Databricks Cost Calculator
Model Databricks DBUs and the cloud VMs underneath them together, compare All-Purpose against Jobs compute on the same workload, and see what Photon and spot instances change. Uses your own DBU rate and instance price.
Interactive notebooks. The expensive one.
| Databricks DBUs7,200 DBUs at $0.56 — 85.3% of the real total | $3,996.00 |
| Cloud instances underneath9 nodes (8 workers + 1 driver) x 400 hours | $691.20 |
| Total | $4,687.20 |
Move this workload to Jobs Compute
Other levers
What this model assumes
- The cloud VMs are the point. Databricks bills DBUs; your cloud provider separately bills the instances those DBUs ran on. The second invoice is usually the larger one, and a calculator that prices only DBUs understates the platform by roughly two thirds.
- Compute types are modelled as ratios, not prices. Databricks does not publish machine-readable DBU rates, and they vary by tier, cloud and contract. What does not vary is the shape — All-Purpose costs several times Jobs for the same DBU — so the model is built on that and takes your actual rate as input.
- The driver node is counted. It runs and bills like a worker, and it cannot go on spot.
- Photon is modelled as faster, not just pricier. It consumes DBUs at a higher rate but finishes sooner, so the net can go either way. The speed-up is an assumption — measure yours before trusting the direction.
- Storage, networking and Unity Catalog are excluded. So is serverless, which prices on a different basis entirely.
- EC2 on-demand (Linux, shared tenancy) — the default instance price only — source, read 4 Aug 2026
US East (N. Virginia) list prices, excluding any negotiated or committed-use discount. Your effective rate is lower if you have one.
Frequently asked questions
What is a DBU and what does it actually cost?
A Databricks Unit is a normalised measure of processing consumed per hour. Databricks bills DBUs; your cloud provider separately bills the virtual machines those DBUs ran on. The second invoice is usually larger than the first, so budgeting from the Databricks bill alone typically captures around a third of the real cost.
Why is All-Purpose compute so much more expensive than Jobs compute?
Because it is priced for interactive use — a cluster you attach a notebook to and leave running. The DBU rate is several times the Jobs rate for identical work. Any workload on a schedule belongs on Jobs compute, and moving it is a configuration change rather than a rewrite. Interactive clusters left running overnight are the classic Databricks bill shock.
Does Photon save money on Databricks?
It can go either way. Photon consumes DBUs at roughly twice the rate but finishes the same work faster, so the net depends entirely on the speed-up your workload actually gets. For scan-heavy SQL it is frequently a win; for workloads dominated by Python UDFs it often is not. Measure the speed-up before assuming the direction.
Do spot instances reduce Databricks costs?
They reduce the cloud half only — DBU rates are the same whether the node is spot or on-demand. Since the cloud half is usually the larger one, that is still a substantial saving. The driver node cannot go on spot without risking the whole cluster, so it stays on-demand.
Why does this calculator ask for my DBU rate?
Databricks does not publish DBU rates as machine-readable data, and the figure varies by tier, cloud and contract. What does not vary is the shape — All-Purpose costs several times Jobs for the same DBU — so this model is built on those ratios and takes your actual rate as an input rather than publishing one nobody verified.
Two invoices, one workload, no single view.
Finitizer joins Databricks consumption to the underlying cloud spend so a job has one cost, not two — and applies the same treatment to BigQuery, Snowflake and Redshift across the whole data estate.