A logistics company runs an on-demand image-processing job several times a day. Each invocation spins up a container that runs for 3 to 8 minutes, then exits. Volume is unpredictable — sometimes zero jobs for hours, sometimes 40 concurrent jobs. The team wants to run these containerized workloads with per-second billing, no infrastructure or node pool to manage, and rapid startup, while avoiding paying for idle capacity. Which compute option best fits these requirements?