Articles

Azure Databricks Pricing: Full Cost Breakdown (2026)

This breakdown covers what each Azure Databricks compute type costs per DBU in 2026, how the VM and storage charges add up, what the Standard tier retirement changed, how serverless compares with classic compute, and what three common workloads cost each month.

Team Eon
Written by
Team Eon
Published: 
Nov 13, 2026
0
 min read

Quick Summary

  • Your bill has two halves: Databricks Units (DBUs) for the software, plus Azure VMs, storage, and networking for the infrastructure under classic compute.
  • Premium is the only tier now. Standard workspaces began moving to Premium automatically on October 1, 2026, which doubles the Jobs Compute rate from $0.15 to $0.30 per DBU.
  • Serverless rates ($0.45 to $0.95 per DBU in East US) include the compute, so they look pricier per DBU but have no VM line or idle cluster time.
  • Pre-purchasing Databricks Commit Units saves 4% to 37% on DBUs only. VM discounts (Spot, reservations, savings plans) apply to classic compute only.
  • Idle time drives the biggest bills: an All-Purpose cluster left running 24/7 costs roughly 3.7 times one that shuts down after the workday.

‍

The first Azure Databricks invoice tends to surprise someone. The DBU charge looks reasonable on its own. Then the VM charges arrive on a separate line: a cluster that nobody shut down ran all weekend, and the storage account under the lakehouse keeps growing.

Azure Databricks pricing is easy to read in pieces and hard to read as one number. Microsoft publishes a rate per DBU for each workload, a DBU count per VM size, separate VM prices, and a handful of add-ons and discounts on top. Few teams add them up before the bill does it for them.

Every rate in this breakdown comes from Microsoft’s official pricing pages, checked November 2026, so you can redo the math for your region and plug in your own cluster sizes.

Azure Databricks pricing by compute type

Azure Databricks pricing runs $0.22 to $0.95 per DBU-hour on the Premium tier in East US, depending on the compute type. 

Classic compute types (All-Purpose, Jobs, SQL Classic, SQL Pro, and classic Lakeflow pipelines) add VM charges on top. Serverless types include compute.

These are pay-as-you-go Premium rates from Microsoft’s pricing page for Azure Databricks, East US, as of November 2026.

Compute type What it’s for Premium DBU rate (East US) VMs billed separately?
All-Purpose Compute Interactive notebooks and shared clusters for exploration and development $0.55 Yes
Jobs Compute Scheduled, automated jobs on clusters that start and stop with the job $0.30 Yes
Jobs Light Compute Simple automated jobs on a lighter Spark runtime with fewer platform features $0.22 Yes
SQL Classic (SQL Compute) SQL warehouses running in your subscription, basic feature set $0.22 Yes
SQL Pro SQL warehouses in your subscription with the full Databricks SQL feature set $0.55 Yes
Serverless SQL SQL warehouses Databricks runs and scales for you $0.70 No
Lakeflow Spark Declarative Pipelines (classic) Pipelines on your own clusters: Core, Pro, or Advanced edition $0.30 / $0.38 / $0.54 Yes
Serverless jobs and pipelines (Automated Serverless Compute) Lakeflow Jobs and pipelines on serverless compute $0.45 No
Serverless notebooks (Interactive Serverless Compute) Notebooks on serverless compute $0.95 No

A few rules sit behind those rates:

  • Photon costs DBUs: Photon-enabled All-Purpose and Jobs clusters keep the same rate per DBU, but enabling Photon increases the DBU count per hour.
  • SQL warehouses bill by size: Each warehouse size uses a fixed DBU count per hour: 4 for 2X-Small, 6 for X-Small, 12 for Small, 24 for Medium, 40 for Large, and up to 528 for 4X-Large.
  • Rates change by region: Central US matches East US on most rows, but serverless jobs cost $0.47 and serverless notebooks $1.00 there. In West Europe, Serverless SQL is $0.91 and SQL Pro is $0.72.
  • Serverless still has extras: Serverless workloads also bill data transfer, NAT gateway, and Private Endpoint hours, on top of DBUs.

Other meters on an Azure Databricks bill

Data engineering and SQL make up most bills, but the same pricing page lists meters for AI and database workloads:

  • Model Training: $0.65 per DBU-hour in East US.
  • Serverless Real-Time Inference (model serving): $0.07 per DBU-hour.
  • Database Serverless Compute (Lakebase): $0.26 per DBU-hour, at a limited-time promotional price that saves 50% through January 31, 2027.
  • Genie Code: The AI coding assistant moved to pay-as-you-go billing on July 8, 2026, with a free monthly allowance per user. Genie One and Genie Agents usage by people (service principals excluded) is free through January 31, 2027.

Managed disks, IPs, and NAT gateways on classic compute

Classic clusters create resources in a managed resource group in your subscription. Azure bills those separately as managed disks, blob storage, and public IP addresses.

Workspaces on the default VNet with secure cluster connectivity also get a NAT gateway that Azure Databricks creates for outbound traffic, and that gateway incurs additional cost. These lines are small next to DBUs and VMs, but they show up every month a workspace exists.

What makes up an Azure Databricks bill

On classic compute, DBUs are only half the bill. Azure also bills you for the virtual machines your clusters run on, plus the managed disks, blob storage, and public IP addresses those clusters use.

Serverless compute folds the machines into the DBU rate instead, because it runs in a compute plane Databricks manages, with no compute resources provisioned in your cloud account.

Your data sits in Azure storage accounts you pay for separately. Add networking, and a typical monthly bill has four parts: DBUs, VMs, storage, and network charges.

Each VM size burns a set number of DBUs an hour (a D8ds v5 uses 2). Take one hour of a Jobs Compute cluster in East US with one driver and four workers, all D8ds v5 VMs:

  1. DBUs: 5 VMs × 2 DBUs = 10 DBUs an hour, × $0.30 = $3.00.
  2. VMs: 5 × $0.452 an hour (Linux pay-as-you-go) = $2.26.
  3. Hourly compute total: $5.26, before storage and networking.

The VMs make up 43% of that hour’s cost, so the DBU rate alone accounts for barely more than half of it.

Standard tier retirement: What changed on October 1, 2026

Azure Databricks retired its Standard tier in two steps. New workspaces had to be Premium starting April 1, 2026, and starting October 1, 2026, remaining Standard workspaces are being upgraded to Premium automatically in a phased rollout.

Microsoft says you won’t lose access to any capabilities, and Premium adds Unity Catalog, Databricks SQL, serverless compute, and Lakeflow pipelines. The rates are the part teams feel. 

The same DBU costs this much on each tier in East US:

Compute type Standard (retired) Premium Increase
All-Purpose Compute $0.40 $0.55 37.5%
Jobs Compute $0.15 $0.30 100%
Jobs Light Compute $0.07 $0.22 About 214%

Jobs Compute is where the change hurts most. A nightly ETL pipeline that used 600 DBUs a month on Standard paid $90 for them. On Premium, the same 600 DBUs cost $180, before any change to the VM line.

Older guides still list $0.15 per DBU for Jobs Compute, which is the retired Standard rate. A budget built on that figure covers about half of the real DBU charge.

How does Azure Databricks billing work?

Azure Databricks billing works by metering every running compute resource per second, multiplying its DBU count by your workload’s rate, and adding the Azure infrastructure under it. 

On a classic cluster, it runs in this order:

  1. You pick a compute type: Starting a notebook cluster puts you on the All-Purpose rate. Running the same code as a scheduled job on a job cluster puts you on the Jobs rate.
  2. You pick VM sizes: The VM size sets the DBUs per hour (0.75 for a DS3 v2, 2 for a D8ds v5, 5.5 for an E16ds v5) and the separate VM price.
  3. The cluster runs: DBU and VM charges accrue from the moment the cluster is up until it terminates, including startup time.
  4. Idle time counts: All-Purpose clusters keep billing until auto-termination kicks in. Idle compute keeps accruing DBU and cloud instance charges for the whole inactivity period.
  5. Storage and network bill on their own meters: Tables in Azure Data Lake Storage Gen2 bill per GB per month, plus read and write operations.

Serverless skips steps 2 and 3. You don’t choose VMs, Databricks allocates the compute, and you pay the serverless DBU rate for the DBUs your workload uses. To see what each workload costs after the fact, query the system.billing.usage table (it can lag by up to 24 hours).

Serverless vs. classic compute: Cost for the same job

The per-DBU rate makes serverless look expensive, but classic rates leave out the VMs.

Take the five-node D8ds v5 job cluster from earlier. It costs $5.26 an hour in DBUs and VMs for 10 DBUs, which works out to about $0.53 per DBU once the VMs are included. Serverless jobs cost $0.45 per DBU with the compute included.

Factor Classic Jobs Compute Serverless jobs
Rate $0.30 per DBU, plus VMs $0.45 per DBU, compute included
All-in cost per DBU (D8ds v5 example) About $0.53 $0.45
Startup You pay for cluster startup minutes Fast in performance-optimized mode; 4 to 6 minutes in standard mode
VM discounts (Spot, reservations, savings plan) Yes No
Pre-purchase (DBCU) discount Yes, on DBUs Yes
Tuning You choose VM sizes, autoscaling, and pools Databricks sizes and scales the compute
Extra charges Managed disks, public IPs, NAT gateway Data transfer, NAT gateway, Private Endpoint hours

Serverless and classic don’t burn DBUs at the same rate for the same work, though. For this job, serverless costs the same as classic if it uses about 11.7 DBUs per hour of runtime. Jobs that stay under that are cheaper on serverless; heavier ones are cheaper on classic.

The only way to know is to run the job both ways and compare the two runs in the billing table.

Two settings move that result:

  • Standard performance mode: Serverless jobs and pipelines can run in standard mode, which uses fewer DBUs in exchange for a 4 to 6 minute startup. Microsoft says it can reduce costs by up to 70% compared with performance-optimized mode. It isn’t available for notebooks.
  • Spot workers on classic: Running the four workers on Spot VMs (about $0.095 an hour instead of $0.452 for a D8ds v5 in East US) cuts the cluster’s VM line by about 63%, since the driver stays on a regular VM. Per worker, Spot is about 79% off. Serverless can’t match that, but Spot VMs can be evicted mid-job.

Notebooks follow a different pattern. An All-Purpose cluster on E8ds v5 VMs costs about $0.76 per DBU all-in, so serverless notebooks ($0.95) run about 25% more per DBU. 

Serverless also has no charge for idle capacity, while a classic cluster bills through its whole idle timeout. For a team that walks away from clusters for hours at a time, serverless often comes out ahead.

Ways to pay for Azure Databricks

Azure Databricks has one discount for the DBU half of the bill and several for the VM half.

Pay-as-you-go

Pay-as-you-go is the default. You pay the list DBU rate per second of use, plus pay-as-you-go VM prices for classic compute, with no commitment.

Databricks Commit Unit (DBCU) pre-purchase

You can save up to 37% over pay-as-you-go DBU prices by pre-purchasing Databricks Commit Units (DBCUs) for one or three years. Usage draws down from the pool at list DBU prices, and the discount grows with the size of the commitment.

Term Smallest plan Largest plan
1 year 12,500 DBCUs for $12,000 (4% off) 2,000,000 DBCUs for $1,340,000 (33% off)
3 years 37,500 DBCUs for $35,250 (6% off) 6,000,000 DBCUs for $3,780,000 (37% off)

The 37% figure only applies to the largest 3-year plan. Mid-size plans land between those points: 100,000 DBCUs for one year costs $89,000 (11% off), and 300,000 DBCUs for three years costs $261,000 (13% off).

Microsoft’s pre-purchase rules are flexible about where the units go. They apply to all Databricks workloads and tiers, they don’t expire hourly the way VM reservations do, and you don’t assign them to a workspace. They cover DBU usage only, though. Compute, storage, and networking stay at their own rates.

Azure savings plan and Reserved Instances

The VMs under classic clusters are regular Azure VMs, so VM discounts apply to them. For a D8ds v5 in East US, an Azure savings plan for compute cuts the hourly rate by about 32% (1 year) or 54% (3 years). A Reserved VM Instance cuts it by about 41% (1 year) or 62% (3 years).

Both pay off only for VMs that run most of the month, such as an always-on cluster or a steady pool. Job clusters that run two hours a night rarely use enough hours to fill a reservation.

Spot VMs

Spot VMs sell unused Azure capacity at deep discounts (about 79% off for a D8ds v5 in East US). They suit worker nodes on fault-tolerant batch jobs. Keep the driver on a regular VM, since evicting the driver ends the job.

Azure Databricks add-ons

Two add-ons sit on top of Premium, and both bill as a percentage of your Azure Databricks spend in the workspace where you turn them on:

Add-on Price What it includes
Enhanced Security and Compliance 10% of list price added to Azure Databricks product spend in the selected workspace Enhanced Security Monitoring and the Compliance Security Profile (required for PCI-DSS, recommended for HIPAA)
Mission Critical 30% of list price (50% off until June 30, 2027) Managed Disaster Recovery plus Enhanced Security and Compliance

Managed Disaster Recovery, which replicates data, Unity Catalog, and workspace assets between regions, has its own meter on top: $0.02 per GB processed in East US. Cross-region data transfer bills separately.

Enhanced Security and Compliance adds about $18 a month to $180 of DBUs. On a workspace spending $50,000 a month on Databricks, it adds $5,000.

Azure Databricks monthly cost examples: 3 workloads

These three bills use East US rates on Premium, pay-as-you-go. The usage figures are assumptions for illustration, so swap in your own. Storage uses the Azure Data Lake Storage Gen2 hot tier at $0.021 per GB a month. Networking (the NAT gateway and any cross-region traffic) isn’t modeled, since it depends on your setup.

Daily ETL pipeline on Jobs Compute

A data engineering team runs one nightly pipeline on a job cluster: one driver and four workers, all D8ds v5, for two hours a night, 30 nights a month.

Component Assumption Monthly cost
DBUs 10 DBUs/hour × 60 hours = 600 DBUs at $0.30 $180.00
VMs 5 × D8ds v5 at $0.452/hour × 60 hours $135.60
Storage 5 TB (5,120 GB) in ADLS Gen2 hot tier $107.52
Total About $423

Moving the four workers to Spot VMs drops the VM line to about $50 and the total to about $338.

Data science team on All-Purpose clusters

A data science team shares one All-Purpose cluster with a driver and three workers on E8ds v5 VMs. They use it eight hours a day on 22 workdays, and it shuts down after an hour of inactivity, so it bills nine hours a day.

Component Assumption Monthly cost
DBUs 11 DBUs/hour × 198 hours = 2,178 DBUs at $0.55 $1,197.90
VMs 4 × E8ds v5 at $0.576/hour × 198 hours $456.19
Storage 2 TB (2,048 GB) in ADLS Gen2 hot tier $43.01
Total About $1,697

Clear the auto-termination box and leave the same cluster running 24/7, and the compute alone comes to about $6,098 a month (about 3.7 times as much). Shortening the idle timeout from 60 minutes to 15 saves the team roughly another $138 a month.

Serverless SQL for a BI team

An analytics team runs dashboards on a Small Serverless SQL warehouse (12 DBUs an hour) that auto-stops outside business hours. It runs 10 hours a day on 22 workdays.

Component Assumption Monthly cost
DBUs (compute included) 12 DBUs/hour × 220 hours = 2,640 DBUs at $0.70 $1,848.00
Storage 10 TB (10,240 GB) in ADLS Gen2 hot tier $215.04
Total About $2,063

If dashboard traffic forces a step up to Medium (24 DBUs), the DBU line doubles to $3,696. A Small warehouse that never stops runs about $6,132 a month in DBUs.

In all three examples, run time and idle time moved the bill much more than storage did. The DBU line was the biggest single cost every time, and on classic compute the VM line made up roughly 28% to 43% of the compute cost.

Should you use Azure Databricks? Matching workloads to cost

Azure Databricks is usually the most direct way to run Spark, Delta Lake, and SQL analytics on data already in Azure storage. Cost depends far more on how you run it than on the rate card.

Azure Databricks fits well if you:

  • Run scheduled data pipelines on Delta Lake tables in ADLS and can use job clusters (and Spot workers) for them.
  • Have SQL and BI users who can share a serverless warehouse that stops when nobody’s querying.
  • Already budget and buy through Azure and want Databricks spend on the same invoice.

Look at alternatives or start smaller if you:

  • Only need a SQL warehouse for a few dashboards and don’t use Spark, notebooks, or ML.
  • Can’t control cluster sprawl yet (no cluster policies, no auto-termination rules), since idle All-Purpose clusters are the fastest way to overspend.
  • Run only a few small transformations a day, where cluster startup minutes can cost more than the work itself.

How to estimate your Azure Databricks bill

  1. List workloads by compute type: Separate scheduled jobs, interactive notebooks, SQL dashboards, and pipelines. Each has its own rate.
  2. Pick VM sizes and hours for classic work: Multiply DBUs per VM by VMs per cluster by hours a month. Add idle time for All-Purpose clusters.
  3. Price both halves: DBUs at the workload rate, plus VMs at Azure’s pay-as-you-go rate for your region. Use the Azure pricing calculator with Premium selected.
  4. Add storage and networking: GB in ADLS by tier, read and write operations, and the NAT gateway if you use the default VNet.
  5. Test one job on serverless: Run a representative job both ways, compare costs in system.billing.usage, then decide which workloads move.

Azure Databricks cost optimization best practices

These habits keep a Databricks bill close to the estimate:

  • Use job clusters for scheduled work: The same notebook costs $0.55 per DBU on an All-Purpose cluster and $0.30 as a job. Move anything on a schedule to Jobs Compute.
  • Set auto-termination on every All-Purpose cluster: The allowed range is 10 to 10,000 minutes. We recommend 15 to 30 minutes for most interactive clusters.
  • Use cluster policies: Cap VM sizes, worker counts, and idle timeouts so one notebook can’t spin up a 20-node cluster.
  • Put workers on Spot: Fault-tolerant batch jobs can cut worker VM costs by about 79%. Keep the driver on a regular VM.
  • Try standard mode for serverless jobs: Jobs that can wait 4 to 6 minutes to start use fewer DBUs.
  • Compact small files: Many small Parquet or Delta files force extra tasks and longer runs. OPTIMIZE compacts them, and well-sized files mean shorter runs on every read.
  • Commit after you measure: Buy DBCUs once you’ve seen two or three months of real usage, sized to the baseline you’d use anyway.

Several Azure cost optimization tools track Databricks spend next to the rest of your Azure resources. The habits that keep it down are the same cloud cost optimization strategies that hold across every Azure service.

Protecting the data behind your Databricks workspace

An Azure Databricks bill pays for compute. The data that compute reads lives somewhere else: the storage accounts holding your Delta tables, the VMs and SQL Server databases feeding them, and the raw files landing every night.

Delta Lake can roll a table back with time travel, but that history lives in the same storage account as the table. A deleted storage account, a mistaken VACUUM, or a compromised subscription takes the history with it.

That’s a posture problem first. Eon brings Cloud Backup Posture Management (CBPM), the category we created, to your Azure environment. It discovers storage accounts, VMs, and SQL Server on Azure VMs continuously, classifies each by data type and environment, and applies backup policy automatically. A new storage account doesn’t sit unprotected until someone tags it.

Backups land in logically air-gapped vaults isolated from production. When something breaks, you restore the objects or the exact database records you need, with no full-environment rebuild. NETGEAR cut a 10TB SQL Server restore from 24 hours to under three that way.

Compared with native cloud backup, Eon reduces backup storage costs by 30% to 50% through global deduplication, and pricing is usage-based per GB per month with no separate retrieval or egress fees. On the native side, Azure Backup pricing still charges the retrieval and egress fees that Eon doesn’t.

Backup data can also feed analytics. Eon keeps backups in open formats that query engines can read, so historical records don’t need a separate copy pipeline. Those records stay queryable through Microsoft Fabric with OneLake, straight from the backup.

Those open formats pay off at ingestion too. Rise, an AdTech platform, uses Eon to compact its stream of 200 billion events a day as it lands in Apache Iceberg, which keeps the lake at roughly half the size of the raw data.

The bottom line on Azure Databricks costs

Azure Databricks pricing comes down to three decisions: which compute type each workload runs on, how long clusters stay up, and which half of the bill you discount. Put scheduled work on job clusters or serverless, shut down idle clusters fast, and commit to DBCUs only once you know your baseline.

The rate card is also higher than it was a year ago for many teams. For anyone who was on Standard, every Jobs Compute DBU now costs twice what it did, so last year’s budget is likely short.

Every table those clusters touch sits in Azure storage, and if you can’t prove that storage is protected, you don’t have coverage.

Do you know which Azure storage accounts behind your analytics are protected right now? Book a demo to see your backup posture across every subscription.

Frequently asked questions

Is Azure Databricks free?

No, but it has a free trial. The Azure Databricks trial gives you free Premium DBUs for 14 days. It needs a pay-as-you-go Azure subscription, and Azure still bills the VMs your clusters run on during the trial.

How is Azure Databricks different from Databricks on AWS or Google Cloud?

The main difference is billing. Azure Databricks is a first-party Azure service, so Microsoft sets the price and bills it on your Azure invoice under your Azure subscription terms. On AWS and Google Cloud, Databricks bills DBUs and the cloud provider bills the instances. The Premium tier on Azure corresponds to the Enterprise tier on AWS and GCP.

Is Databricks expensive?

Databricks can be expensive when clusters sit idle or run on the wrong compute type, but the rates themselves are moderate. A nightly two-hour pipeline on a five-node job cluster costs about $316 a month in DBUs and VMs in East US. The same hardware left running 24/7 as an All-Purpose cluster costs over $5,000.

What does one DBU cost on Azure Databricks?

One DBU costs between $0.22 and $0.95 on Azure Databricks Premium in East US, depending on the workload. Jobs Compute is $0.30, All-Purpose Compute is $0.55, Serverless SQL is $0.70, and serverless notebooks are $0.95. Classic compute adds VM charges on top.

Does Azure Databricks charge for idle clusters?

Yes. A classic cluster keeps billing DBUs and VM charges while it sits idle, until auto-termination shuts it down. Serverless SQL warehouses bill while they run, so set auto-stop to end idle time quickly.

FAQ

No items found.
Team Eon
Team Eon
>100% ROI in the first year

SoFi automated multi-region resilience and regulatory alignment across five AWS regions with Eon’s cloud-native platform, cutting recovery time from a day to minutes and achieving over 100% ROI.

Read case study
88% faster recovery, 35% savings

NETGEAR replaced its legacy backup provider with Eon's cloud-native platform, cutting a 10TB recovery from 24 hours to under three and reducing backup storage costs by 35% in under a week.

Read case study

Turn your backups into usable data

Eon turns your backups into instantly searchable, usable data so you can recover exactly what you need without delays.

  • Instantly search backup data with Global Search
  • Find any backup data with Global Search
  • Restore only what you need
See eon in action