What is chargeback?

Chargeback is the FinOps practice of billing cloud costs back to the teams that incurred them, so each team's Databricks spend lands in its own budget. It requires attribution both sides trust: shared clusters, serverless usage, and platform overhead all need a defensible split before a charge will survive scrutiny.

The mechanics are accounting; the hard part is the attribution underneath. Dedicated, tagged compute charges back cleanly. Databricks environments also produce the other kind: an all-purpose cluster shared by four teams, serverless SQL running under one service principal, platform overhead that belongs to nobody’s job. Each needs an allocation policy (by observed usage, fixed split, or proportional share) that the charged teams accept.

Most organizations get there in two steps: run showback first, publishing each team’s spend without moving budgets, and switch to chargeback once the numbers have survived a few months of questions. The attribution work is identical; only the consequences change.

The test of a working chargeback model is the dispute path. If a team can see which rule produced their charge and how shared costs were split, disagreements are policy conversations. If they can’t, the model erodes one contested invoice at a time.

See these numbers in your own workspaces.