What is the price difference between the H100 and H200?
The H200 is the same Hopper architecture as the H100 but with more and faster memory, so it rents at a premium, commonly a moderate step above the H100 rate for the same provider and tier. The cheapest verified H100 we track is $1.99/hr. Whether the H200 is worth the extra depends entirely on whether your workload is memory-bound.
| GPU | VRAM | $/hr | Where | |
|---|---|---|---|---|
| H100 NVL | 94 GB | $1.40 | Microsoft Azure spot / interruptible | Rent → |
| H100 PCIe | 94 GB | $1.99 | RunPod community cloud | Rent → |
| H100 (unspecified) | 94 GB | $1.99 | Voltage Park on-demand | Rent → |
| H100 (unspecified) | 94 GB | $2.01 | Spheron on-demand | Rent → |
| H100 SXM | 94 GB | $2.04 | Microsoft Azure spot / interruptible | Rent → |
| H100 SXM | 94 GB | $2.30 | Vultr on-demand | Rent → |
| H100 (unspecified) | 94 GB | $2.50 | Hyperstack on-demand | Rent → |
| H100 NVL | 94 GB | $2.59 | RunPod community cloud | Rent → |
The core compute of the two cards is similar. What the H200 adds is a jump in memory capacity and bandwidth, which matters most for large-model inference and for keeping big context windows in VRAM. If your job is bottlenecked on memory rather than raw math, the H200 can finish faster and end up cheaper per unit of work despite the higher hourly rate.
For training and inference that already fit comfortably on an H100, the extra memory sits unused and the premium is wasted. In that case the H100, or even an A100, is the better value.
Because both cards float on the same supply and demand, the gap between them narrows and widens over time. The table below shows current H100 rates across providers; compare them against live H200 listings before committing to the newer card.
Related questions
- How much does an H100 cost per hour?
- What is the cheapest B200 rental?
- What is the cheapest 80GB GPU to rent?
Numbers on this page come from today's verified snapshot. Full table on the homepage; method in the methodology.