For teams fine-tuning and running research on shared fabric.
- NVIDIA H200, 141 GB
- 3.2 Tb/s InfiniBand, shared spine
- Slurm or Kubernetes, managed
- Live in about 3 days
- Email support, 1 working day
Power, cooling, fabric and checkpoint storage are in the rate. Longer terms lower it.
For teams fine-tuning and running research on shared fabric.
For labs training frontier models on their own fabric.
For governments and regulated industry that need a hall of their own.
All plans run in Hall A, Narvik, on the same hydro power. The differences are fabric, control and support.
| Feature | Pod | Dedicated | Sovereign |
|---|---|---|---|
| Compute | |||
| GPU | H200 | H200 · B300 | H200 · B300 |
| Cluster size | 64 – 512 | 512 – 4,096 | 4,096+ |
| Fabric | Shared spine | Dedicated | Dedicated, isolated |
| Bare metal | |||
| Data | |||
| Checkpoint storage | 200 TB | 2 PB | Custom |
| Customer-held keys | |||
| Air-gapped option | |||
| Support | |||
| Response time | 1 day | 15 min | 15 min |
| Named engineer | |||
| Uptime SLA | 99.9% | 99.99% | 99.995% |
Power, cooling, rack space, InfiniBand fabric, checkpoint storage and 24/7 hands on site. Egress inside the EEA is free.
Yes. Move from on-demand to a 1 or 3 year term at any time; the lower rate applies from the next billing month.
We invoice monthly in EUR or USD. Rates are fixed for the whole term, independent of the NO4 power price.
No setup fee on Pod or Dedicated. Sovereign builds with a private cage carry a one-off fit-out cost, agreed up front.