Roundup· 1 min
What an H100 costs per hour, by provider
Prices checked on each vendor's pricing page in November 2026. Hourly equivalents for per-second and per-minute billing are our own calculation.
AI Tools Hub Editorial · published · updated
H100 hourly rates
| Provider | Form | Per hour |
|---|---|---|
| Together AI | Cluster, preemptible | $1.99 |
| Together AI | Cluster, reserved 3–180 days | $3.19–$3.69 |
| Together AI | Cluster, on demand | $3.99 |
| Modal | Per-second (H100 SXM5) | ~$3.95 |
| Together AI | Dedicated inference | $5.49 |
| Replicate | H100 | $5.49 |
| Baseten | Dedicated deployment | ~$6.50 |
| Fireworks | On demand H100 80GB | $8.00 |
Three bands
Preemptible at $1.99 suits work that can be interrupted: batch embedding, offline evaluation, overnight training. You must build the resume logic yourself.
Reserved suits predictable load and saves roughly 20% against on demand. On demand suits experiments and unknown load, and is where you should start.
Check before comparing rates
- Whether the GPU is actually available
- Whether pinning a region adds a premium — Fireworks charges 1.5x
- Separate storage and egress charges
- Minimum terms and cancellation
FAQ
Can preemptible be used in production?
Not where a user is waiting. It suits batch and training work that can restart.