GPU cloud pricing compared
Rented accelerators for training and inference, billed by the GPU-hour. The spread between the specialist hosts and the hyperscalers is the widest of any category on this site — often 3x to 5x for the identical card — because the specialists sell capacity and the hyperscalers sell it bundled with everything else.
List prices checked
Rate cards
All 12 GPU cloud providers in the dataset, sorted by the axis this category is actually billed on.
| Service | Provider | Entry /mo | Egress /GB | Storage /GB-mo |
|---|---|---|---|---|
| Vast.ai H100 | Vast.ai | — | — | — |
| Together AI H100 | Together AI | — | — | — |
| Hyperstack H100 | Hyperstack | — | — | — |
| RunPod H100 | RunPod | — | — | — |
| Paperspace H100 | Paperspace | — | — | — |
| CoreWeave H100 | CoreWeave | — | — | — |
| Lambda H100 | Lambda | — | — | — |
| Modal H100 | Modal | — | — | — |
| Replicate H100 | Replicate | — | — | — |
| Google Cloud A3 | — | — | — | |
| Azure ND H100 v5 | Microsoft | — | — | — |
| AWS EC2 P5 | Amazon Web Services | — | — | — |
Ranked by real monthly cost: Growing SaaS
A funded product serving a few thousand daily users, with an image-heavy frontend and two application servers. Each provider's own published rates, applied to identical usage.
| # | Service | Monthly | Breakdown |
|---|---|---|---|
| 1 | Vast.ai H100 | $2,409 | GPU $2,409 |
| 2 | Together AI H100 | $2,555 | GPU $2,555 |
| 3 | Hyperstack H100 | $2,774 | GPU $2,774 |
| 4 | RunPod H100 | $2,905 | GPU $2,905 |
| 5 | Paperspace H100 | $3,270 | GPU $3,270 |
| 6 | CoreWeave H100 | $3,373 | GPU $3,373 |
| 7 | Lambda H100 | $3,635 | GPU $3,635 |
| 8 | Modal H100 | $5,767 | GPU $5,767 |
| 9 | Replicate H100 | $7,358 | GPU $7,358 |
| 10 | Google Cloud A3 | $10,191 | GPU $10,191 |
| 11 | Azure ND H100 v5 | $10,191 | GPU $10,191 |
| 12 | AWS EC2 P5 | $11,052 | GPU $11,052 |
Excludes committed-use discounts, regional variation, and per-request charges. Services that do not publish a rate for every axis this category bills on are listed last rather than shown as cheap.
Questions
- What is the cheapest GPU cloud provider in 2026?
- For the "Growing SaaS" workload — a funded product serving a few thousand daily users, with an image-heavy frontend and two application servers. — Vast.ai H100 comes out cheapest at $2,409 per month on published list prices. Ranked purely on the headline rate the order is different, which is why both tables are on this page.
- What does an H100 actually cost per hour, and how much am I paying for the hyperscaler's name?
- The second table answers it directly: it applies each provider's published rates to one identical workload and ranks by the resulting monthly bill. Change the workload and the ranking changes — that is the point, and it is why a single "best" list is usually wrong.
- Are these gpu cloud prices current?
- Each row is stamped with the month it was last checked against the provider's own pricing page, and every service page links straight to that page. Providers change list prices without notice, so treat these as a starting point and confirm before you commit.