No shared tenancy
Your cluster is yours alone. Dedicated hardware with full root access and no noisy neighbors.
No spot interruptions
Reserved capacity means your workloads run uninterrupted, 24/7, for the duration of your contract.
Latest NVIDIA GPUs
H100, H200, B200, and B300 GPUs with NVLink and InfiniBand interconnect.
Flexible scale
From 16-node clusters for focused training to 128+ node deployments for frontier-scale workloads.
Who are dedicated clusters for?
- ML teams running large-scale distributed training jobs
- AI companies that need guaranteed GPU capacity for production inference
- Enterprises with compliance or data residency requirements
- Research labs working on frontier models or large experiments
- Inference Engine users who need a dedicated deployment of a model or workload
- GPU Instance users who have outgrown on-demand and need reserved capacity
How it compares
| On-demand instances | Dedicated clusters | |
|---|---|---|
| Infrastructure | Shared multi-tenant | Dedicated single-tenant |
| GPUs | H100, A100, L40S, RTX 4090 | H100, H200, B200, B300 |
| Networking | Standard | InfiniBand / NVLink |
| Scale | 1–8 GPUs | 16–128+ nodes |
| Commitment | None | 12–24 months |
| Pricing | Hourly on-demand | Fixed monthly rate |
| Setup | 60 seconds | 1–2 weeks |
| Access | Self-serve | Contact sales |
Get started
Contact sales
Tell us about your GPU requirements and we’ll put together a proposal within 24 hours.
- Number of GPUs / nodes needed
- GPU type preference (H100, H200, B200, B300)
- Preferred region or data center location
- Contract duration (12 or 24 months)
- Target start date