The serverless promise and its bill

Serverless is genuinely magical when it fits: you write a function, it runs, you pay by the millisecond. The bill is small at first. Then traffic grows, or one inefficient function gets called more, and the bill stops being small in a hurry.

Where the bill comes from

• Cold-start times that get billed but provide no value.
• Inefficient functions multiplied by traffic spikes.
• Per-invocation data transfer that adds up at volume.
• Database connections that don’t pool well in stateless functions.

When it’s worth it anyway

Sporadic workloads. Background tasks with unpredictable timing. Early stage where ops time is the constraint, not money. Don’t use it everywhere, but don’t avoid it on principle either.

Serverless is right until traffic is steady. Then it’s just a server you didn’t optimize.