Simple GPU pricing. Pay for what you run.
Rates are per GPU per hour and billed per second. Final prices are set by each host, so the marketplace shows live offers at or above the starting prices below.
| GPU | VRAM | Starting price / hr | Best for |
|---|
Three ways to rent
On-demand
Fixed price, guaranteed until you stop it. Best for production, demos and anything you can't afford to lose.
Interruptible
Bid on spare capacity at a steep discount. Your instance can pause if a higher bid arrives, so save checkpoints often.
Reserved
Commit to a month and get a lower rate on the same on-demand instance. Pay up front and run it as long as you like.
Estimate your cost
Other charges
| Item | Price |
|---|---|
| Compute | Per second, at the hourly rate shown |
| Storage (instance running or stopped) | |
| Data in | Free |
| Data out | |
| Minimum top-up | |
| Platform fee |
How HostGPU compares
| GPU, per hour | HostGPU (from) | RunPod Community | RunPod Secure | Lambda |
|---|---|---|---|---|
| RTX 4090 | $0.34 | $0.34 | $0.69 | Not offered |
| A100 80 GB | $0.67 | $1.39 (PCIe) | n/a | $1.29 |
| H100 80 GB | $1.49 | $1.99 (PCIe) | $2.89 (PCIe) | $3.99 (SXM) |
Competitor prices are published rates as of August 2026 and change often. Check each provider before you decide.
Questions
Why do prices vary between listings?
Each host sets their own price based on hardware, location, bandwidth and reliability. Verified datacenter hosts usually cost more than home or small-farm hosts.
What happens when my credit runs out?
Running instances are stopped and your data is kept for a grace period. Add credit and restart. Enable auto top-up to avoid interruptions.
Do stopped instances still cost money?
You only pay storage for stopped instances. GPU compute stops billing the second you stop.
Can I get a refund?
Unused credit can be refunded to your original payment method within 30 days. Used compute is not refundable.
Do you offer volume discounts?
Yes. For 8+ GPUs or monthly spend over $5,000, contact us for custom rates.