Azure Databricks cost: DBUs on top of the underlying VMs

Quick answer

Databricks cost is two parts: the DBU rate (Databricks Unit, priced per workload type and tier) plus the Azure VM cost of the cluster running it. So a cluster costs the VMs plus a DBU surcharge. Use job clusters over all-purpose, Spot VMs for workers, and autoscaling with autotermination to control both parts.

Databricks on Azure has a two-part bill that surprises teams who look only at one side. You pay Azure for the virtual machines in the cluster, and you pay Databricks a DBU rate for the platform on top. Both scale with cluster size and runtime, so the levers reduce both together.

DBUs plus VMs

A DBU is Databricks unit of processing, consumed per hour based on cluster size, and priced by workload type (jobs, all-purpose, SQL) and tier (Standard, Premium). On top of the DBU charge you pay the standard Azure cost of the underlying VMs. A cluster therefore costs its VMs plus a DBU surcharge proportional to its size and uptime.

Job clusters beat all-purpose

Cluster type	DBU rate	Best for
Job cluster	Lower	Scheduled and automated jobs
All-purpose cluster	Higher	Interactive notebooks, exploration

Job clusters, created for a job and terminated after, carry a lower DBU rate than all-purpose interactive clusters and do not sit idle. Running scheduled work on job clusters rather than a shared all-purpose cluster cuts both the DBU rate and idle time.

Spot workers and autotermination

Put worker nodes on Spot (preemptible) VMs to cut the VM portion sharply for fault-tolerant jobs, keeping the driver on-demand. Enable autoscaling so clusters size to the work and autotermination so idle interactive clusters shut down rather than billing DBUs and VMs for nothing. The Spot and autoscaling logic is the same as EMR cost optimization.

Controlling Databricks cost

Run automated work on job clusters, put workers on Spot, enable autoscaling and autotermination, right-size clusters to the job, and use the appropriate tier rather than Premium everywhere. Because cost is VMs plus DBUs, reducing cluster size and uptime lowers both.

FAQ

How is Azure Databricks priced?

Two parts: a DBU (Databricks Unit) rate consumed per hour based on cluster size and priced by workload type and tier, plus the standard Azure VM cost of the cluster. A cluster costs its VMs plus a DBU surcharge, both scaling with size and runtime.

What is a DBU?

A Databricks Unit, the platform's unit of processing consumed per hour based on cluster size. DBU rates vary by workload type (jobs, all-purpose, SQL) and tier (Standard, Premium). It is charged on top of the underlying Azure VM cost.

Are job clusters cheaper than all-purpose clusters?

Yes. Job clusters carry a lower DBU rate than all-purpose interactive clusters and terminate after the job rather than sitting idle. Running scheduled and automated work on job clusters reduces both the DBU rate and idle time.

Can I use Spot VMs with Databricks?

Yes, for worker nodes on fault-tolerant jobs, which cuts the VM portion of the bill sharply. Keep the driver node on-demand for stability. Because workers reschedule when reclaimed, the risk is low and the saving significant.

How do I reduce Databricks cost?

Run automated work on job clusters, put workers on Spot VMs, enable autoscaling and autotermination so idle clusters shut down, right-size clusters to the job, and use the appropriate tier rather than Premium by default. Reducing cluster size and uptime lowers both DBUs and VM cost.

Does C3X estimate Databricks cost?

C3X prices the underlying Azure VMs of a Databricks cluster from your Terraform. The DBU surcharge and cluster runtime depend on usage, which you provide as assumptions for a complete estimate, similar to the EMR surcharge model.

What to do next

Price the VMs behind your clusters before you launch them. C3X reads your Terraform and prices your resources against a live catalog. Start with the quickstart.