Serverless Cost: When It Wins, When It Doesn’t

Serverless economics are uniquely shaped. The threshold is well-defined but rarely calculated.

Serverless cost shape

Per-request cost is high; scaled-to-zero is free. The crossover happens at a specific request rate per service.

Below crossover: serverless is unbeatable. Above: provisioned wins.

Four thresholds

• Lambda crossover: ~10M requests/month for typical compute.
• Fargate crossover: ~24/7 utilization at 70%+.
• API Gateway crossover: ~50M req/month vs ALB.
• DynamoDB on-demand crossover: ~50K writes/day.

Migration patterns

When serverless stops paying: migrate to provisioned (ECS, EKS, EC2). Same code; different runtime.

Migration is mechanical for stateless; harder for stateful.

Hybrid posture

Many orgs: serverless for sporadic; provisioned for sustained. Each serves what it serves best.

The hybrid is the realistic posture.

Antipatterns

• Serverless forever despite high volume. 3-5x bill premium.
• Provisioned for sporadic. Pay for idle.
• No threshold tracking. Crossover passes silently.

What to do this week

Three moves. (1) Apply this lever to your highest-spend workload. (2) Measure the dollar impact for one month. (3) Roll the practice out to the next two services if the savings hold.