How the Databricks bill works

This guide explains what a DBU is and what sets its price. It also covers why classic compute sends you a second bill for virtual machines and how serverless changes that, using Azure list prices and a worked example.

The bill in short

Databricks charges in Databricks Units (DBUs), billed per second, and your Databricks charge is the DBUs used times the price per DBU. That price depends on the type of compute, the pricing tier, the cloud and, for some products, the region. On classic compute your cloud provider also bills the virtual machines, disks and networking, while on serverless the compute is inside a higher DBU price.

DBUs and what they cost

A DBU is how Databricks measures, and prices, the work your compute does. Microsoft's pricing page describes it as "a unit of processing capability per hour, billed on a per-second usage".

  • On classic compute, every VM type has a DBU rating per hour. On Azure, a DS13 v2 VM (8 vCPUs, 56 GiB) is rated at 2 DBUs per hour. A cluster of four of them uses 8 DBUs for every hour it runs. The driver counts like any other node.
  • A SQL warehouse has a DBU rate per hour set by its size. On Azure a 2X-Small serverless warehouse is 4 DBUs per hour and a Medium is 24.
  • Other serverless products bill the DBUs your work uses.

Four things set the DBU price

  1. The product, or SKU. Each type of compute has its own price. Jobs Compute is for automated work on a cluster the job starts and stops. All-Purpose Compute is everything else on a cluster, such as notebooks, and costs more per DBU. SQL warehouses, Lakeflow pipelines, serverless compute and model serving each have their own rates.
  2. The pricing tier. The Premium tier costs more per DBU than the old Standard tier did.
  3. The cloud. Azure, AWS and Google Cloud each have their own price lists.
  4. The region. On Azure the classic jobs and all-purpose prices were the same in the regions we checked, but SQL Pro and serverless prices differ by region, per the Azure Retail Prices API: Serverless SQL was $0.70 per DBU-hour in East US 2 and $0.91 in West Europe.

Here are Azure Databricks Premium list prices in East US 2. They are list prices as of November 2026, before any discount you have negotiated, and they change. Check the pricing page for your region.

Azure Databricks, Premium tier, East US 2, pay-as-you-go. List price as of November 2026.
Workload (SKU)ComputeUSD per DBU-hour
Jobs Compute (and Jobs Compute with Photon)Classic$0.30
All-Purpose Compute (and with Photon)Classic$0.55
Jobs Light ComputeClassic$0.22
SQL Compute (classic SQL warehouse)Classic$0.22
SQL Pro Compute (pro SQL warehouse)Classic$0.55
Lakeflow Spark Declarative Pipelines: Core / Pro / AdvancedClassic$0.30 / 0.38 / 0.54
Serverless SQLServerless$0.70
Automated Serverless Compute (serverless jobs and pipelines)Serverless$0.45
Interactive Serverless Compute (serverless notebooks)Serverless$0.95

Note that the serverless prices also pay for the machines. The classic prices do not.

The Standard tier is gone on Azure

Azure Databricks used to offer two tiers, Standard and Premium. Microsoft stopped new Standard workspaces on April 1, 2026 and retired the tier on October 1, 2026. Any remaining Standard workspaces are upgraded to Premium automatically, in a phased rollout. If you still had Standard workspaces, your DBU prices rose to the Premium rates when they were upgraded: Jobs Compute, for example, was $0.15 per DBU-hour on Standard and is $0.30 on Premium.

Classic compute has two bills

With classic compute (clusters you configure, and classic or pro SQL warehouses), the machines run in your own cloud account, in what Databricks calls the classic compute plane. Two companies bill you for the same hour.

  • Databricks bills DBUs.
  • Your cloud provider bills the virtual machines, plus the disks attached to them, public IP addresses, storage and networking.

The VM charge can be about as large as the DBU charge: in Microsoft's own pricing example below, the two are almost equal. Microsoft bills the VMs from the moment they start until the cores are released, so start-up time is on the VM bill too.

AWS works the same way: AWS bills the instances, separately from the DBUs. That is why the Databricks pricing calculator says its classic estimates exclude EC2 instances.

How it shows up on the Azure bill

Azure Databricks is an Azure service, so the DBUs and the VMs both land on your Azure invoice, as separate meters. The VM side lives in the managed resource group.

When you create a workspace, Azure creates a second resource group for it, which Databricks manages. It holds the workspace storage account. When a cluster starts, its virtual machines, disks, IP addresses and network interfaces are created there too. You can see it in the Azure portal, but you can't modify it or create your own VMs in it: Databricks manages what is inside.

So in Azure cost analysis, Azure Databricks shows your DBUs, and the virtual machines, storage and networking in the managed resource group show the rest. Databricks tags these resources (cluster ID, creator and more), and the tags reach Azure cost analysis, so you can tie VM costs back to a cluster.

Serverless puts the compute in the DBU price

With serverless compute, the machines run in Databricks' own account, not yours. Databricks provisions and scales them, and their cost is part of the serverless DBU price: the Databricks pricing calculator says serverless estimates include compute infrastructure costs. Databricks says serverless has no charge for idle capacity: serverless notebooks and jobs bill while work runs (notebooks add a small session fee) and stop a few minutes after it ends. Serverless SQL warehouses work differently: they bill a fixed DBU rate per hour for their size, per running cluster, for as long as they are running, so the idle minutes before a warehouse auto-stops are billed too.

Serverless is not entirely one line, though. Microsoft notes that data transfer, NAT gateway and private endpoint hours are billed on top of the DBUs.

Classic computeServerless compute
Where the machines runYour cloud accountThe Databricks account
What you pay DatabricksDBUs, at a lower rateDBUs, at a higher rate that includes the compute
What you pay the cloud providerVMs, disks, IP addresses, storage, networkingNo VMs; some networking (data transfer, NAT gateway, private endpoints)
On Azure, where it showsAzure bill: Databricks meters plus VM, disk and network metersAzure bill: Databricks meters, plus any networking
Who sizes the machinesYou choose the VM type and countDatabricks (you choose a size for SQL warehouses)
Spot VMs and Azure savings plansApply to the VM partNot applicable

Compare the full cost of a job, DBUs plus VMs, not the DBU rate.

Photon and DBUs per hour

Photon is Databricks' faster query engine. On classic compute it is billed through the DBU rate: a VM running Photon uses DBUs at a different, higher rate than the same VM without it. Microsoft's pricing page states that "Enabling Photon will increase DBU count." On Azure the price per DBU for the Photon SKUs of Jobs and All-Purpose Compute is the same as without Photon, so the extra cost comes from the extra DBUs per hour.

The VM price does not change with Photon. So whether a run costs more overall depends on how much shorter it gets: a shorter run uses fewer VM hours and fewer hours of DBUs.

Spot VMs and commitments

There are two separate levers, one for each half of the classic bill.

  • Spot VMs lower the VM price. They use spare cloud capacity at a discount, and Azure can take them back with 30 seconds' notice. On Azure Databricks you can turn on spot instances for a cluster: by default the driver stays on-demand, the workers are spot, and Databricks tries to replace workers that are evicted. The DBU charge stays the same.
  • Pre-purchased DBUs lower the DBU price. On Azure you can buy Databricks Commit Units (DBCUs) for one or three years, for up to 37% off pay-as-you-go DBU prices. Your DBU usage draws down from the pre-purchase at list price. It covers DBUs only: Microsoft states that compute, storage and networking are charged separately.

The VM half can also use Azure savings plans for compute. And if you have a negotiated contract, your prices may differ from the list prices in this guide.

A worked example

These are round numbers to show how the sums work, not a quote. Real VM and DBU rates vary by VM type, region, tier and contract. The DBU rates are the East US 2 Premium list prices above; the VM price is a made-up round figure, close to what a DS13 v2 cost in East US 2 on pay-as-you-go.

Say a nightly job runs on a classic jobs cluster of 4 nodes (one driver and 3 workers). Each VM is rated at 2 DBUs per hour and costs $0.60 an hour. The job runs 3 hours a night, 30 nights a month.

One nightly job for a month, at the rates above
StepWorkingResult
Cluster hours3 hours × 30 runs90 hours
VM hours90 hours × 4 VMs360 VM hours
DBUs360 VM hours × 2 DBUs per hour720 DBUs
Databricks charge720 DBUs × $0.30 (Jobs Compute)$216
Cloud VM charge360 VM hours × $0.60$216
Total per monthDatabricks charge plus VM charge, before disks, IPs and storage$432

In this example the VM charge and the Databricks charge are the same, $216 each.

Now run the same work on an all-purpose cluster instead. The VMs and DBUs are the same, but the price per DBU is $0.55: 720 DBUs × $0.55 = $396, plus $216 of VMs, which comes to $612 a month for the same result.

Microsoft's own pricing page has a bigger example with real list prices: ten DS13 v2 VMs for 100 hours in East US 2 cost $598 in VMs plus $600 in Jobs Compute DBUs, or $1,100 in DBUs as All-Purpose Compute.

You can't convert the example to serverless with a formula, because serverless doesn't use your VM ratings: Databricks decides what to run it on. The way to compare is to run the work both ways and read the cost of each from the bill.

Where to see the numbers

  • The account console Usage page. Account admins can see usage across workspaces there and import ready-made usage dashboards.
  • System tables. system.billing.usage records billable Databricks usage for the whole account, and system.billing.list_prices keeps a history of each SKU's published list price. Both are about the Databricks side: they hold no cloud VM charges, and list prices are before your discount.
  • Microsoft Cost Management on Azure. Cost analysis shows the whole Azure bill, including the VMs, disks and networking in the managed resource group, and can group by the tags Databricks puts on them.

Once you can see where the money goes, the next question is what to change. On the code side, PySpark habits that make Databricks jobs cost more walks through the common ones. For everything outside the code, ClusterBite finds the changes, prices them in dollars a month and measures what each one saved (how we measure). It is read-only and works with Azure Databricks today. You pay 30% of the measured saving for the first three months after each change; the pricing page has the details.

Get started

Sources

Checked in November 2026.

Written by Marthinus Bosman for ClusterBite.