How does serverless differ from traditional servers or compute?

Summary

• With serverless compute, Databricks manages all the underlying infrastructure automatically, so you focus on your data and workloads instead of provisioning and tuning clusters; with traditional (classic) compute, you configure clusters yourself, including instance types, autoscaling, and policies.
• Serverless compute starts in seconds because Databricks keeps warm pools of instances ready, rather than waiting for a cluster to provision.
• Serverless scales automatically, both horizontally (more executors) and vertically (larger executors), and scales down when work completes, so you pay only for the compute you actually use with no idle capacity to reserve.
• Databricks automatically applies runtime upgrades and platform optimizations to serverless compute, so there are no manual version upgrades or Spark configurations to maintain.
• Serverless is available across SQL warehouses, jobs, notebooks, and model serving, and Databricks reports that most customers achieve a 25% or greater reduction in compute spend with serverless.

How does serverless differ from traditional servers or compute?

The core difference is who manages the infrastructure. With serverless compute, Databricks provisions, scales, and maintains the underlying resources for you: the platform monitors what a workload needs, allocates exactly that, and scales it back down when demand falls. With traditional or classic compute, you decide and manage those resources yourself, including instance types, autoscaling ranges, cluster policies, and runtime upgrades. Serverless is designed to remove that operational overhead so teams can focus on data and analytics rather than on running clusters.

Why Databricks serverless compute

• No infrastructure to manage. Databricks handles infrastructure automatically, so administrators no longer need to adjust quotas, create and maintain network resources, or connect billing sources. With classic compute you configure worker and driver instance types, compute policies, and scaling parameters manually.
• Startup in seconds. Databricks keeps warm pools of instances ready, so serverless compute starts in seconds rather than waiting for a cluster to provision, giving near-zero startup latency and improved query concurrency.
• Automatic scaling. Serverless autoscaling responds automatically to workload needs and scales both horizontally (more executors) and vertically (larger executors). Classic compute autoscaling scales horizontally, and you select instance types yourself.
• Scale to zero and pay per use. Serverless scales down automatically when a workload completes, so you pay only for the compute you actually use and there is no idle capacity to reserve. With classic compute you pay for provisioned capacity, including idle time, which often means reserving resources for peak demand.
• Automatic updates. Databricks manages Spark runtime upgrades and applies platform-level optimizations to serverless compute without customer intervention, while classic compute requires you to plan and run Databricks Runtime upgrades yourself.
• Simple, broad coverage. There is no need to pick instance types, scaling parameters, or Spark configs; it is handled for you. Serverless is available for SQL warehouses, jobs, notebooks, and model serving, with Photon and autoscaling enabled automatically for serverless jobs.

Because serverless removes idle capacity and management overhead and uses a single, simplified bill, Databricks reports that most customers achieve a 25% or greater reduction in compute spend with serverless.

Getting started

• Read What is serverless computing? for the concept and benefits.
• Compare options in Serverless versus classic compute to see when to use each.
• Review Use serverless architectures and Choose compute for your jobs.

FAQs

What does serverless compute mean on Databricks?

It means Databricks fully manages the underlying infrastructure: it provisions, scales, and maintains compute automatically, so you run workloads without configuring or operating clusters yourself.

Do I still choose instance types and cluster sizes with serverless?

No. With serverless you do not pick instance types, scaling parameters, or Spark configurations; Databricks handles all of that. Those choices remain manual with classic compute.

How does serverless billing work?

Serverless scales to zero when a workload finishes and you pay only for the compute you actually use, so there is no idle capacity to reserve or pay for.

Which workloads can run on serverless?

Serverless compute is available for SQL warehouses, jobs, notebooks, and model serving, so most common Databricks workloads can run without managing clusters.