NVIDIA H200 Rental Price per Hour (2026)
VRAM 141 GB. Prices collected 2026-10-23 from public platform pages and APIs. Checkout price prevails.
Lowest listed price: $3.59/hour at RunPod. Collected 2026-10-23.
International platforms (USD)
| Platform | Price per GPU hour | Collected | Note | Source |
|---|---|---|---|---|
| RunPod | $3.59 | 2026-10-23 | On-demand list price (H200 Community Cloud); availability and surcharges per official site. | source |
| Vast.ai | $4.277 | 2026-10-23 | Lowest marketplace offer for Vast.ai H200; host reliability not guaranteed. 5 matching listings. | source |
| RunPod | $5.29 | 2026-10-23 | On-demand list price (H200 Secure Cloud); availability and surcharges per official site. | source |
H100 vs H200: which should you rent?
Both are Hopper-generation cards. The key difference is memory: H100 has 80 GB, H200 has 141 GB. More memory means you can fit larger models, use higher precision, run longer context, or need fewer cards to reach the same capacity.
On price, H200 hourly rates at the same platform are typically 1.5–2× H100 in the data we collect (see the table above for current numbers). The practical rule: estimate your VRAM requirement first — if 80 GB fits, rent H100; only move to H200 when you actually need the capacity.
How is the price calculated?
GPU cloud platforms charge per GPU-hour. The listed price is per card; multi-GPU instances multiply it. Storage, network egress, and setup fees are usually billed separately. Long-running jobs are often cheaper with monthly commitments, while short experiments are better on on-demand hourly pricing.
Check how much VRAM your model needs first — see the VRAM calculator. If you are comparing self-hosting against renting, start from the hourly rate above.