Runpod

What it does

Runpod is a cloud platform for AI workloads. It offers on-demand GPU Pods, serverless endpoints for API inference, and multi-node clusters. The company says it supports over 30 GPU SKUs across 31 global regions, all managed from one account.

Runpod provides GPU compute for AI work through three main products. Pods give you dedicated GPU instances. Serverless runs API-based workloads on GPU endpoints that scale from zero, so an idle endpoint costs nothing. Clusters handle multi-node jobs for distributed training. The site says you can launch a GPU pod in under a minute and that Serverless scales to thousands of workers with sub-200ms cold starts. Pricing is listed per hour or per second depending on the GPU and cloud type, and the pricing page shows rates for cards such as B300, H200, B200, H100, A100, L40S, and RTX models. The homepage also mentions persistent network storage with no egress fees. Before you commit, check regional availability, storage terms, and whether your framework fits the supported containers. The platform is broad, so a smaller project may find a simpler host cheaper.