A startup guide to GPU cloud for AI inference: compare latency, memory, billing, self-managed Compute, managed endpoints, and scaling responsibilities.