How does serverless pricing work?

Summary

• Databricks serverless compute uses a pay-per-use model billed at per-second granularity, with no up-front costs and no charge for idle capacity.
• You are billed in DBUs (Databricks Units) — a normalized unit of processing capability per hour — for the compute you actually consume, while Databricks manages and autoscales the underlying infrastructure.
• Serverless Notebooks and Jobs bill only when a command or query runs (plus a small session fee for the REPL VM); charges stop automatically after a few minutes of idle time.
• Serverless SQL warehouses bill at a fixed DBU-per-hour rate based on warehouse size, and other serverless capabilities (jobs, Lakeflow pipelines, predictive optimization, Databricks Apps) each carry their own DBU rate.
• You can monitor exact consumption and cost by querying the system.billing.usage system table.

How does serverless pricing work?

Serverless pricing is a pay-per-use model: you pay only for the compute you actually consume, billed at per-second granularity, with no up-front costs and no charge for idle capacity. Databricks runs and scales the underlying infrastructure for you, so instead of paying for provisioned servers that sit idle, you pay for the processing your workloads use while they run. Consumption is measured in DBUs (Databricks Units), a normalized unit of processing capability per hour, and each serverless service has a published DBU rate. For teams that want volume discounts, Databricks also offers committed-use contracts.

Why Databricks serverless pricing is predictable

• Pay only for what you consume. Serverless compute is billed per second of use with no up-front commitment. Databricks manages the infrastructure and uses real-time scaling to add and remove resources based on demand; the autoscaler winds resources down when they are no longer needed, so you are not charged for idle time.
• DBUs are the unit of measure. A DBU is a normalized unit of processing capability per hour. Your bill reflects the DBUs your workloads consume multiplied by the price of the relevant serverless SKU, so cost tracks directly with usage.
• Notebooks and Jobs bill only when work runs. With Serverless Notebooks and Jobs you are not billed for simply viewing a notebook. A notebook connected to serverless incurs a small session fee for its REPL VM, and running a command or query adds compute charges. Charges stop automatically once the workload has been idle for a few minutes.
• SQL warehouses bill by size. Serverless SQL warehouses bill at a fixed DBU-per-hour rate determined by the warehouse size you choose, so you can size for the concurrency and performance you need.
• Coverage across the platform. Many serverless capabilities carry their own DBU rate, including Serverless Jobs, Lakeflow (Spark Declarative) Pipelines, Predictive Optimization, Data Quality Monitoring, Fine-Grained Access Control, materialized views and streaming tables in Databricks SQL, Databricks Apps, and serverless feature-store materialization.
• Transparent cost monitoring. You can see exactly what you consumed by querying the system.billing.usage system table, which breaks DBU consumption and charges down by user, workload, and other dimensions. Databricks recommends benchmarking representative workloads and reviewing billing data to estimate ongoing costs.

Getting started

• Review the Databricks pricing page for current serverless rates and SKUs.
• Learn how to connect to and use serverless compute.
• Understand serverless workspace billing and costs and how charges appear.
• See per-product rates such as Lakeflow Jobs pricing.

FAQs

What is a DBU?

A DBU (Databricks Unit) is a normalized unit of processing capability per hour used to measure and price the compute your workloads consume. Your serverless bill is the DBUs used multiplied by the price of the relevant SKU.

Do I pay for idle serverless compute?

No. Databricks automatically scales serverless resources down when they are not in use, and for Serverless Notebooks and Jobs charges stop after a few minutes of inactivity, so you are not billed for idle time.

How am I billed for Serverless Notebooks and Jobs?

You are not billed for viewing a notebook. Connecting a notebook to serverless incurs a small session fee for the REPL VM, and running a command or query adds compute charges that stop automatically once the workload is idle for a few minutes.

How are serverless SQL warehouses priced?

Serverless SQL warehouses are billed at a fixed DBU-per-hour rate based on the warehouse size you select.

How can I track serverless costs?

Query the system.billing.usage system table to see DBU consumption and charges by user, workload, and other dimensions. Databricks recommends benchmarking representative workloads and analyzing billing data to estimate costs.