H200 cloud GPU price comparison
| provider | $/gpu/hr | type | |
|---|---|---|---|
| RunPod | $0.50 | community | view → |
| Lambda | $2.29 | on-demand | view → |
| Vast.ai | $2.88 | on-demand | view → |
| RunPod | $3.59 | community | view → |
| RunPod | $3.79 | on-demand | view → |
| RunPod | $4.39 | on-demand | view → |
What does an H200 cost to rent?
As of 2026-07-05, the cheapest H200 rental is $0.50 per GPU per hour at RunPod. Prices on this page are collected automatically from each provider's public pricing and refreshed daily, so the table above reflects the current market rather than a stale blog snapshot. The overall floor of $0.50/hr is a community offer, which can be interrupted; the cheapest regular on-demand price is $2.29/hr at Lambda.
About the H200
The NVIDIA H200 pairs the Hopper architecture with 141 GB of HBM3e, nearly doubling the H100's memory bandwidth. It shines for large-model inference and memory-bound training workloads. It ships with 143 GB of VRAM per GPU.
On-demand vs community and spot pricing
On-demand instances run at a fixed rate until you stop them. Community and spot offers (marketplaces like Vast.ai, RunPod community cloud) are often dramatically cheaper but can be interrupted or vary in reliability between hosts. For long training runs, weigh the price gap against the cost of checkpoint restarts; for interruptible inference and experimentation, the cheaper tiers are usually the rational pick.