Explanation
Why the waste happens and who it affects.
For databases that are active only briefly within an hour, or for limited hours per month, Serverless can provide significantly lower cost because it bills per second for the compute used, never less than the configured minimum vCores and memory while the database is online, and only for storage while the database is auto-paused (auto-pause is supported only in the General Purpose tier). The economic break-even point between Provisioned and Serverless depends on workload activity patterns. If monthly active time falls below the conceptual break-even range, Serverless is more cost-effective. If active time regularly exceeds that range, Provisioned may be more appropriate. This inefficiency typically appears when teams default to Provisioned compute without evaluating workload behavior over time.
Billing model
The pricing dimensions that drive this cost.
Provisioned compute is billed per vCore-hour regardless of usage. Serverless compute is billed per second for the vCores and memory used, with a floor at the configured minimum vCores and memory while the database is online, plus storage; compute charges stop only while the database is auto-paused, which is supported only in the General Purpose tier. Azure Hybrid Benefit and reservation discounts do not apply to serverless, so compare against the discounted provisioned price when estimating the break-even point. Selecting Provisioned for workloads with low or sporadic utilization results in paying for unused capacity.
How to detect
4 checks to find it in your estate.
- Review whether database activity is intermittent or consistently low throughout the day
- Assess whether monthly active time is substantially lower than the equivalent cost of Provisioned compute
- Evaluate whether workload patterns include extended idle periods suitable for auto-pause
- Check for development, testing, or sporadic workloads that do not require continuous compute availability
How to fix
4 ways to remove the waste.
- Move the database to the Serverless compute tier when workloads are intermittent or have low active time
- Use Serverless auto-pause capabilities to eliminate compute charges during idle periods
- Periodically reassess compute model selection as application usage patterns evolve
- Apply workload profiling to ensure that Serverless performance meets application needs before migration
Documentation
Vendor references for pricing and configuration.