Runpod is a cloud computing platform for AI workloads, made by Runpod Inc. It rents GPUs (graphics processors, used to train and run AI models) in three forms: Pods, on-demand GPU machines in 31 regions; Serverless, GPU endpoints behind an API; and Clusters of several GPU machines for distributed jobs.
With Runpod Serverless, a developer packages a workload as a container and deploys it as an autoscaling endpoint. Requests go to workers; Runpod adds workers when requests arrive in bursts and scales down to zero when none arrive, so idle time costs nothing. FlashBoot keeps cold starts, the delay before a new worker can answer, under 200 milliseconds, and active workers stay pre-warmed. Deployments can come straight from GitHub, with rollback to earlier container versions. Persistent network storage keeps files between runs. Supported languages include Python, Node.js, Go, Rust and C++, with frameworks such as PyTorch. Typical uses are inference, fine-tuning and compute-heavy tasks such as rendering.
Runpod is a hosted service billed per second, from worker start to full stop. The price depends on the GPU type.