Databricks DBU Cost: DBU Pricing per Hour and Cost per DBU for Every Compute SKU

The rate card only becomes a budget once you multiply it by consumption, and consumption is where the tier decision either costs nothing or doubles the bill. Three worked examples, all using the Azure East US rates above and all counting the Databricks software charge only.

A small pipeline team. Say 2,000 Jobs Compute DBU-hours a month. On Standard that is $300 a month. On Premium it is $600. The tier delta alone is $3,600 a year, and that is the real price of Unity Catalog for this team, even though Unity Catalog itself has no price.

A mid-size platform. Say 20,000 Jobs Compute DBU-hours plus 5,000 All-purpose DBU-hours a month. Standard: $3,000 plus $2,000, so $5,000. Premium: $6,000 plus $2,750, so $8,750. The delta is $3,750 a month, or $45,000 a year, and notice that almost all of it comes from the jobs line rather than the notebooks line.

An analytics-first organization. Say 15,000 SQL Analytics DBU-hours a month and little else. Standard: $3,300. Premium: $3,300. The tier delta is zero. This team gets Unity Catalog, access controls and audit logs for no incremental DBU cost at all, which is the opposite of what a flat 1.5x rule of thumb would have predicted.

The lesson is that the governance tier is priced against your workload mix, not against your headcount or your data volume. Before you accept or reject Premium on cost grounds, pull your own consumption by SKU and apply the multipliers in the table. Two organizations with identical Databricks bills can face wildly different upgrade costs.

One caveat that matters for every number on this page: the DBU rate is not your bill. The virtual machines, disks and storage the cluster runs on are billed separately by the cloud provider, and for many workloads that infrastructure line is comparable to or larger than the DBU line. If cloud and SaaS spend across the business is the actual problem rather than Databricks specifically, that is a job for a read-only cloud cost management platform rather than a per-SKU rate table.

We build Datatrail, so read this section knowing that. We do not sell cost management and we will not claim to cut your Databricks bill by a percentage.

What we do have an opinion about is the one category of DBU waste that rate tuning cannot reach. Cluster right-sizing, auto-termination and moving interactive work onto Jobs Compute are all well covered and you should do all three. They make each hour cheaper. They do nothing about hours that should not run at all.

In most estates of any age there is a tail of scheduled work that still runs every night to refresh tables nobody reads. A dashboard was retired, a model was replaced, an experiment ended, and the pipeline that fed it was never switched off because nobody could prove it was safe to switch off. Every one of those runs bills DBUs at full rate forever. Finding them is not a cost problem, it is a lineage problem: you need to know, for each table, what actually reads it downstream, all the way into the BI layer where the consumption really ends.

Datatrail connects to Databricks with a read-only role, parses query history and dbt artifacts into column-level lineage that refreshes itself, and follows the column past the edge of the workspace into the models and dashboards that depend on it. The same graph drives impact analysis, so before you delete a job you can see exactly what reads its output and what would break. The Databricks-specific detail, including how we fill the gaps Unity Catalog documents, is in Databricks data lineage.

If you are comparing the wider field rather than a single platform, the category guides are data lineage tools and data observability tools.

Every rate on this page came from the Azure Retail Prices API, which is public, needs no key and returns the same list prices Microsoft bills against. We filtered it to Azure Databricks, East US, Consumption pricing, on 9 September 2026. The call returned 43 SKUs, each carrying a retail price, a unit of measure and an effective start date.

BASE=https://prices.azure.com/api/retail/prices
curl "$BASE?\$filter=serviceName%20eq%20'Azure%20Databricks'"

Two limits are worth stating plainly. These are Azure list prices for one region, and Databricks on AWS and Google Cloud is billed by Databricks directly on a different card, so do not carry these figures across clouds. And list is not what a large account pays: committed-use discounts are negotiated and are not visible in any public feed.

What the feed does give you is a dated, primary source, which is more than a rate quoted in an article can offer. If you are checking our work, the same query run today should return the same numbers, and if it does not, ours are the ones that are stale.