Runpod

Runpod is a web-based AI infrastructure platform for AI developers, researchers, and teams that need cloud GPUs for training, inference, agents, and batch processing.

Key capabilities include:

• On-demand GPU pods launched as virtual instances with configurable RAM and vCPU, supporting container-based deployments such as notebooks, inference services, and custom Docker images (GPU options include A10G, A100, RTX 3090, H100, with listed starting prices as low as $0.17/hour for some GPUs).
• Serverless inference endpoints designed for production traffic with auto-scaling, zero cold starts with active workers, and scaling to over 1,000 requests/second (per published case studies).
• Persistent storage options including network volumes and pod disks, billed separately from compute (network volumes $0.05–$0.14/GB/month, pod volumes $0.10–$0.20/GB/month), plus zero egress fees.
• Deployment helpers such as pre-configured AI templates, idle container auto-shutdown, and multi-region availability (listed across 31 global regions).

Pricing: Paid, with pay-as-you-go per-second billing for compute, plus spot (preemptible/community cloud), on-demand (secure cloud), and savings plans (3–6 month commitments) for discounts.

Notable for combining low-friction GPU access (pods, serverless endpoints, and clusters) with cost-focused billing such as zero egress fees and optional spot capacity, aimed at reducing GPU infrastructure overhead compared to large cloud providers.