Serverless platforms—AWS Lambda, Google Cloud Functions, Azure Functions, Cloudflare Workers, Vercel Functions—run code in response to events without exposing the underlying machines to the developer. Billing is per request and per millisecond of execution, which means idle code costs nothing. The model fits webhooks, scheduled jobs, image processing pipelines, and increasingly the API layer of full applications when paired with managed databases like DynamoDB, Aurora Serverless, or PlanetScale.
The tradeoffs are real and worth naming. Cold starts can add hundreds of milliseconds to the first request after idle periods—painful for latency-sensitive APIs, largely solved on edge runtimes like Cloudflare Workers, still relevant on classic Lambda. Local development and testing remain awkward; observability has improved but still feels less mature than what container-based shops are used to. The convenience of pay-per-use also flips into a liability under sustained high traffic, where a properly sized container often costs less per request.
A practical heuristic: serverless wins for workloads with low average load and unpredictable spikes, for glue code between SaaS systems, and for new products where the team would rather ship than tune autoscaling. It loses when traffic is steady and high, when long-running connections matter (WebSockets, gRPC streams), or when regulatory constraints rule out the runtime’s compliance posture for the data involved.