About RunPod

RunPod rents GPUs by the second for machine learning work. It is not a general VPS host. You pick a GPU, pick a container image, and get a running pod, usually in well under a minute.

The thing that makes it different from a normal cloud is the split between two pools. Secure Cloud runs in real data centers with the reliability you would expect. Community Cloud is a marketplace of hardware from third-party operators, often at half the price. Community pods can and do disappear when the host reclaims the machine, so the tradeoff is explicit rather than hidden.

For training runs you can checkpoint, or for inference you can restart, Community is good value. For anything you cannot afford to have vanish, use Secure.

Pricing

Billing is per second with no monthly minimum, so a listed monthly price does not really apply. As a rough guide, consumer cards like the RTX 4090 sit around $0.35/hr on Community, an A100 80GB around $1.90/hr on Secure, and an H100 around $2.80/hr. Serverless endpoints bill only while a request is being processed, which suits bursty inference far better than a pod sitting idle.

Storage is billed separately, and this is the line people miss. A stopped pod still charges for its volume. If you spin up a 500GB workspace and forget it, you keep paying.

GPU pricing moves faster than almost anything else in hosting, so treat these as indicative and check the current rates.

Service Details

Pods are containers, not VMs. You get root inside the container and a Jupyter or SSH endpoint, but you are not running your own kernel, so anything needing kernel modules or custom drivers will not work. Templates cover PyTorch, TensorFlow, ComfyUI, text-generation-webui, and Ollama, and you can supply your own image from any registry.

The API and CLI let you create and destroy pods programmatically, which is the sane way to use an hourly GPU. Network volumes persist across pods in the same region, so you can keep a model cache without paying for an idle machine.

Ubuntu is the base for essentially every template. If you are new to running models locally, our AI and ML self-hosted apps directory and the Docker and Podman guide cover the groundwork.

Available in: United States, Netherlands, Iceland, Romania, Sweden, Canada

Learn More

For more information about RunPod, visit their website at https://www.runpod.io/

Plan	Price/mo	CPU	RAM	Storage	Notes
RTX 4090 (Community)	—	8 vCPU	32GB	80GB NVMe	Around $0.35/hr, third-party hosts
A100 80GB (Secure)	—	16 vCPU	128GB	200GB NVMe	Around $1.90/hr, tier 3 or 4 data centers
H100 80GB (Secure)	—	24 vCPU	250GB	500GB NVMe	Around $2.80/hr