-GPU / hour all-inclusive (datacenter, power, network) | Itemized rack space + power + network |
| Settlement | Monthly take-or-pay reservation | Reservation, lower entry barrier |
| Power model | Bundled within contract cap, no overload penalties | Passthrough metering |
| Network access | Built-in VPN / private line to customer site | Customer-owned cross-connects |
| Idle billing | GPUs always available, no extra space fee | Rack space billed regardless of utilization |
| Best fit | Stable long-running training workloads | Inference spikes, extending your own datacenter |
Enterprise account features
Account management and billing transparency
Transparent pricing, team-level accounting, audit-ready — standard for the enterprise tier
Sub-account permissions
Each engineer signs in with their own account; quotas and resource caps are controlled by admins.
Job execution logs
Every job submission, GPU config, and usage is recorded — exportable for audit.
Cross-team resource sharing
GPU pools auto-allocate across teams by priority — no idle waste.
Transparent billing, no hidden fees
The contract price is the invoice price. Monthly statements. No traffic, API, or surprise surcharges.
Taiwan datacenter sites
Three ISO 27001-certified datacenters
All training, inference, and storage stay inside Taiwan — zero cross-border transfer
| Site | Region | Capacity | Status | Certifications |
|---|---|---|---|---|
| Taipei — Neihu | North | 12 MW | Operational | ISO 27001 · Tier 3 |
| Taichung — Wuri | Central | 8 MW | Operational | ISO 27001 · Tier 3 |
| Kaohsiung — Qianzhen | South | 6 MW | Operational | ISO 27001 · Tier 3 |
FAQ
Can I rent a GPU server for just one month?
Yes, we offer flexible monthly plans with no long-term contracts. You can terminate or renew anytime based on your needs.
What's the difference between GPU cloud and dedicated servers?
GPU cloud is suitable for flexible needs, billed by usage; dedicated servers are suitable for scenarios requiring 100% performance and exclusive resources.
How long does deployment take?
GPU cloud can be started within minutes; dedicated servers can be deployed within 48 hours.
Ready to Start AI Computing?
Our technical team can provide customized AI computing solutions based on your needs