Scheduled jobs (ETL)
Jobs on a schedule, running on Lakeflow Jobs compute.
Free tool · no sign-up
Databricks cost calculator
Estimate your monthly Databricks cost from the workloads you run — scheduled jobs, SQL analytics, interactive clusters, streaming — instead of a DBU number you don't have yet. The estimate uses current AWS, Azure, or GCP list rates and models what a rate card misses: idle time, autoscaling overshoot, retries, and auto-stop tails.
Runs in your browser · your inputs stay on the page · estimates, not quotes
Describe what you run
Toggle the workloads that exist in your environment and size them roughly. Rough is fine — the point is the shape of the bill, not the last dollar.
Start from a scenario
Each preset is a complete estimate of a typical platform at that stage — copy or share it as-is. Change anything and the scenario becomes Custom.
Cloud
List rates: AWS June 2026 · Azure and GCP, July 2026.
What runs in your environment — click to include:
Size each one in its section below — jump to Scheduled jobs ↓
SQL analytics / BI
Dashboards and ad-hoc queries on a SQL Serverless warehouse.
Interactive / data science
Notebooks and exploration on shared all-purpose clusters.
Streaming
Long-running structured streaming jobs on jobs compute.
Where the estimate comes from
Every line of the model, using the rates and assumptions below.
| Workload | DBUs/mo | Monthly total | |||||
|---|---|---|---|---|---|---|---|
| Scheduled jobs (ETL) Jobs Compute · $0.15/DBU · 6.85 DBU/hr | 2,466 | $1,220 | |||||
| SQL analytics / BI SQL Serverless · $0.70/DBU · 12 DBU/hr | 2,598 | $2,000 | |||||
| Interactive / data science All-Purpose Compute · $0.55/DBU · 1.38 DBU/hr | 538 | $557 | |||||
| Streaming Jobs Compute · $0.15/DBU · 1.38 DBU/hr | 1,007 | $475 | |||||
| Total | 5,602 | $3,778 | |||||
| Committed-use discount (0%) | −$0 | ||||||
| After discount | $3,778 | ||||||
Monthly total = Databricks DBU cost + cloud VM cost + the behavior layer (idle, overshoot, retries, auto-stop tails); the committed-use discount applies to the DBU layer including its behavior share, never to cloud VMs. Rounded for readability; the panel range applies a −20% / +25% band.
How the estimate is calculated
A Databricks bill has two layers. Databricks meters your compute in DBUs and charges DBUs consumed × the SKU rate. On classic compute, your cloud provider separately bills the VMs those clusters run on. Serverless SKUs fold the infrastructure into the DBU rate, so there is no second line.
This calculator derives hours from what you describe, turns hours into DBUs using the cluster or warehouse size, applies the rate, adds the VM layer where it applies, and then adds the behavior layer on top.
Compute rates
| Compute SKU | AWS $/DBU | Azure $/DBU | GCP $/DBU | Cloud infra |
|---|---|---|---|---|
| Jobs Compute (classic) | $0.15 | $0.30 | $0.15 | Billed separately |
| All-Purpose Compute (classic) | $0.55 | $0.55 | $0.55 | Billed separately |
| SQL Serverless | $0.70 | $0.70 | $0.70 | Included in the DBU rate |
List rates: AWS (June 2026), Azure and GCP (July 2026), Premium tier, pay-as-you-go, in USD.
Cluster and warehouse assumptions
| Cluster preset | AWS DBU/hr | Azure DBU/hr | GCP DBU/hr |
|---|---|---|---|
| Small · 1 driver + 1 worker · 4 vCPU each | 1.38 | 2.00 | 1.44 |
| Medium · 1 driver + 4 workers · 8 vCPU each | 6.85 | 10.00 | 7.20 |
| Large · 1 driver + 12 workers · 8 vCPU each | 17.81 | 26.00 | 18.72 |
Node families and their on-demand VM prices: m5.xlarge $0.192/hr and m5.2xlarge $0.384/hr on AWS (EC2 us-east-1), D4s v5 $0.192/hr and D8s v5 $0.384/hr on Azure (East US pay-as-you-go), n2-standard-4 $0.1942/hr and n2-standard-8 $0.3885/hr on GCP (Compute Engine us-central1). SQL Serverless warehouses use published DBU/hr (2X-Small 4 · X-Small 6 · Small 12 · Medium 24 · Large 40) on all three, with infrastructure inside the rate.
The behavior layer
Four adjustments, each a percentage on top of the metered base. They are the difference between what a rate card predicts and what an environment actually runs.
- Retries · Scheduled jobs · default 5% Reruns after a failed task. The compute is billed for both attempts.
- Autoscaling overshoot · Scheduled jobs and streaming · default 10% A cluster scales up faster than it scales back down, so some of the added capacity is billed while idle. Cluster start-up minutes, billed before a run does useful work, sit here too.
- Auto-stop tail · SQL warehouses · default 10% A warehouse keeps billing after the last query until auto-stop fires.
- Idle share · Interactive clusters · default 25% Shared clusters keep billing between sessions until auto-termination kicks in.
List rates change and differ by cloud, region, and tier. We check these against the Databricks pricing pages; last verified June 2026 for AWS, July 2026 for Azure, and July 2026 for GCP.
Why your invoice will still differ
- Committed-use discounts: Databricks commit contracts (DBCUs) buy consumption up front and lower your effective DBU rate below list, which is what the discount control in the panel approximates.
- Region and cloud: DBU rates and VM prices both move across AWS, Azure, and GCP, and across regions inside each one, though the structure of the bill stays identical.
- Photon: Photon changes how many DBUs an hour of compute consumes, so a faster job can still land on a different number than a rate-card estimate suggests.
- Infrastructure and storage: Spot instances and autoscaling reshape the VM layer, and storage, networking, and egress are billed by your cloud on top of everything modelled here.
The gap that is left is behavior, and that one you can see. Once a workload runs, system tables record actual consumption; that is the layer this estimate can only approximate.
An estimate is a model. A bill is behavior.
This page can tell you what a setup like yours bills at list rates. It cannot tell you which cluster idled through the weekend, which job doubled its runtime after a schema change, or which team owns the spike. LakeSentry reads your Databricks system tables — cost metadata only, never your data — and turns them into per-job, per-team attribution with anomaly alerts. Free tier, unlimited workspaces, no card.
Free
$0€0
Standard
$499€499/mo
Pro
$849€849/mo
No per-DBU tax. No per-workspace fees. Paid tiers billed annually.
Frequently asked questions
How much does Databricks cost per month?
There is no single number — Databricks bills per second of consumption. As scenario illustrations at June 2026 AWS list rates: the starter preset on this page lands around $450–710 a month, the growing-platform preset around $3,000–4,700, and the at-scale preset around $37,000–58,000. Your mix of workloads decides it.
What does a DBU cost?
A DBU is the unit Databricks meters per second of compute. On AWS Premium list rates (June 2026), jobs compute bills $0.15 per DBU, all-purpose compute $0.55, and SQL Serverless $0.70 with cloud infrastructure included. Tier, cloud, and region all shift the rate — see our pricing guide for the full card.
Does this estimate include cloud VM costs?
Yes. For classic compute (jobs, all-purpose, streaming) the estimate adds the separately-billed VM layer at on-demand list prices for the assumed node types — EC2 us-east-1, Azure East US, or Compute Engine us-central1, depending on the cloud you select. For SQL Serverless, infrastructure is already folded into the DBU rate, so no extra line is added. Storage and networking are excluded.
Why will my actual bill differ from this estimate?
Because a bill reflects behavior, not just rates: idle clusters, autoscaling overshoot, retries, and auto-stop tails. This calculator models them as adjustable percentages, but your real numbers are your own. Regions and Photon consumption still move the totals in both directions, and committed-use discounts can be applied with the discount control.
Does the calculator cover Azure and GCP?
Yes — the cloud switch applies each cloud's Premium list rates. The structure of the bill is identical everywhere, but the numbers are not: jobs compute bills $0.30 per DBU on Azure versus $0.15 on AWS and GCP, and each cloud rates its node types differently in DBU/hr, so the same workloads land on different totals. Rates: AWS June 2026, Azure and GCP July 2026.
How is this different from the official Databricks calculator?
The official calculator is good at one thing: multiplying a DBU quantity you already know by a SKU rate. This one starts a step earlier — you describe workloads, and it derives hours, DBUs, and the environment behavior that a clean rate-card estimate leaves out. It is an estimate either way; this one just fails in more honest ways.
Know where every Databricks dollar goes
Connect a workspace in about 15 minutes. Free tier, unlimited workspaces, no card.