Do I pay only when my application is running?

Summary

• Yes. With Databricks serverless compute you are billed only when compute is assigned to your workload, and not for the time to acquire and set up instances.
• Idle stops the meter. For serverless notebooks and jobs, running a command or query incurs charges that stop automatically after the workload has been idle for a few minutes.
• Scale-to-zero means the intelligent serverless autoscaler winds resources down when they are no longer needed, so there is no charge for idle capacity.
• Usage is measured in DBUs (Databricks Units), and serverless SQL warehouses and Model Serving both scale automatically with demand and terminate when idle.
• You can monitor and cap spend with the system.billing.usage system table, cost dashboards, and budget alerts and policies.

Do I pay only when my application is running?

Yes. On the Databricks Platform, serverless compute uses a pay-per-use model: you are billed only when compute is assigned to your workload, and you are not billed for the time it takes to acquire and set up compute instances. When your workload stops or goes idle, the meter stops. Usage is measured in DBUs (Databricks Units), a normalized measure of processing power, while Databricks manages and autoscales the underlying infrastructure.

Why serverless bills only for work done

• Billed for work, not idle time. You pay only when compute is actively assigned to a workload. For serverless notebooks and jobs, you are not billed simply for viewing a notebook; running a command or query incurs charges that stop automatically once the workload has been idle for a few minutes.
• Scale-to-zero. The intelligent serverless autoscaler provisions the right amount of capacity to meet demand and reduces provisioned capacity in graduated steps after a few minutes when it predicts resources are no longer needed. This eliminates charges for idle capacity, so you pay only for resources actively processing your workloads.
• Serverless SQL warehouses. Serverless SQL warehouses start and scale up in seconds and terminate when idle, so interactive analytics and BI are available on demand without paying for a warehouse that is sitting unused.
• Model Serving. Model Serving runs on serverless compute and automatically scales up and down to meet changes in demand, so you are not paying for capacity when there is no traffic.
• Unified cost visibility and controls. Databricks consolidates cost components into a unified billing view. You can monitor usage with the system.billing.usage system table, use pre-built cost dashboards, and set budget alerts and policies to keep spend in check.

Getting started

• Read What is serverless computing? for how pay-per-use and scale-to-zero work.
• Turn on serverless compute for notebooks, jobs, SQL warehouses, and Model Serving.
• Review cost optimization best practices to right-size and control spend.
• Set up budgets and alerts so you are notified before spend exceeds a threshold.

FAQs

Am I charged when my workload is idle?

No. For serverless notebooks and jobs, charges stop automatically after the workload has been idle for a few minutes, and scale-to-zero removes charges for idle capacity.

How is serverless usage measured?

Usage is measured in DBUs (Databricks Units), a normalized measure of processing power, billed only for the compute assigned to your workloads.

Do serverless SQL warehouses and Model Serving also stop billing when idle?

Yes. Serverless SQL warehouses terminate when idle, and Model Serving scales automatically with demand, so you are not paying for unused capacity.

How can I keep serverless spend under control?

Monitor usage with the system.billing.usage system table and cost dashboards, and set budget alerts and policies to cap and track spend.