Serverless
A model where the provider allocates and scales compute per request, and you are billed for execution rather than for provisioned capacity.
The genuine advantages are scale-to-zero, per-request billing, and no operating system to patch. For spiky, event-driven and low-baseline workloads the economics are hard to beat.
The genuine constraints are equally concrete. Cold starts add latency to the first request after idle. Execution has a time limit. Local state does not survive. Connection-pool exhaustion is a recurring problem, because a thousand concurrent functions try to open a thousand database connections — which is what connection proxies exist to fix. And at steady high utilisation, per-request pricing crosses over and becomes more expensive than a reserved instance.
The honest rule of thumb: serverless for event-driven and bursty work, containers for sustained throughput, and do the crossover arithmetic rather than assuming.