Cloud GPU providers
37 providers, the ones with most GPUs first, with their cheapest on-demand H100 and A100 rates. Open a provider for every price, its billing rules and its other GPUs.
| Provider / locations | GPU models | H100 on-demand | A100 on-demand | Cheapest $/hr | Offers | Checked | |
|---|---|---|---|---|---|---|---|
| RunPod–Prices → | 22 | $2.89 | $1.79 | $0.25RTX A4000 · secure cloud | 58 | 2026-10-12 | Visit → |
| LiumAustralia, Canada, Finland, Germany, Greece, Hong Kong, Ireland, Israel, Japan, Kazakhstan, Lithuania, Romania, Singapore, Ukraine, United Kingdom, United StatesPrices → | 13 | – | – | $0.12RTX A4000 · marketplace host | 74 | 2026-10-12 | Visit → |
| Spheron–Prices → | 12 | $2.65 | $1.43 | $0.48RTX 4090 · on-demand | 18 | 2026-10-12 | Visit → |
| SaladCloud–Prices → | 12 | – | – | $0.065RTX 5060 · community cloud | 13 | 2026-10-12 | Visit → |
| Massed Compute–Prices → | 11 | $2.73 | $1.35 | $0.35A30 · on-demand | 14 | 2026-10-12 | Visit → |
| DataCrunch–Prices → | 11 | $4.05 | $1.41 | $0.212Tesla V100 · on-demand | 12 | 2026-10-12 | Visit → |
| Modal–Prices → | 10 | – | – | $0.590Tesla T4 · serverless | 11 | 2026-10-12 | Visit → |
| Hyperstack–Prices → | 9 | $2.50 | $1.35 | $0.15RTX A4000 · on-demand | 21 | 2026-10-12 | Visit → |
| TensorDock–Prices → | 9 | $2.25 | $1.50 | $0.10RTX A4000 · on-demand | 11 | 2026-10-12 | Visit → |
| Cerebrium–Prices → | 9 | – | – | $0.590Tesla T4 · serverless | 10 | 2026-10-12 | Visit → |
| Lambda–Prices → | 8 | $3.29 | $1.99 | $0.69Quadro RTX 6000 · on-demand | 22 | 2026-10-12 | Visit → |
| Google Cloud–Prices → | 8 | $9.80 | $2.93 | $0.35Tesla T4 · on-demand | 18 | 2026-10-12 | Visit → |
| Daytona–Prices → | 8 | $3.95 | – | $0.99RTX 4090 · on-demand | 16 | 2026-10-12 | Visit → |
| DigitalOcean–Prices → | 8 | $4.41 | – | $1.57RTX 6000 Ada · on-demand | 14 | 2026-10-12 | Visit → |
| Koyeb–Prices → | 8 | – | – | $0.70L4 · serverless | 9 | 2026-10-12 | Visit → |
| CoreWeave–Prices → | 8 | $6.16 | $2.70 | $1.25L40 · on-demand | 8 | 2026-10-12 | Visit → |
| OVHcloud–Prices → | 8 | $2.99 | $3.07 | $0.60Quadro RTX 5000 · on-demand | 8 | 2026-10-12 | Visit → |
| Vultr–Prices → | 7 | – | $2.40 | $1.67L40S · on-demand | 7 | 2026-10-12 | Visit → |
| Baseten–Prices → | 6 | – | – | $0.631Tesla T4 · serverless | 6 | 2026-10-12 | Visit → |
| fal–Prices → | 6 | – | – | $4.00RTX PRO 6000 · serverless | 6 | 2026-10-12 | Visit → |
| Nebius–Prices → | 6 | $4.50 | – | $1.35L40S · on-demand | 6 | 2026-10-12 | Visit → |
| AWSus-east-1Prices → | 5 | $6.88 | $2.74 | $0.805L4 · on-demand | 12 | 2026-10-12 | Visit → |
| Crusoe–Prices → | 5 | $3.90 | $2.00 | $1.50L40S · on-demand | 6 | 2026-10-12 | Visit → |
| Jarvislabs–Prices → | 5 | $3.49 | $0.89 | $0.44L4 · on-demand | 6 | 2026-10-12 | Visit → |
| Fireworks AI–Prices → | 5 | – | – | $8.00H100 · serverless | 5 | 2026-10-12 | Visit → |
| Replicate–Prices → | 5 | – | – | $0.81Tesla T4 · serverless | 5 | 2026-10-12 | Visit → |
| Together AI–Prices → | 4 | $3.99 | – | $3.99H100 · on-demand | 8 | 2026-10-12 | Visit → |
| Prime Intellect–Prices → | 4 | $2.43 | – | $2.43H100 · on-demand | 5 | 2026-10-12 | Visit → |
| Scaleway–Prices → | 4 | $3.21 | – | $0.882L4 · on-demand | 5 | 2026-10-12 | Visit → |
| GMI Cloud–Prices → | 4 | $2.60 | – | $2.60H100 · on-demand | 4 | 2026-10-12 | Visit → |
| Thunder Compute–Prices → | 4 | $3.20 | $1.09 | $0.35RTX A6000 · on-demand | 4 | 2026-10-12 | Visit → |
| Impossible CloudAPAC, EU, North America, USPrices → | 3 | – | – | $2.95H100 SXM · reserved / committed | 12 | last 2026-10-10 | |
| Latitude.shUnited StatesPrices → | 3 | $3.37 | – | $3.37H100 · on-demand | 9 | 2026-10-12 | |
| Microsoft AzureeastusPrices → | 2 | $6.98 | $3.40 | $3.40A100 SXM 40GB · on-demand | 22 | 2026-10-12 | Visit → |
| Synteq–Prices → | 2 | – | – | $1.21RTX PRO 6000 · reserved / committed | 2 | 2026-10-12 | |
| Hot Aisle–Prices → | 1 | – | – | $2.99MI300X · on-demand | 1 | 2026-10-12 | Visit → |
| Voltage Park–Prices → | 1 | $1.99 | – | $1.99H100 · on-demand | 1 | 2026-10-12 | Visit → |
H100 and A100 columns are each provider's cheapest on-demand rate for any variant of that GPU; – means no on-demand rate is recorded there. The cheapest rate prefers on-demand capacity and otherwise names its billing type. Referral status never changes the order. Every tracked provider has a price page.
Retired providers: price history
SF Compute's recorded public prices. Prices have not been public since 25 September 2026; collection ended on 4 October 2026.