# Databricks Pricing: DBUs, Compute Types and Monthly Cost Examples

Databricks charges for the processing your workloads consume, with different rates for jobs, SQL and interactive notebooks. On AWS Premium in US East, Jobs Serverless costs USD 0.35 per DBU and SQL Serverless costs USD 0.70 per DBU; classic compute adds a separate cloud infrastructure bill. The worked examples below range from USD 163.95 for a light startup workload to USD 3,019 for a larger SaaS workload, before storage operations, networking and optional services. [Jobs pricing](https://www.databricks.com/product/pricing/lakeflow-jobs) and [SQL pricing](https://www.databricks.com/product/pricing/databricks-lakehouse).

Canonical URL: https://www.martechdevs.com/blog/databricks-pricing

By [Jafrin Akter Alo](https://www.martechdevs.com/authors/jafrin-akter-alo)

Data Engineer and Co-founder at [Martechdevs](https://www.martechdevs.com) and [Growth and Analytics Partners](https://www.growthandanalytics.com)

[LinkedIn](https://linkedin.com/in/jafrinalo)

Last updated: 2026-11-11

Next step: [Compare Snowflake and Databricks](https://www.martechdevs.com/blog/snowflake-vs-databricks). Match the platform to SQL, engineering and model workloads.

## What are you billed for in Databricks?

Databricks bills compute in Databricks Units, or DBUs, which measure processing consumption rather than people or stored rows. Multiply the DBUs consumed by the rate for the actual product, tier, cloud and region. Then add separately billed storage, networking and enabled services. Classic compute also needs the cloud provider's machine costs.

USD pay-as-you-go list rates below apply to AWS US East, Northern Virginia, for the named tier. A rate from another cloud, product or region is a different comparison. The official [Jobs](https://www.databricks.com/product/pricing/lakeflow-jobs), [SQL](https://www.databricks.com/product/pricing/databricks-lakehouse) and [interactive compute](https://www.databricks.com/product/pricing/datascience-ml) rate cards supply these figures.

| Compute product | Premium, USD/DBU | Enterprise, USD/DBU | Underlying compute included? |
| --- | ---: | ---: | --- |
| Jobs Classic | 0.15 | 0.20 | No; add cloud infrastructure |
| Jobs Serverless | 0.35 | 0.45 | Yes |
| SQL Classic | 0.22 | 0.22 | No; add cloud infrastructure |
| SQL Pro | 0.55 | 0.55 | No; add cloud infrastructure |
| SQL Serverless | 0.70 | 0.70 | Yes |
| All-Purpose Classic | 0.55 | 0.65 | No; add cloud infrastructure |
| Interactive Serverless | 0.75 | 0.95 | Yes |

DBUs are consumption units. A warehouse consuming 12 DBUs an hour for ten billable hours uses 120 DBUs. At USD 0.70 per DBU, its compute costs USD 84. Do not multiply those 120 measured DBUs by hours again.

For a marketing stack, separate ingestion, transformation, SQL reporting and interactive exploration before estimating. They run at different frequencies and can use different products. Our [data warehouse implementation service](https://www.martechdevs.com/data-warehouse-cdp) scopes those workloads alongside identity, reporting and activation, so the cost model reflects the work the business actually needs.

## What will Databricks cost a company like yours?

Estimate Databricks from workload consumption, then use company size only to describe the scenario. The examples below combine SQL warehouse runtime with assumed monthly job and notebook DBUs. They are reproducible planning examples at list price, not client results, benchmarks or claims that every company in a size band needs the same compute.

All four examples use AWS Premium in Northern Virginia, one SQL cluster at the stated size, no negotiated discount and no platform add-ons. SQL sizes and DBUs per hour come from the [Databricks calculator](https://www.databricks.com/product/pricing/product-pricing/instance-types). Jobs and notebook consumption are explicit assumptions to replace with measured usage. Default storage uses USD 0.023 per GiB-month of stored data; operations and networking are extra. [Default storage pricing](https://www.databricks.com/product/pricing/storage).

| Company situation | SQL assumption | Other monthly usage assumptions | Listed monthly subtotal |
| --- | --- | --- | ---: |
| Five-person startup, light reporting | 2X-Small, 4 DBU/hour, 44 billable hours | 100 Jobs Serverless DBUs; 250 GiB stored; no notebook compute | USD 163.95 |
| 20-person SaaS, daily reporting | X-Small, 6 DBU/hour, 132 billable hours | 500 Jobs Serverless DBUs; 100 Interactive Serverless DBUs; 1,000 GiB stored | USD 827.40 |
| 50 to 100 person B2B SaaS | Small, 12 DBU/hour, 180 billable hours | 1,920 Jobs Serverless DBUs; 960 Interactive Serverless DBUs; 5,000 GiB stored | USD 3,019 |
| 200 to 500 person business, broader usage | Medium, 24 DBU/hour, 300 billable hours | 6,000 Jobs Serverless DBUs; 2,000 Interactive Serverless DBUs; 20,000 GiB stored | USD 9,100 |

The subtotal covers the named serverless compute and stored-data volume only. Add storage operations, transfer, connectivity, model serving, managed services, support and any other contracted charges that apply. A subtotal with explicit boundaries is usable; a single number labelled as the whole platform bill hides those choices.

For the 50 to 100 person example:

- SQL: `12 × 180 × 0.70 = USD 1,512`.
- Jobs: `1,920 × 0.35 = USD 672`.
- Notebooks: `960 × 0.75 = USD 720`.
- Stored data: `5,000 × 0.023 = USD 115`.
- Listed monthly subtotal: `1,512 + 672 + 720 + 115 = USD 3,019`.

Use the [Databricks cost calculator](https://www.martechdevs.com/tools/databricks-cost-calculator) to replace these assumptions. A full month of stored data uses the vendor's storage convention; Databricks measures its billed GB in binary units, or GiB, and prorates storage using a 730-hour month. [Storage billing details](https://www.databricks.com/product/pricing/storage).

### How do you estimate Jobs Serverless DBUs before purchasing?

Run representative workloads and measure the resulting DBUs. Include a normal incremental run, a large first load, a retry and a peak-period run. Serverless chooses resources for the workload, so input row count alone cannot establish the consumption. Databricks recommends measuring representative jobs in its [Jobs pricing explanation](https://www.databricks.com/product/pricing/lakeflow-jobs).

Suppose a successful daily job consumes 40 DBUs. Thirty executions imply 1,200 DBUs, or USD 420 at the example Premium serverless rate. If four additional reruns also consume 40 DBUs each, the compute becomes USD 476. Record the measured cost of reruns separately; a retry may process a different amount of work.

## What makes a Databricks bill jump?

Longer runtime, larger or additional SQL clusters, repeated processing and the wrong compute product are the main levers in these examples. Other products can add their own meters. Diagnose the change by workload and billing SKU before changing an account-wide setting, so a repair targets the source of the increase.

| Change in the example workload | Calculation | Cost effect |
| --- | --- | ---: |
| Small SQL warehouse runs 720 hours instead of 180 | Additional 540 hours × 12 DBU/hour × USD 0.70 | USD 4,536 more SQL compute |
| Small warehouse increases to Medium for the same 180 hours | Additional 12 DBU/hour × 180 × USD 0.70 | USD 1,512 more, before any runtime improvement |
| Job usage doubles from 1,920 to 3,840 DBUs | Additional 1,920 × USD 0.35 | USD 672 more |
| Classic workload uses All-Purpose rather than Jobs at equal 1,000 DBUs | USD 0.55 versus USD 0.15 per DBU | USD 400 more platform cost, with cloud costs still separate |

These are arithmetic sensitivities using the [official calculator](https://www.databricks.com/product/pricing/product-pricing/instance-types) and [compute rate cards](https://www.databricks.com/product/pricing/datascience-ml). Equal runtime and equal DBUs are assumptions for isolating a price effect. A larger warehouse may complete the work faster, so measure cost per completed workload before deciding to resize it.

SQL concurrency can start additional clusters. The relevant billable hours then include those clusters, not only the elapsed time the warehouse was open. Databricks describes the relationship between size, concurrency and autoscaling in its [warehouse sizing guide](https://docs.databricks.com/aws/en/compute/sql-warehouse/warehouse-behavior).

AI features and pipeline maintenance can create separate charges. SQL calls to generative AI functions use model-serving billing, while materialized-view and streaming-table updates use serverless pipeline billing. Do not assume the SQL warehouse DBUs include every feature invoked from SQL. [SQL rate-card billing explanation](https://www.databricks.com/product/pricing/databricks-lakehouse).

## Does adding people or adding usage cost more?

The compute products above charge for consumption, so adding an account is not equivalent to adding a fixed monthly compute seat. Costs rise when users run more queries, keep compute active, create more workloads or increase concurrency. Budget SQL, jobs, notebooks and additional AI services separately instead of multiplying headcount by one platform price.

A dashboard with 500 occasional readers can cost less to operate than repeated large notebook experiments by five engineers. The useful denominator is a delivered output: a completed load, a refreshed business model or an answered report query. Headcount helps predict demand, but usage establishes the charge. Product-specific rates remain in the [Databricks pricing catalogue](https://www.databricks.com/product/pricing).

Query the billing records by workspace, product and owner. `system.billing.usage` provides the consumed quantity, unit, product and resource metadata. Keep the usage unit in the grouping because DBUs, storage and other metered quantities cannot be added as if they were interchangeable. [Billable usage reference](https://docs.databricks.com/aws/en/admin/system-tables/billing).

## Which Databricks plan and compute type fit your situation?

Choose the platform tier for required controls, then choose compute for each workload's behavior and compatibility. Premium and Enterprise prices are not interchangeable, and a serverless job is not simply a classic job with a different unit price. Compare total cost and required functionality on the same input and acceptance criteria.

| Situation | First option to evaluate | What decides the fit |
| --- | --- | --- |
| Learning or a personal prototype | Free Edition | Fair-use quotas and lack of an SLA are acceptable |
| Scheduled supported transformations | Jobs Serverless | Measured DBUs, completion time and supported APIs |
| Work needing unsupported serverless behavior | Jobs Classic | Required configuration plus the full cloud bill |
| SQL reporting and dashboards | SQL Serverless | Query performance, concurrency and region availability |
| Enterprise security or operating requirements | Appropriate tier and add-ons | Exact controls and contract scope |

Serverless notebooks and jobs require compatible code and access patterns. For example, the current limitations include no R support or Spark RDD APIs, and external data access uses Unity Catalog. Check the actual workload before estimating a migration saving. [Serverless limitations](https://docs.databricks.com/aws/en/compute/serverless/limitations).

Free Edition provides one 2X-Small SQL warehouse and a maximum of five concurrent job tasks, subject to fair-use limits. It has no guaranteed reliability, support or SLA. Use it to learn and assess the workflow; a business reporting dependency needs its own reliability decision. [Free Edition limits](https://docs.databricks.com/aws/en/getting-started/free-edition-limitations).

For a warehouse-focused team choosing a platform, our [Snowflake versus Databricks comparison](https://www.martechdevs.com/blog/snowflake-vs-databricks) separates BI, engineering and model workloads. Do that architecture comparison before optimizing the rate of a platform you may not need.

![Check serverless compatibility, measure its DBUs or price classic compute with cloud infrastructure, add other charges and test outputs, runtime and budget. Failed checks return to workload design.](https://www.martechdevs.com/assets/blog/databricks-pricing.svg)

Diagram connections:

- Define workload inputs and completion target to Serverless supports required behavior?. 
- Serverless supports required behavior? to Measure serverless DBUs on representative runs: Yes. 
- Serverless supports required behavior? to Price classic DBUs and cloud infrastructure: No. 
- Measure serverless DBUs on representative runs to Add storage networking and enabled services. 
- Price classic DBUs and cloud infrastructure to Add storage networking and enabled services. 
- Add storage networking and enabled services to Outputs runtime and budget acceptable?. 
- Outputs runtime and budget acceptable? to Change workload compute or processing frequency: No. 
- Change workload compute or processing frequency to Define workload inputs and completion target. 
- Outputs runtime and budget acceptable? to Assign cost owner and monitor usage: Yes

## How do you keep Databricks costs in check?

Set controls around when compute runs, what it processes and who owns the bill. Start with the largest repeated workloads and idle resources, then measure the cost per successful output after each change. Cutting the DBU rate without preserving runtime, data quality or completion reliability can increase the overall operating cost.

1. Configure SQL Auto Stop deliberately. Idle warehouses continue billing until stopped. Serverless defaults to ten minutes; the UI allows a five-minute minimum and the API can set one minute. Balance the setting against real query gaps and restart behavior. [Warehouse settings](https://docs.databricks.com/aws/en/compute/sql-warehouse/create).
2. Limit SQL cluster scaling to the concurrency the service needs, then inspect queued queries and spill. A small configuration that makes critical reports unusable is not a successful saving. [Sizing and queue guidance](https://docs.databricks.com/aws/en/compute/sql-warehouse/warehouse-behavior).
3. Run scheduled production transformations on the appropriate job compute. Compare actual job completion costs before moving work off an interactive cluster; include cloud machines in the classic calculation.
4. Make backfills deliberate. Select the affected partitions or time range, confirm an idempotent write strategy, and estimate the rerun before starting it. Our [incremental backfill guide](https://www.martechdevs.com/blog/dbt-incremental-backfills) explains the repair decisions behind that estimate.
5. Tag workloads by team, environment and purpose. Use tags to allocate spending, without placing personal or sensitive information in tag values. [Usage attribution](https://docs.databricks.com/aws/en/admin/usage/system-tables).
6. Set budget alerts and assign someone to act on them. Ordinary compute-budget monitoring does not establish a hard stop for SQL or jobs; the documented blocking option is scoped to Unity Gateway requests. [Budget controls](https://docs.databricks.com/aws/en/admin/account-settings/budgets).

For a monthly list-price report, join usage with the historical matching prices. Match the SKU, cloud and unit, apply the price validity interval, and use the effective list price. Retain corrections and retractions so a replaced billing record is not counted twice. The result is a list-price estimate to reconcile with the actual invoice, which can include negotiated terms and other charges. [Pricing table](https://docs.databricks.com/aws/en/admin/system-tables/pricing) and [usage record types](https://docs.databricks.com/aws/en/admin/system-tables/billing).

Before approving the budget:

## Checklist

- [ ] Separate SQL, jobs, interactive work, storage and additional products.
- [ ] Record the actual cloud, region, tier and billing SKU for each rate.
- [ ] Measure normal, peak, first-load and retry consumption.
- [ ] Include classic cloud infrastructure or confirm serverless compute inclusion.
- [ ] Include storage operations, transfer and enabled platform add-ons.
- [ ] Assign owners, alerts and a response for unexpected usage.

Use the [marketing data audit skill](https://www.martechdevs.com/skills/marketing-data-audit) for the upstream marketing data, warehouse and reporting evidence that informs workload scope. Databricks-specific billing remains in its system tables and cloud invoice.

## What should you watch for in a Databricks contract?

Compare the committed amount with expected eligible usage and confirm exactly which products, clouds and workspaces consume it. Ask how unused commitments, excess consumption, credits, support and add-ons are handled. A discount percentage is incomplete without those terms and a workload forecast that includes the next full operating year.

Databricks offers pay-as-you-go pricing and quoted committed-use arrangements. Do not invent a standard discount or assume every service draws from the same commitment. Platform add-ons can calculate their charge from list-price product spend in the enabled workspaces before discounts and credits. [Platform tiers and add-ons](https://www.databricks.com/product/pricing/platform-addons).

Keep these items explicit in the order form:

- Applicable SKU, cloud, region and platform tier.
- Minimum commitment, term, consumption eligibility and excess-use rates.
- Whether unused value expires, carries forward or has another agreed treatment.
- Support fees and any minimums.
- Enabled security or reliability add-ons and their calculation base.
- Promotional credits, eligible products and expiration.

The [Databricks for Startups programme](https://www.databricks.com/product/startups) advertises up to USD 200,000 in credits for qualifying startups across Databricks and Neon. The maximum is not an entitlement for every applicant. Confirm the approved allocation and eligible usage before deducting anything from the budget.

At unchanged usage and list prices, the USD 3,019 monthly example becomes USD 36,228 for a full year of its named components, before the excluded charges. Show that unsubsidized year next to any first-year credit benefit. It makes the budget useful when credits expire or the next contract starts.

## How do you read the Databricks rate card?

Read every line as a combination of product, consumption unit, cloud, region and tier. USD per DBU is a rate; DBUs per hour is a consumption speed; measured monthly DBUs is a quantity already accumulated. The final product cost uses the quantity once, with its matching rate and any separate infrastructure charges.

| Rate-card term | Plain meaning | Budget implication |
| --- | --- | --- |
| USD/DBU | Price of one consumed processing unit | Multiply by measured DBUs |
| DBU/hour | Consumption rate while that configuration runs | Multiply by billable runtime and relevant cluster count |
| Classic compute | Compute infrastructure billed by the cloud provider | Include driver, workers and other cloud resources |
| Serverless compute included | Underlying compute included in that service rate | Keep storage, networking and other products separate |
| DSU | Databricks Storage Unit | Include stored data and billed storage operations |

The [official calculator](https://www.databricks.com/product/pricing/product-pricing/instance-types) gives a Small SQL Serverless warehouse at 12 DBUs/hour in the example region, while the [SQL rate card](https://www.databricks.com/product/pricing/databricks-lakehouse) gives its per-DBU price. Use both values rather than treating USD 0.70 as an hourly warehouse price.

For default storage, one GiB stored for the billed month consumes one DSU, and request operations can add DSUs. Storage lifecycle charges can continue for up to 30 days after deletion. External cloud storage has its own provider bill and should not be silently priced as Databricks default storage. [Storage rate-card explanation](https://www.databricks.com/product/pricing/storage).

For serverless networking, review data transfer and connectivity charges for the actual architecture. Cross-region movement, private endpoints and data processing can create charges beyond compute. A service being serverless does not establish that moving data is free. [Networking rate card](https://www.databricks.com/product/pricing/data-transfer-connectivity).

## What should you do before setting a Databricks budget?

Price and run a representative workload, including its normal update and repair paths, before making a large commitment. Compare the measured bill with the estimate, account for the missing components, and attach an owner to each recurring workload. That turns a rate-card comparison into an operating budget that can be reviewed against actual delivery.

Use our [Databricks cost calculator](https://www.martechdevs.com/tools/databricks-cost-calculator) for the arithmetic, [Snowflake versus Databricks guide](https://www.martechdevs.com/blog/snowflake-vs-databricks) for the platform decision and [data warehouse cost guide](https://www.martechdevs.com/blog/data-warehouse-cost) for the surrounding pipeline and operating costs. Our [data warehouse and CDP service](https://www.martechdevs.com/data-warehouse-cdp) covers the implementation when the models, access and reporting need to be built together.

## Frequently asked questions

### Is a Databricks DBU the same as a virtual machine hour?

No. A DBU measures Databricks processing consumption, while a virtual machine hour measures infrastructure runtime. A machine or SQL warehouse can consume several DBUs per hour. Classic compute generally has both a Databricks platform charge and a cloud infrastructure charge; serverless compute includes the underlying compute in its service rate. [Databricks calculator](https://www.databricks.com/product/pricing/product-pricing/instance-types).

### Is Azure Databricks priced the same as AWS Databricks?

No. The cloud, region, tier and product determine the rate. Azure Databricks is billed through Microsoft, and Databricks states that Azure Premium corresponds to Enterprise on AWS and GCP. Compare the required capabilities and actual regional rates rather than applying an AWS Premium quote to an Azure deployment. [Databricks tier qualification](https://www.databricks.com/product/pricing/lakeflow-jobs).

### Does stopping a SQL warehouse stop every Databricks charge?

No. Stopping the warehouse ends that warehouse's ongoing compute consumption, but stored data, other workloads and applicable services can continue billing. Check jobs, notebooks, storage lifecycle and networking separately. A warehouse Auto Stop setting is useful for idle SQL spend, while the account's total bill requires a broader inventory. [Warehouse settings](https://docs.databricks.com/aws/en/compute/sql-warehouse/create).

### Can I use Free Edition as a production budget baseline?

Free Edition is a poor baseline for a business workload that needs predictable availability or support. It has fair-use quotas, limited compute and no service-level agreement. Establish the paid deployment needed for your runtime, security and support requirements, then treat free access as a learning or evaluation benefit. [Free Edition limitations](https://docs.databricks.com/aws/en/getting-started/free-edition-limitations).

### Will a lower DBU rate always make a job cheaper?

No. Total cost depends on consumed DBUs, runtime, cloud infrastructure and other relevant charges. A cheaper unit rate can lose its advantage if the job consumes more units or requires more operating work. Compare the same inputs and accepted outputs across the options, including retries and peak runs. [Measuring serverless consumption](https://www.databricks.com/product/pricing/lakeflow-jobs).

## Sources

1. [Databricks pricing catalogue](https://www.databricks.com/product/pricing)

2. [Lakeflow Jobs pricing](https://www.databricks.com/product/pricing/lakeflow-jobs)

3. [SQL pricing](https://www.databricks.com/product/pricing/databricks-lakehouse)

4. [Interactive compute pricing](https://www.databricks.com/product/pricing/datascience-ml)

5. [Databricks pricing calculator](https://www.databricks.com/product/pricing/product-pricing/instance-types)

6. [Databricks storage pricing](https://www.databricks.com/product/pricing/storage)

7. [Data transfer and connectivity pricing](https://www.databricks.com/product/pricing/data-transfer-connectivity)

8. [Platform tiers and add-ons](https://www.databricks.com/product/pricing/platform-addons)

9. [SQL warehouse sizing and scaling](https://docs.databricks.com/aws/en/compute/sql-warehouse/warehouse-behavior)

10. [SQL warehouse creation and Auto Stop](https://docs.databricks.com/aws/en/compute/sql-warehouse/create)

11. [Serverless compute limitations](https://docs.databricks.com/aws/en/compute/serverless/limitations)

12. [Free Edition limitations](https://docs.databricks.com/aws/en/getting-started/free-edition-limitations)

13. [Billable usage system table](https://docs.databricks.com/aws/en/admin/system-tables/billing)

14. [Pricing system table](https://docs.databricks.com/aws/en/admin/system-tables/pricing)

15. [Monitor costs with system tables](https://docs.databricks.com/aws/en/admin/usage/system-tables)

16. [Create and monitor budgets](https://docs.databricks.com/aws/en/admin/account-settings/budgets)

17. [Databricks for Startups](https://www.databricks.com/product/startups)
