RunPod offers three core primitives: Pods (on-demand GPU instances across 31 regions and 30+ SKUs, from B200s to RTX 4090s), Serverless (API-based GPU endpoints with sub-200ms cold starts via FlashBoot, auto-scaling from 0 to thousands of workers, and zero idle cost), and Clusters (multi-node GPU clusters for distributed training). The platform is container-native — developers bring their own Docker images, frameworks, and code. A managed orchestration layer handles queuing, failover, real-time logging, and metrics with no custom framework required. RunPod is also an OpenAI infrastructure partner and is SOC 2 Type II certified.