RunPod

RunPod carved out a niche in the cloud GPU market by aggregating compute capacity from both its own data centres and third-party data centre partners into a single marketplace, offering GPU rental at prices well below the major cloud providers. Founded in 2022, the platform has attracted AI researchers, fine-tuners, and inference operators who need GPU access on demand without committing to reserved instances. The Serverless GPU offering — where customers pay per compute unit consumed by inference workloads rather than per hour of reserved instance — addresses the bursty, variable nature of inference traffic in a way that traditional reserved GPU instances do not.

RunPod's pod architecture gives users direct GPU access through either a web-based terminal and Jupyter interface or SSH access to the underlying pod container. The platform provides pre-built Docker image templates for common ML frameworks — PyTorch, TensorFlow, ComfyUI, Stable Diffusion, and others — reducing the setup time for common workloads. Persistent volumes provide storage that survives pod restarts, and network volumes allow data sharing across pods. The RunPod API enables programmatic pod creation and termination, making it usable for batch compute workflows that spawn and destroy pods based on job queues.

RunPod service disruptions reflect the marketplace architecture's specific failure modes. GPU availability on popular configurations — particularly H100 and A100 instances — can drop to zero during demand spikes, returning "no capacity available" errors even when the API and management plane are functioning normally. Community Cloud pods, which run on third-party hardware contributed by data centre partners, experience connectivity issues when individual partner locations have network problems — pods may become unreachable over SSH or HTTP even when the RunPod management platform itself is healthy. The RunPod API may return provisioning errors or timeouts during control plane load, and pods stuck in a "starting" state require manual intervention to resolve. Persistent volume I/O degradation during storage backend issues causes training checkpoints and model files to write slowly or fail.

Outage.gg tracks RunPod service status using real-time community reports from ML engineers and AI developers. If pods are not starting, GPU capacity is unavailable, or the RunPod API is returning errors, the live status page shows current impact from the RunPod user community.