Serverless computing promised to eliminate infrastructure management entirely and has delivered on that promise in many contexts. But like any architectural pattern, serverless has a specific sweet spot — and applying it outside that zone creates problems that outweigh the benefits.
Where Serverless Excels
Serverless shines for event-driven workloads with variable traffic patterns. API backends that see spiky load, data processing pipelines triggered by file uploads or queue events, scheduled jobs, and webhook handlers are all excellent serverless candidates. The auto-scaling-to-zero model provides both operational simplicity and cost efficiency for these use cases.
Where Serverless Struggles
Serverless runs into trouble with long-running computations (execution time limits), applications requiring persistent connections (WebSockets, gRPC streaming), workloads that need predictable latency (cold start variance), and systems where compute costs are predictably high (the economics invert at sustained high throughput). Knowing these limits saves you from painful architectural reversals.