Runpod

Runpod is a GPU cloud built for people who train, fine-tune and serve AI models. It offers three ways to rent compute from one account. Pods are on-demand GPU machines that start in about a minute, with SSH, a web terminal and HTTP ports for tools like Jupyter Lab, and run on either the cheaper Community Cloud or the Secure Cloud. Serverless turns a container into an API endpoint backed by GPU workers that autoscale with the request queue, with logs, metrics, builds and releases in the console. Clusters provide multi-node GPU capacity for distributed training.

The hardware range runs from consumer cards such as the RTX 3090, 4090 and 5090 through the A40, L40S and A100 to H100, H200, B200 and B300, priced per hour or per second, with the exact rate shown per GPU on the pricing page. In October 2026 Runpod made Batch Inference generally available on Serverless: large volumes of jobs that can wait run on dedicated batch workers, separate from real-time traffic, at up to 30% off on-demand pricing depending on GPU type.

The Runpod Hub lists open-source models and worker templates (Whisper, Stable Diffusion and others) that deploy in a few clicks, and a Python SDK and rp CLI cover deployment from code. Runpod also publishes MCP servers and agent skills so coding agents can manage GPU workloads. It suits independent developers and startups that want raw GPU access without long contracts.