When managed databases stop being cheaper than self-hosted

Managed databases remove a large operational burden. They also introduce cost lines that are easy to underestimate: provisioned IOPS, storage that grows with the data plus the backups, cross-region replicas, and connection or proxy charges. At some combination of size, IOPS and replication the managed option becomes more expensive than a well-run self-hosted alternative on the same cloud.

I do not start from ideology. I start from a side-by-side monthly cost model that includes:

• Instance / node price
• Storage and IOPS (or the managed equivalent)
• Backup storage and retention
• Cross-region or multi-AZ replication
• Connection pooling / proxy if required
• The engineering time still needed for schema changes, major version upgrades and incident response

When the managed total is higher and the team already has the operational maturity to run the database, the self-hosted path is worth evaluating. When the managed total is lower or the team does not want the operational load, managed stays.

Typical tipping points I have observed

• High sustained IOPS on a large dataset
• Multiple cross-region read replicas
• Very long backup retention with frequent restores
• Connection counts that force the next larger managed tier or an expensive proxy

A simple spreadsheet with the current managed bill on one side and a realistic self-hosted configuration (including the extra engineering hours) on the other is enough to make the decision data-driven instead of emotional.

Managed is still the right default for most teams. The moment the bill shows that storage, IOPS or replication dominate, it is time to run the comparison rather than assume managed is always cheaper.

I revisit the comparison whenever the data size or the IOPS requirement roughly doubles, or when a new region is added. The numbers change; the framework stays the same.