In what way do serverless compute resources differ from classic compute resources in Databricks?

Serverless compute resources in Databricks significantly diverge from classic compute resources primarily in their operational model. Serverless compute resources are designed to be ephemeral, meaning they are provisioned automatically and can scale according to the workload without the user having to manage or maintain the underlying infrastructure. They are dynamically allocated when needed and terminate when the processing is complete, which allows users to pay only for the compute resources they actually use.

This contrasts with the traditional compute resources, which involve dedicated instances that are often running continuously, leading to a reserved capacity for specific customers. In the case of classic compute resources, users must explicitly manage the scaling and provisioning, which can lead to underutilization or over-provisioning.

Therefore, option B accurately reflects that serverless compute resources are not always running but are instead utilized dynamically as needed, contrasting sharply with the reserved and always-on nature of classic resources. This flexibility makes serverless a scalable and cost-efficient choice for many workloads in Databricks. The other options mischaracterize the nature and accessibility of serverless compute resources within the Databricks ecosystem.