lakehouse

Databricks

Databricks is a unified analytics platform built on Apache Spark. This profile covers what it does, how it is priced, and where it stops being the right choice.

What the product actually does

Databricks runs Spark and, more recently, a SQL warehouse and a model training environment on top of object storage. The defining idea is the lakehouse: one copy of the data in cheap storage, queried by engines that understand a table format rather than by a database that owns the bytes.

The notebooks are the interface most people meet first, and they are a genuine productivity win for exploratory work. The same workspace holds the scheduled jobs, the cluster definitions and the lineage between them.

How pricing works once you are inside

The unit is the DBU — a Databricks Unit — multiplied by the cluster's instance type and consumed per second while the cluster runs. A cluster that idles still bills, and an interactive cluster left running over a weekend is the classic first invoice.

The mitigation is job clusters and autoscaling, both of which exist and both of which require deliberate configuration. The default settings are not the cheap ones.

Where it breaks down

Cost control is genuinely hard. The DBU model means the bill is a function of how people use the workspace, and usage grows naturally as more teams join. Organisations that do not enforce cluster policies spend more than they expect.

The platform is also heavy. A team with one small pipeline does not need a lakehouse, and the operational surface — workspace administration, cluster policies, Unity Catalog — is real work.

What switching costs

Moderate to high. The data is in open formats and moves; the notebooks, jobs, cluster configurations and governance metadata do not.

Sources

Databricks public documentation and pricing page, read October 2026.