Databricks cost glossary

The vocabulary of Databricks cost work, defined precisely — with the numbers that matter. Start with the definition; take the deep dive when you need the mechanics.

Compute

All-purpose compute

All-purpose compute is Databricks compute built for interactive work: notebooks, exploration, debugging. Clusters stay up between commands so people can iterate, and the DBU rate runs roughly 3–4× jobs compute on the same tier. The premium buys interactivity — capability that bills whether or not anyone is using it.

FinOps practice

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.

FinOps practice

Cost center

A cost center is an organizational unit — a team, department, or project — that costs are assigned to for budgeting and accounting. In Databricks, cost centers rarely map one-to-one to workspaces or clusters, so attribution (tags, identity mappings, allocation rules) is what actually connects the bill to them.

Pricing & billing

DBU (Databricks Unit)

A DBU (Databricks Unit) is the unit Databricks uses to meter compute consumption. Every running cluster, SQL warehouse, and serving endpoint consumes DBUs per hour based on its size and configuration. The Databricks bill is DBUs consumed multiplied by the rate for that compute type, plus the underlying cloud infrastructure cost.

Pricing & billing

DBU pricing

DBU pricing is the per-unit rate Databricks charges for each DBU consumed. Rates vary by compute type, pricing tier, and cloud: on AWS Premium list prices (June 2026), jobs compute runs about $0.15 per DBU, all-purpose compute $0.55, and SQL Serverless $0.70. The same work can cost 3–4× more purely by compute type.

Compute

Jobs compute

Jobs compute is Databricks compute provisioned for scheduled, non-interactive workloads: a cluster spins up for the job run and terminates when it finishes. It carries the lowest DBU rate of the classic compute types — about $0.15 per DBU on AWS Premium — because the capacity exists only for the duration of the work.

Compute

Photon

Photon is Databricks' vectorized, C++-based query engine, available on clusters and SQL warehouses. Enabling it consumes DBUs at a higher rate for the same instance — often around 2×, varying by SKU — in exchange for faster execution. It lowers cost only when wall-clock time drops by more than that premium.

FinOps practice

Showback

Showback is the FinOps practice of reporting each team's cloud spend without billing it: visibility first, accountability through transparency rather than invoices. Most organizations run showback as the stage before chargeback — the same attribution and team-level rollups, minus the budget transfer, while teams learn to trust the numbers.

Databricks platform

System tables

System tables are Databricks-managed tables in Unity Catalog that record platform metadata: billing usage, cluster and warehouse configurations, job runs, query history. The cost-relevant ones, like system.billing.usage, retain a rolling 365-day window and are the raw source behind any Databricks cost analysis — native or third-party.

Databricks platform

Unity Catalog

Unity Catalog is Databricks' unified governance layer: one place where catalogs, schemas, tables, models, and permissions are defined across workspaces. For cost work it matters twice over — the system tables that record billing and usage live in Unity Catalog, and the tags attribution relies on are governed through it.

Definitions are the easy part.

Seeing these concepts in your own bill is what LakeSentry does. The free tier covers unlimited workspaces with three months of history. No card required.

Start Free