How can I control or reduce my Databricks lakehouse costs?

Summary

• Databricks uses consumption-based pricing measured in Databricks Units (DBUs) with per-second billing, so you pay only for what you use — the main levers for control are observing, attributing, controlling, and optimizing that consumption. See Databricks pricing.
• See your spend. Billing system tables record granular usage with tags and identity, and prebuilt cost AI/BI dashboards give account admins centralized spend monitoring.
• Attribute cost. Tags (key:value) propagate to billing records, and serverless usage policies (Public Preview) auto-tag serverless usage so you can attribute spend to teams and projects.
• Set guardrails. Budgets and budget alerts, budget policies, and compute policies let you cap spend, enforce tags, and control which compute users can create; a default serverless notebook timeout guards against runaway execution.
• Optimize compute. Serverless compute scales to zero when idle, autoscaling and auto-termination right-size clusters, and predictive optimization automates storage upkeep. Follow the cost optimization guide.

How can I control or reduce my Databricks lakehouse costs?

The Databricks lakehouse is billed on a consumption-based model: usage is measured in Databricks Units (DBUs) at per-second granularity, so you pay only for the compute and resources you actually use, with no upfront license. That means controlling cost is about managing consumption, and it works best as four stages: see your spend, attribute it to teams and projects, put guardrails in place, and optimize the compute that drives it. Databricks provides built-in tooling for each stage. For current rates, see Databricks pricing.

Why the Databricks Lakehouse gives you control over cost

• Consumption-based, pay-as-you-go. You pay only for what you use, billed per second in DBUs, with no idle license to carry. Because the rate depends on the workload and compute type, matching the right compute to each job is the primary cost lever. See Databricks pricing.
• Cost observability with system tables and dashboards. Billing system tables — such as system.billing.usage — store granular usage logs including resource metadata, custom tags, and identity. Account admins can import prebuilt cost management AI/BI dashboards into any Unity Catalog-enabled workspace for centralized spend monitoring, and the Governance Hub (in Beta) consolidates spend and cost drivers on a dedicated Cost page. See cost management tools on Databricks.
• Cost attribution with tags. Tags formatted as key:value pairs propagate to billing records so you can attribute cost to users, groups, and projects. Serverless usage policies (in Public Preview) automatically tag serverless compute for the users they are assigned to, and admins can enforce custom tagging through compute policies. See monitor and control cost.
• Spending guardrails. Set account-wide or filtered budgets with alerts to get notified when a threshold is crossed; use budget policies to enforce tags on serverless usage and govern who can consume serverless resources; and use compute policies to control the type and size of compute users can create. A default execution timeout on serverless notebooks protects against long-running queries consuming excessive resources.
• Serverless scale-to-zero and autoscaling. serverless compute is fully managed and scales to zero when idle, so you are billed only while a workload runs. Autoscaling and auto-termination right-size and release resources automatically, and Standard mode is cost-optimized for workloads that do not need Performance mode.
• Continuous optimization. Predictive optimization automates data layout and storage maintenance so you do not pay to manage it by hand, Photon (the vectorized query engine) runs by default on SQL warehouses and serverless compute to accelerate SQL and analytics workloads, and running scheduled production work on jobs compute keeps steady workloads efficient. The cost optimization guide describes how to right-size compute and keep spend efficient.

Getting started

• Review Databricks pricing to understand DBU rates by product and tier.
• Import a cost management dashboard or query the billing system tables to see where your spend goes.
• Apply tags and budget policies for attribution, and set budgets with alerts to cap and monitor spend.
• Turn on serverless compute so idle compute scales to zero, and follow the cost optimization guide to right-size workloads.

FAQs

How is Databricks lakehouse cost measured?

Usage is measured in Databricks Units (DBUs) and billed on a consumption-based, pay-as-you-go model at per-second granularity, so cost reflects the specific workloads you run rather than a fixed license fee.

How do I see where my Databricks spend is going?

Use the billing system tables, such as system.billing.usage, and the prebuilt cost management AI/BI dashboards for centralized monitoring. The Governance Hub (in Beta) consolidates spend and cost drivers on a dedicated Cost page.

How do I stop runaway spend?

Set budgets with alerts, use compute and budget policies to control which resources users can create and enforce tagging, and rely on serverless scale-to-zero and the default serverless notebook timeout to avoid paying for idle or long-running compute.

Does serverless compute reduce cost?

You pay only while a workload is running, and serverless scales to zero when idle. Autoscaling and auto-termination release capacity automatically, so there is no idle compute to pay for.