Per-second billing
You pay for the seconds your instance runs. Stopped instances only pay for storage.
GPU compute, priced like a marketplace.
Rent H100s, A100s and RTX cards from verified hosts worldwide. Billed per second, no contracts, a running instance in under a minute.
- Per second
- billing, no minimums
- $5
- minimum top-up
- 0
- contracts
| GPU | VRAM | Location | Reliability | Price / hr |
|---|
Sample listings. Connect your backend to show live offers. See pricing
From sign-up to training in four steps
Add credit
Top up with card or crypto. Start from $5.
Pick a GPU
Filter by model, memory, location and price. Compare hosts side by side.
Choose a template
PyTorch, TensorFlow, ComfyUI, Jupyter or your own Docker image.
Connect
Open Jupyter, SSH in, or call your instance's API. Stop it any time and billing stops with it.
Built for people who run real workloads
Interruptible pricing
Bid on spare capacity and save up to 70% on training jobs that can checkpoint.
Verified hosts
Every host passes hardware, bandwidth and uptime checks, with a public reliability score.
Full root access
Bring your own Docker image, mount persistent volumes and use SSH keys.
Developer API
Search offers, launch and destroy instances from the CLI or REST API.
Global locations
Choose hosts in North America, Europe and Asia to stay close to your data.
H100 price per GPU-hour
HostGPU shows starting prices. Competitor figures are published on-demand rates (RunPod, Lambda) as of August 2026. See full rate card
Got idle GPUs? Put them to work.
List your rigs on HostGPU, set your own price and get paid for every hour they run. Install our host agent, pass verification and start earning.