Why the mismatch happens
Network volumes are storage in one datacenter. On RunPod, for example, you choose the datacenter when you create the volume, GPU options depend on where the volume is, and data does not sync automatically between volumes in different datacenters (checked 2026-11-17). When that datacenter runs out of your GPU type, the volume is stuck there with your weights on it.
How to avoid it
- Pick the datacenter by GPU, not by price. Check where your GPU type is usually free before creating the volume; one with steady stock beats one that is cheaper but often empty.
- Keep a master copy elsewhere. Store weights and datasets in S3-compatible object storage or the Hugging Face Hub, so a new volume in another datacenter can be filled with one command.
- Script the second datacenter. Keep a ready sync script (rsync over SSH, or the provider's S3 API) so moving takes minutes of your time, not an afternoon.
- Watch the costs. Two volumes means paying for storage twice; RunPod lists $0.07/GB per month for the first 1 TB and $0.05/GB above that.
- For steady work, reserve. If you use the same GPUs every day, a reservation in one named datacenter removes the availability gamble.
Volume-based cloud vs reserved dedicated server
| Network volume + on-demand GPU | Strata reserved dedicated server | |
|---|---|---|
| Where data lives | Volume in one datacenter | Local NVMe on your server, in the datacenter named in the contract |
| GPU availability | Whatever is free in that datacenter now | Allocated to you for the whole term |
| Moving regions | Copy data yourself | Choose the region when you reserve |
| Best for | Occasional, bursty use | Daily training or production inference |
Strata list prices are the same in every region; availability differs. Prices as of 2026-11-17.
Strata regions
Strata GPU sources H100, H200, B200 and RTX PRO 6000 capacity in Europe (Frankfurt, Amsterdam, Stockholm, London), North America (Ashburn, Dallas), the Middle East (Dubai, Riyadh) and Asia-Pacific (Mumbai, Singapore, Tokyo). Dedicated nodes are reserved for 3–36 months at 3–14% below on demand; an 8×H100 node is $18,863 a month at the on-demand rate. Clusters are usually ready in a few days to two weeks.


