What you get for the price
The B200 roughly doubles H100 compute in BF16 and FP8, adds FP4 for cheaper inference and has 192 GB of HBM3e at 8 TB/s. For large training runs it shortens time to result; for inference with FP4 it can lower cost per token below Hopper GPUs.
How to get a quote
Send the number of GPUs, term, preferred region and start date through the capacity request form. We reply with available clusters and a fixed monthly price.
B200 on-demand prices elsewhere
| Provider | B200 on demand, per GPU-hour |
|---|---|
| Strata GPU (dedicated) | On request |
| RunPod Community Cloud | $5.98 |
| RunPod Secure Cloud | $7.99 |
| Lambda (8-GPU instance) | $6.69 |
| CoreWeave (8-GPU node) | $8.60 |
List prices from each provider's pricing page, checked 2026-10-23. 8-GPU node prices divided by 8.

