The NVIDIA H100 (Hopper, 80 GB HBM3) rents from $1.71/hr, averaged across 12 cloud providers, updated 2026-08-14. On-demand pricing has risen 8% over the past 70 days.
The workhorse of the current AI build-out — strong FP16/FP8 training and inference, and the most widely available data-center GPU. Hopper · 80 GB HBM3 · 3.35 TB/s memory bandwidth · launched 2022.
▮ on-demand /hr · ▮ spot /hr
| Date | On-demand | Spot |
|---|---|---|
| 2026-08-14 | $4.00 | $1.71 |
| 2026-08-13 | $3.89 | $1.66 |
| 2026-08-12 | $4.13 | $1.77 |
| 2026-08-11 | $4.10 | $1.69 |
| 2026-08-10 | $3.95 | $1.61 |
| 2026-08-09 | $4.02 | $1.67 |
| 2026-08-08 | $4.05 | $1.68 |
| 2026-08-07 | $4.09 | $1.74 |
| 2026-08-05 | $4.07 | $1.69 |
| 2026-08-04 | $4.01 | $1.72 |
| 2026-08-03 | $4.02 | $1.86 |
| 2026-08-02 | $3.96 | $1.82 |
| 2026-08-01 | $3.95 | $1.73 |
| 2026-07-31 | $4.02 | $1.92 |
| GPU | On-demand /hr | Spot /hr | Memory | |
|---|---|---|---|---|
| H100 | $4.00 | $1.71 | 80 GB HBM3 | |
| B200 | $6.84 | $3.74 | 192 GB HBM3e | H100 vs B200 → |
| H200 | $4.37 | $2.98 | 141 GB HBM3e | H100 vs H200 → |
| MI355X | $8.60 | $5.29 | 288 GB HBM3e | H100 vs MI355X → |
| MI325X | $3.18 | $2.00 | 256 GB HBM3e | H100 vs MI325X → |
| MI300X | $3.56 | $1.65 | 192 GB HBM3 | H100 vs MI300X → |
The H100 averages $4.00/hr on-demand, $1.71/hr spot, and $3.49/hr reserved across 12 cloud providers (updated 2026-08-14).
The lowest tier is spot at about $1.71/hr on average; reserved (1–3 year commitment) runs ~$3.49/hr. Compare live per-provider rates on the Bargo GPU price tracker.
At on-demand $4.00/hr, one H100 runs about $2,920/month (730 hours); reserved pricing lowers that to roughly $2,548/month.
On-demand, the H100 is $4.00/hr vs the B200 at $6.84/hr — cheaper per hour. See the full H100 vs B200 comparison for specs and cost-per-job context.
Spot ($1.71/hr) is interruptible spare capacity at a discount to on-demand ($4.00/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.