Why does Power BI Service report different System Storage for two semantic models identical

Hi everyone,

I'm investigating an unexpected difference in Power BI Service → Workspace → Settings → System Storage between two semantic models.

Both models were created from the same PBIX. The only difference is the data source:

• Model A → Amazon Redshift
• Model B → ClickHouse

To verify that the models are equivalent, I compared them using DAX Studio (VertiPaq Analyzer).

Comparison results

• Same number of rows (~25.9 million)
• Same tables
• Same columns
• Same partitions (1 partition, 25 segments)
• Same column cardinality
• Nearly identical table sizes
• Nearly identical dictionary sizes
• Nearly identical relationship sizes
• Overall VertiPaq size differs by only a few MB

I also verified that:

• Both models refresh successfully.
• Both were published to a brand-new empty workspace.
• The same behavior occurs in the new workspace.

However, Workspace → Settings → System Storage reports significantly different storage usage for the two semantic models.

Since the VertiPaq models are almost identical, I expected the System Storage values to also be nearly identical.

My questions

1. What exactly is included in System Storage besides the VertiPaq model?
2. Is System Storage expected to depend on the data source connector (Redshift vs. ClickHouse)?
3. Are processing artifacts, refresh snapshots, internal metadata, indexes, or other service-side objects included in this calculation?
4. Has anyone observed similar behavior where two nearly identical semantic models consume different System Storage in Power BI Service?

I've exhausted the usual optimization checks (cardinality, dictionaries, partitions, relationships, VertiPaq size), so I'm trying to understand whether this is expected behavior or if there's another storage component that VertiPaq Analyzer does not expose.

Any insights or references to Microsoft documentation would be greatly appreciated.

Thank you!