The NVIDIA H200 (Hopper, 141 GB HBM3e) rents from $2.62/hr, averaged across 50 cloud providers, updated 2026-09-28. On-demand pricing has risen 25% over the past 90 days.
An H100 with ~1.8× the memory and bandwidth — same core compute, but fits larger models and serves long-context inference more efficiently. Hopper · 141 GB HBM3e · 4.8 TB/s memory bandwidth · launched 2024.
▮ on-demand /hr · ▮ spot /hr
| Date | On-demand | Spot |
|---|---|---|
| 2026-09-28 | $4.48 | $2.62 |
| 2026-09-27 | $4.38 | $2.62 |
| 2026-09-25 | $4.42 | $2.99 |
| 2026-09-24 | $4.50 | $3.36 |
| 2026-09-23 | $4.50 | $4.66 |
| 2026-09-22 | $4.46 | $2.53 |
| 2026-09-21 | $4.40 | $2.62 |
| 2026-09-20 | $4.42 | $2.62 |
| 2026-09-19 | $4.44 | $3.29 |
| 2026-09-18 | $4.47 | $2.62 |
| 2026-09-17 | $4.40 | $2.53 |
| 2026-09-16 | $4.47 | $3.79 |
| 2026-09-15 | $4.40 | $3.11 |
| 2026-09-14 | $4.35 | $2.53 |
| GPU | On-demand /hr | Spot /hr | Memory | |
|---|---|---|---|---|
| H200 | $4.48 | $2.62 | 141 GB HBM3e | |
| B200 | $6.49 | $4.78 | 192 GB HBM3e | H200 vs B200 → |
| H100 | $3.39 | $2.00 | 80 GB HBM3 | H200 vs H100 → |
| MI355X | $5.99 | $2.97 | 288 GB HBM3e | H200 vs MI355X → |
| MI325X | $3.06 | $— | 256 GB HBM3e | H200 vs MI325X → |
| MI300X | $4.41 | $1.60 | 192 GB HBM3 | H200 vs MI300X → |
The H200 averages $4.48/hr on-demand, $2.62/hr spot, and $3.83/hr reserved across 50 cloud providers (updated 2026-09-28).
The lowest tier is spot at about $2.62/hr on average; reserved (1–3 year commitment) runs ~$3.83/hr. Compare live per-provider rates on the Bargo GPU price tracker.
At on-demand $4.48/hr, one H200 runs about $3,270/month (730 hours); reserved pricing lowers that to roughly $2,796/month.
On-demand, the H200 is $4.48/hr vs the H100 at $3.39/hr — pricier per hour. See the full H200 vs H100 comparison for specs and cost-per-job context.
Spot ($2.62/hr) is interruptible spare capacity at a discount to on-demand ($4.48/hr); use spot for checkpointed or fault-tolerant jobs and on-demand or reserved for long, uninterrupted runs.
Cross-provider averages from public cloud GPU listings. Informational only.