Is serverless more expensive than classic compute?

Summary

• Pay-as-you-go by design. Serverless compute is billed only for the compute actively doing work, measured in Databricks Units (DBUs), with no charge for idle or provisioned capacity.
• Cost tracks your workload. Whether serverless is the lower-cost option for a given job depends on its usage pattern; because serverless bills for work done and scales down automatically, cost follows actual demand rather than an always-on footprint.
• Intelligent autoscaling and scale-to-zero. Databricks manages the infrastructure and scales capacity up and down in real time, winding resources down after a few minutes of idle so you are not paying for unused headroom.
• DBU-based, tiered pricing. Rates vary by product and platform tier; see the Databricks pricing pages for current rates.
• Built-in cost controls. Budget policies (automatic tagging), budgets and alerts, prebuilt usage dashboards, and the system.billing.usage table let you attribute, monitor, and cap serverless spend.

How serverless compute is priced

There is no single yes-or-no answer, because serverless compute is billed on a pay-as-you-go model: you pay only for the compute that is actively doing work, measured in Databricks Units (DBUs), and you are not billed for idle or provisioned capacity. Whether serverless is the lower-cost choice for a given workload depends on that workload's usage pattern. Because serverless bills for work done and winds down automatically, its cost tracks actual demand rather than an always-on footprint. The choice between serverless and classic compute for a given workload comes down to usage patterns and operational preferences rather than a fixed price gap. For current rates, see the Databricks pricing pages.

Why the serverless cost model works this way

• You pay for work done, not idle time. Serverless bills only when compute is assigned to your workload, not for the time to acquire and set up instances, and charges stop automatically once a workload has been idle for a few minutes. Connecting a notebook to serverless incurs only a small session fee until you run a command. See the serverless compute GA announcement.
• DBU-based, usage-driven pricing. Databricks charges in DBUs, which measure the processing power used to run your workload. Rates vary by product and platform tier; the platform tiers and add-ons pricing page lists current rates.
• Intelligent autoscaling and scale-to-zero. Databricks manages the underlying infrastructure and uses real-time signals about your workload to scale capacity up and down in graduated steps, then winds resources down when they are no longer needed, so you are not paying for unused headroom.
• Cost controls and attribution. Budget policies automatically apply tags to serverless resources for chargeback and reporting; budgets and alerts notify you when spending crosses a threshold; and prebuilt usage dashboards plus the system.billing.usage table give detailed, queryable cost visibility. See attributing serverless costs with budget policies and cost management with billing system tables.
• Guardrails against overspend. Defaults such as an execution timeout on serverless notebooks help contain long-running work, and admins can tune these controls. See cost savings with serverless compute.

Getting started

• Review current rates on the Databricks pricing pages.
• Set up budgets and budget policies to attribute and cap serverless spend, and import the prebuilt usage dashboards.
• Monitor detailed usage with the billing system tables.
• Read the serverless compute GA announcement for how billing works.

FAQs

How is serverless compute billed?

On a pay-as-you-go model measured in DBUs: you pay only for compute actively doing work, with no charge for idle or provisioned capacity, and charges stop after a workload has been idle for a few minutes.

What determines my serverless cost?

Your usage pattern. Because serverless bills for work done and scales down automatically, cost tracks actual demand and the DBU rate for the product and tier you use, rather than a fixed, always-on footprint.

Can I control and predict serverless spend?

Yes. Budget policies apply tags automatically for chargeback, budgets and alerts flag overspend, and prebuilt dashboards plus the system.billing.usage table give detailed cost visibility.

Where do I find serverless pricing rates?

On the Databricks pricing pages, including the platform tiers and add-ons page, which list current DBU rates by product and tier.