GPU Cloud Price Comparison
On-demand hourly rates, read from each provider's published pricing page on 2 September 2026. Pick a GPU and the hours you expect to run it.
Some vendor links on this page are affiliate links: if you sign up through them we may earn a commission, at no extra cost to you. The pricing and rankings shown here are calculated from published rates and are not affected by those arrangements.
| Provider | Per hour | 730h / month | vs cheapest | Notes |
|---|---|---|---|---|
| RunPod | $3.29 | $2402 | — | Secure Cloud |
| Modal | $3.95 | $2884 | +$482 | billed per second |
| Lambda | $3.99 | $2913 | +$511 | from; up to $4.29 |
At 730 hours a month, running a H100 SXM on RunPod rather than Lambda saves $511 a month, or $6132 a year.
What these numbers do and do not include
These are on-demand rates for a single GPU, before reserved capacity, committed-use contracts or spot pricing — any of which can cut the figure substantially for a predictable workload. Storage, egress and idle time are excluded.
For bursty work, billing granularity matters as much as the rate. Modal bills per second, so a job that runs nine minutes costs nine minutes. An hourly-billed instance left running between jobs can easily cost more at a lower advertised rate than a per-second one at a higher rate.
RunPod figures are Secure Cloud. Its Community Cloud tier is cheaper but runs on hosts of varying provenance — a different risk profile, not simply a better price.
Vast.ai has no row above because it has no fixed rate to put there: it is a marketplace of individual hosts bidding against each other, so its prices move continuously and a figure printed here would be wrong by the time you read it. It is usually the cheapest way to get a given GPU, particularly on interruptible instances, at the cost of variable host quality and the risk of being pre-empted. Worth checking live before committing to any rate above.