On-demand, the B200 rents for $6.77/hr and the H200 for $4.17/hr — the H200 is ~62% cheaper per hour, averaged across cloud providers.
| B200 | H200 | |
|---|---|---|
| On-demand /hr | $6.77 | $4.17 |
| Spot /hr | $3.51 | $2.81 |
| Reserved /hr | $4.67 | $3.42 |
| Per month (on-demand) | $4,942 | $3,044 |
| B200 | H200 | |
|---|---|---|
| Vendor | NVIDIA | NVIDIA |
| Architecture | Blackwell | Hopper |
| Memory | 192 GB HBM3e | 141 GB HBM3e |
| Memory bandwidth | 8 TB/s | 4.8 TB/s |
| Launched | 2025 | 2024 |
On-demand, the B200 averages $6.77/hr vs the H200 at $4.17/hr — the H200 is about 62% cheaper per hour. Spot rates are $3.51/hr vs $2.81/hr.
B200 costs more per hour but offers 192 GB HBM3e memory and newer architecture — for large-model training the higher hourly rate often yields a lower cost-per-job. For memory-bound inference or budget runs, the H200 is usually the better value.
At on-demand rates that's about $4,942/mo for the B200 and $3,044/mo for the H200 (730 hours).
See full pricing & history: B200 price · H200 price · all GPUs.
Cross-provider averages from public cloud GPU listings. Informational only.