“Serverless” sounds cheaper because it scales to zero, but that is only half the story. Whether serverless or provisioned wins on cost depends entirely on your traffic shape. Here is the framework, with real June-2026 pricing.
The two billing models
| Model | Examples | You pay for |
|---|---|---|
| Serverless / usage | Upstash Redis, MongoDB Flex, DataStax Astra, MSK Serverless, ClickHouse Cloud | Requests, throughput or compute-seconds actually used |
| Provisioned / instance | RDS MySQL, ElastiCache nodes, MSK brokers, Instaclustr | A sized instance/node, billed per hour whether busy or idle |
When serverless wins
Serverless wins when your resource would otherwise sit mostly idle:
- Spiky traffic — a cache that is busy at peak and quiet overnight.
- Low duty cycle — a side project or internal tool used a few hours a day.
- Unpredictable load — you can’t size an instance confidently.
Upstash Redis at $0.20 per 100K commands costs almost nothing for a low-traffic app, where a provisioned ElastiCache node would cost ~$12/month minimum, busy or not.
When provisioned wins
Provisioned wins when the resource is consistently busy:
- Steady high throughput — a cache or database hammered 24/7.
- Predictable load — you can size and reserve capacity.
- Reserved discounts — RDS and ElastiCache cut up to ~69% with a 1-3 year commitment.
A serverless service billing per request becomes expensive at sustained high volume; a flat hourly instance is cheaper once it is busy most of the time.
The crossover
There is no universal break-even, but a useful rule of thumb is 30-50% sustained utilisation. Below that, usage-based pricing usually wins; above it, a reserved provisioned instance typically wins. Kafka is the clearest example: MSK Serverless ($0.75/cluster-hour plus throughput) is great for bursty streams, but a provisioned 3-broker cluster (~$447/month) is cheaper for a steady high-volume pipeline.
Bottom line
Don’t assume serverless is cheaper — match the model to your traffic. Estimate both in the cost estimator, compare specific offerings in our head-to-head comparisons, and read how to cut managed-database costs. Prices are a June-2026 snapshot; verify on each vendor’s pricing page.