Databricks sells its platform on a pay-as-you-go basis with no up-front costs, charging only for the products a team uses at per-second granularity, according to the company’s pricing page. For B2B buyers, the practical work is not choosing a plan. It is knowing which meter each product runs on, what the published rates leave out, and who issues the invoice.
What it is and who it’s for
Databricks describes its pricing model as pay-as-you-go with no up-front costs, where customers pay for the products they use. There is no single platform rate to approve: the pricing page breaks the catalog into product families, each with its own entry rate, so the first budgeting question is which of those families a team expects to consume.
Databricks answers the scope question directly in its FAQ: “Databricks pricing is based on your compute usage.” The same answer draws a boundary around what the Databricks bill covers, adding that “Storage, networking and related costs will vary depending on the services you choose and your cloud service provider.”
The model applies to two different commercial surfaces. Teams buying from Databricks work from its published Price List. Teams buying Azure Databricks do not: Databricks notes that pricing for Azure Databricks is set by Microsoft and appears on its own page only for convenient reference, and its billing FAQ says charges for Azure Databricks usage are billed directly by Microsoft and governed by the customer’s Azure subscription terms.
How it works
The unit behind most of the catalog is the DBU. Databricks defines a Databricks Unit as “a normalized unit of processing power on the Databricks Lakehouse Platform used for measurement and pricing purposes,” and 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.
That definition is why a per-DBU rate is not a per-hour rate. Two workloads running for the same wall-clock time can draw different DBU counts, because consumption tracks processing metrics rather than elapsed time alone.
The DBU is also not the only meter on the page. Databricks prices its Operational Database product per CU rather than per DBU. It defines a Databricks Storage Unit as a standardized unit for measuring storage usage across multiple Databricks products, and says DSU consumption may vary by region, storage type, volume of data stored, or transactions. Data transfer and networking products are measured in GB, which Databricks defines as 2^30 bytes, and private connectivity endpoints are counted in hours provisioned, rounded up to the nearest whole hour.
Four published terms turn those meters into a number. The Price List is the per-cloud-provider listing of Databricks Platform Services products and SKUs. List Price is the undiscounted price shown for each SKU. A SKU Group is a Cross Service or Service Specific bundle whose included product SKUs are documented in a separate listing. A Promotional Discount is generally-applicable discount pricing Databricks may offer from time to time for specific services, identified on the Price List or communicated to eligible customers.
Pricing and cost considerations
Databricks publishes an entry rate per product family. These are the figures on its pricing page:
| Product family | Published entry rate |
|---|---|
| Artificial Intelligence | Starting at $0.07 / DBU |
| Genie | Starting at $0.07 / DBU beyond free usage |
| Data Engineering | Starting at $0.15 / DBU |
| Data Warehousing | Starting at $0.22 / DBU |
| Interactive workloads | Starting at $0.40 / DBU |
| Operational Database | Starting at $0.069 / CU |
Two qualifiers sit on every row. Each is a “starting at” figure, so it is a floor rather than the rate a specific configuration pays. And each is a List Price, which Databricks identifies as undiscounted, while noting that pricing displayed for SKUs identified as promotional is temporarily discounted.
Geography moves the same numbers again. Databricks says its prices and cloud infrastructure prices “may vary based on geographic region and cloud service provider,” and points buyers to the pricing pages for individual products. A rate quoted for one region and one cloud is not portable to another.
The Platform row carries no headline rate at all. Databricks lists it as cross-platform capability spanning governance, management, security, managed services, tiers and add-ons, data transfer and connectivity, storage, and Delta Share, so those items have to be looked up in the relevant category listing rather than read off a single entry figure.
On the payment side, pay-as-you-go accounts are billed monthly. Databricks says the credit card on file is charged monthly based on Platform Services usage at Price List rates, that a customer may be charged more frequently for accrued fees with the remaining balance billed at the end of the month, and that other arrangements such as billing by invoice or an annual plan are available on request.
How to choose
Start with the seller, not the rate card. If the deployment is Azure Databricks, Microsoft sets the price and issues the bill, so the commercial review belongs to the Azure subscription rather than to Databricks’ own list.
Next, decide whether the workload is stable enough to commit. Databricks says Committed Use Contracts offer access to discounts and other benefits in exchange for committing to certain levels of usage, that larger commitments bring greater benefits, and that those benefits include options to use commitments flexibly across multiple clouds. Usage predictability is therefore the dividing line between treating Databricks as on-demand spend and negotiating it as a commitment purchase. The same trade-off shows up across metered data platforms, as our Snowflake pricing guide covers for credit-based billing.
Then match the meter to the workload. A team whose spend concentrates in interactive data science starts from the catalog’s highest published per-DBU figure at $0.40; a team whose spend sits in the AI or Genie families starts from $0.07. Because DBU consumption tracks processing metrics, that gap is an input to a usage model rather than a ranking of total cost. For how metered billing compares with seat and tier structures, see our guide to SaaS pricing models.
Teams that want to test before committing should be clear on what the trial does not cover. Databricks says the free trial includes access to the Data Intelligence Platform and allows colleagues to be invited to the account, but also that a customer who configures Databricks to work with their own cloud account will still be charged by that cloud provider for resources, such as compute instances, used within the account during the free trial.
Limitations and gotchas
The published rates are floors, not quotes. Every product figure is a “starting at” number at undiscounted List Price, so no row on the pricing page states what a given workload will actually be billed.
The compute-only framing also does not fully square with the rest of the page. The FAQ says pricing is “based on your compute usage,” yet the same page lists Storage among its Platform capabilities and defines a storage unit for measuring storage usage across products. Buyers should confirm which storage charges land on the Databricks invoice and which land on the cloud provider’s, rather than reading that FAQ line as a guarantee that Databricks never meters storage.
SKU Groups can mislead a buyer who reads them as complete bundles. Databricks states that certain product exclusions may apply, as detailed in the SKU Groups listing itself.
Trial credits carry more conditions than the trial headline suggests. Databricks says the credits are not transferable, cannot be combined with other promotions or negotiated terms, and are not available to customers who previously accepted them. Eligibility, the amount of credits, and when they expire are determined solely by Databricks following account creation, and “A customer’s trial ends when their credits are exhausted.”
Cancellation does not stop accrued charges. Databricks says a pay-as-you-go account can be cancelled at any time, and that upon cancellation all usage up to the termination date will be charged in full.
One clause is easy to miss entirely. Databricks states that in the event of conflict between the customer’s Master Cloud Services Agreement and the billing description on the pricing page, the pricing-page description governs. A buyer reconciling contract language against the public page should not assume the signed agreement is the controlling text here.
FAQ
Can a promotional discount be stacked on a Committed Use Contract discount?
Databricks does not state a general rule. It says it will publish or communicate additional details for each Promotional Discount, including eligibility criteria, duration, and whether that discount cumulates with Committed Use Contract discounts, so stacking is settled per promotion rather than by policy.
What happens when the credits in a free trial run out?
The trial ends. Databricks also says that at the end of a trial the customer is automatically subscribed to the plan they had been using during it, and may need to input a valid payment method to continue.
Is there a free option outside the trial?
Yes, in a limited form. Databricks Community Edition is described by the company as a free, limited functionality platform designed for anyone who wants to learn Spark.
Where are Databricks prices published for Azure China regions?
Not on the main Databricks pricing page. Databricks directs buyers to azure.cn/pricing for Databricks offerings in Azure China.
