Free cost calculator
Databricks cost calculator
Estimate your monthly Databricks bill from compute type, cluster size, DBU rate, cloud VM cost, spot instances and Photon.
How this is calculated
DBUs = nodes (workers + driver) × DBUs per node-hour × Photon multiplier × hours × days.
Databricks cost = DBUs × price per DBU.
Cloud VMs (classic compute only) = worker node-hours × VM price × (1 − spot share × spot discount) + driver node-hours × VM price. The driver stays on-demand. Serverless types have no separate VM cost.
Not included: cluster start-up time, storage (S3, ADLS, GCS), networking and NAT, Unity Catalog or model serving charges, and committed-use discounts.
Guides to cut this bill
- Cutting a Databricks bill →
- Spark executor memory and OOM →
- Fixing Spark data skew →
- Unity Catalog in practice →
Comparing platforms instead? Try theside-by-side cost estimatoror see all calculators.
FAQ
What is a DBU and how much does it cost?
A Databricks Unit (DBU) is the unit Databricks bills compute in. Each instance type uses a fixed number of DBUs per hour, and the price per DBU depends on the workload type and tier: on AWS Premium in the US, Jobs Compute is around $0.15, All-Purpose around $0.55 and SQL Serverless around $0.70. Check your own rates, which vary by cloud, region and contract.
Do I pay for the cloud VMs as well?
On classic compute, yes: your cloud provider bills the VMs separately from the DBUs, and the VM cost is often as large as the DBU cost. Serverless SKUs include the compute in the DBU price, so there is no separate VM bill.
Why is All-Purpose compute so much more expensive than Jobs?
All-Purpose clusters are for interactive notebooks and cost roughly three to four times more per DBU than Jobs clusters. Running scheduled production work on All-Purpose clusters is one of the most common causes of a high Databricks bill.
Does Photon make jobs cheaper?
Photon raises the DBU rate per node-hour (often about double), so it only saves money when it makes the job finish faster by more than that multiple. Test it on your real workload and compare total cost, not just run time.