Databricks DBU pricing is the way Databricks turns platform usage into charges. Each workload emits usage in a unit, mostly the Databricks Unit (DBU), against a SKU. Each SKU has a list price that depends on the workload type, the platform tier, the cloud and the region.[1] The Price List shows “Databricks’ undiscounted price for each SKU (“List Price”)”. Contract discounts, commitments and promotions are applied on top of it.[1] This article covers the units, the main SKU families, the tiers and add-ons, and how usage is measured. The commercial wrapper is covered in Databricks committed-use contracts and billing.
Editions
Platform tiers
Databricks on AWS has “two platform tiers: Premium and Enterprise”. New accounts start on Premium and can be upgraded from the account console.[15] The Standard tier has been retired. The pricing FAQ says that all Standard workspaces were to be upgraded to Premium on 2025-10-01 on AWS and Google Cloud, and on 2026-10-01 on Azure. Until then, existing Standard workspaces paid USD 0.10 per DBU for Jobs Compute on AWS and Google Cloud and USD 0.15 on Azure.[2] The tier names are not equivalent across clouds. The Premium tier on Azure Databricks corresponds to the Enterprise tier on AWS and Google Cloud.[2] Catalog proof: Platform tiers are Premium and Enterprise; Standard tier retired.
The tier changes the per-DBU price of many SKUs. At retrieval, Jobs Compute on AWS was USD 0.15 per DBU on Premium and USD 0.20 on Enterprise. Jobs Serverless was USD 0.35 and USD 0.45. All-Purpose Compute was USD 0.55 and USD 0.65, and All-Purpose Serverless was USD 0.75 and USD 0.95.[2][4] The SQL warehouse prices were the same on both tiers: USD 0.22 for SQL Classic, USD 0.55 for SQL Pro and USD 0.70 for SQL Serverless.[3]
Add-ons
Two add-ons are priced as a percentage of Product Spend instead of per unit.
| Add-on | What it adds | Price at retrieval |
|---|---|---|
| Enhanced Security and Compliance | Enhanced security and controls for compliance needs | 15% of Product Spend before discounts[6] |
| Mission Critical | Managed Disaster Recovery plus Enhanced Security and Compliance | 30% of Product Spend, promotionally 15% until 2027-06-30[6] |
Product Spend is “calculated based on product spend at list incurred in the specific workspaces where the add-on is enabled (turned on), before the application of any discounts, usage credits, add-on uplifts, or support fees”. A workspace with Mission Critical is not also charged for Enhanced Security and Compliance.[6] Because the uplift is calculated on list price, it does not fall when contract discounts rise. Catalog proof: Platform add-ons are charged as a percentage of Product Spend at list.
Metrics
| Unit | Catalog row | Applies to |
|---|---|---|
| DBU, a normalized unit of processing power | Databricks Unit (DBU) | Jobs, SQL, All-Purpose, Pipelines, serverless, model serving, managed services |
| DSU, a standardized unit of storage usage | Databricks Storage Unit (DSU) | Databricks default storage |
| Capacity Unit hour | Lakebase Capacity Unit hour | Lakebase Postgres compute |
| GB (2^30 bytes) and endpoint hour | Gigabyte, Private Connectivity Endpoint hour | Data transfer and connectivity |
| Percentage of Product Spend | Product Spend | Platform add-ons |
Databricks says the number of DBUs a workload consumes “is driven by processing metrics, which may include the compute resources used and the amount of data processed”.[1] The billable usage table records the unit for each row in its usage fields. Its usage_type column distinguishes usage types such as COMPUTE_TIME, STORAGE_SPACE, NETWORK_BYTE, NETWORK_HOUR, API_OPERATION, TOKEN, GPU_TIME and ANSWER.[11] Catalog proof: DBU is the normalized unit for pricing compute.
Counting / floors
Classic compute
Classic clusters run in the customer’s own cloud account. The DBU price excludes the virtual machines: the Jobs page lists “Plus underlying compute costs billed by cloud provider”, and the customer pays the cloud provider separately for instances, “including idle time when leveraging instance pools”.[2] Photon is an option on classic clusters and raises the DBU emission rate. The rate is 2.9 times that of a non-Photon cluster for Jobs and Pipelines and 2 times for All-Purpose clusters.[2][4][5] Bills list classic jobs as “Jobs Compute” and Photon jobs as “Jobs Compute (Photon)”.[2] Catalog proof: Classic compute VMs are billed separately by the cloud provider; Photon raises the DBU emission rate on classic compute.
Lakeflow Pipelines on classic compute come in three feature levels with rising prices: Core at USD 0.20, Pro at USD 0.25 and Advanced at USD 0.36 per DBU on AWS Premium at retrieval. Advanced adds change data capture and data quality expectations.[5] The DLT Core Compute SKU row records the Core price.
Serverless compute
Serverless prices include the underlying compute. Databricks chooses and scales the instances.[2] There is no charge for start-up time, and “Serverless compute is billed with a one-minute minimum”. Photon is always on in serverless, and its effect is built into the serverless DBU rate.[2] Because customers cannot size serverless capacity, Databricks recommends running representative workloads and measuring the DBUs they consume.[2][14] Catalog proof: Serverless compute has a one-minute minimum and excludes VM start-up time.
Some features run on serverless infrastructure even in accounts that do not use serverless notebooks or jobs. Data quality monitoring and predictive optimization “are billed under the serverless jobs SKU”.[14] Materialized views, Lakeflow Connect, AI Search and clean rooms also produce serverless usage records.[13] The Managed Services price page lists data quality monitoring, predictive optimization, fine-grained access control and data classification at USD 0.35 per DBU on AWS Premium. It also explains that DBUs for Lakehouse Monitoring are doubled before pricing.[16] Catalog proof: Some background features bill under the serverless jobs SKU.
Storage, Lakebase, serving and networking
Default storage is priced per DSU, USD 0.023 on AWS at retrieval. Stored data counts as 1 DSU per GB-month. On AWS and Google Cloud, Tier 1 operations (on AWS, PUT, COPY, POST and LIST requests) count as 0.2174 DSU per 1,000 and Tier 2 operations (GET, SELECT and other requests) as 0.0174 DSU per 1,000.[7] Lakebase compute is billed on the Capacity Unit hours actually used between a minimum and maximum autoscaling range. Always-On instances never scale to zero and carry a minimum usage charge, and storage, point-in-time restore and snapshots are billed per GB-month.[8] Model Serving bills CPU serving per DBU. GPU serving is billed per GPU instance hour through published DBU-per-hour rates, for example 10.48 DBUs per hour for a T4 and 800 DBUs per hour for 8 H100 GPUs.[9] Serverless products can also incur charges for private connectivity per GB and per endpoint hour, public connectivity per GB and data egress per GB.[17] Catalog proof: Default storage is priced per DSU.
Virtualization & partitioning
There is no processor, core or partitioning rule. On classic compute, the cluster’s instance types and node count determine how many DBUs it emits. The billable usage table records the node_type for non-serverless usage but does not distinguish driver and worker nodes.[11]
Cloud / BYOL
Prices “may vary based on geographic region and cloud service provider”.[1] Azure Databricks prices are set by Microsoft, and Databricks shows them only “for convenient reference”. Model Serving explains its regional prices as reflecting “the regional cost of infrastructure supporting our serverless products”.[1][9] There is no bring-your-own-licence path. On classic compute the customer always pays its cloud provider for the instances, under its own cloud agreement and any cloud discounts it holds.[2]
Programs
SKU Groups
The Price List groups SKUs into “Cross Service SKU Groups” and “Service Specific SKU Groups”.[1] Examples of cross-service groups are the AWS Cross-service SKU Group and the GCP Cross-service SKU Group. Service-specific groups include AWS Jobs Compute, AWS Serverless SQL Compute and AWS Databricks Storage.[10] “A small subset of SKUs are excluded from Cross-Service SKU groups”, in particular services with large third-party or infrastructure components and some pre-GA services. The listing excludes data transfer and egress, Databricks Storage, model training, real-time inference, and OpenAI, Anthropic and Gemini model serving SKUs.[10] When a contract prices a discount by SKU Group, these exclusions decide which usage gets the discount. Catalog proof: Excluded SKUs do not draw on Cross-service SKU Groups.
Promotional Discounts
Databricks may offer “generally-applicable promotional discount pricing for specific Databricks Services”. These are identified on the Price List, together with whether they combine with Committed Use Contract discounts.[1] At retrieval, promotions were shown for Lakebase compute (50% until 2027-01-31), Lakehouse RT warehouses (30% until 2027-01-31) and the Mission Critical add-on.[8][3][6] See Promotional Discount.
Measuring consumption
The billable usage system table, system.billing.usage, centralizes usage for the whole account. Records are “typically available within 12 hours”.[11] Serverless usage can take up to 24 hours to appear.[14] Each record carries the SKU, cloud, usage unit, quantity, workspace, custom tags and identity metadata. The record_type column marks corrections as retractions or restatements of earlier records.[11] Prices are in system.billing.list_prices, which adds a record “each time there is a change to a SKU price”. Its pricing column has default, promotional and effective_list keys, and the effective list price is “used for calculating the cost”.[12] Serverless usage policies attach tags to serverless usage for chargeback. They do not tag classic compute.[18] Catalog proof: Billable usage is recorded in system.billing.usage by SKU and unit; List price history is kept in system.billing.list_prices; Serverless usage policies tag serverless usage for attribution.
A usage quantity multiplied by the list price gives list cost, not invoiced cost. The invoiced amount depends on the discounts and commitments in the Order. That is why system-table cost estimates and Databricks budgets, which use list prices, can differ from invoices.
Out of scope
This page does not cover Azure Databricks prices, which Microsoft sets, SAP Databricks capacity unit pricing within SAP Business Data Cloud, per-token pricing of Foundation Model APIs and proprietary model serving, Databricks Apps, clean rooms, or Agent Bricks pricing. Each of these has its own pricing page on databricks.com.