Databricks Pricing Explained: The DBU Consumption Model
If you are budgeting a Databricks deployment, the first thing to understand is that there is no price list with named plans. Databricks bills pay-as-you-go, with no up-front cost, metered per second. You pay for what you run. The meter counts a single unit called the DBU, short for Databricks Unit, and your bill is the number of DBUs your workloads burn multiplied by a per-DBU rate that depends on the type of work and where it runs.
That model is flexible, but it makes cost estimation harder than picking a tier off a page. This breakdown walks through how the consumption model works, what a DBU actually measures, the starting per-DBU rates by workload, what falls outside the DBU charge, how Committed-Use discounts lower the rate, and why Azure Databricks shows up on a Microsoft invoice instead of a Databricks one.
The Consumption Model
Databricks does not sell seats or subscription tiers. It sells consumption. The moment a cluster spins up and starts processing, the meter runs; when the cluster terminates, it stops. Billing is calculated per second, so a job that runs for ninety seconds is billed for ninety seconds, not rounded up to an hour. There is no minimum commitment to start and no up-front license fee.
This is the same metered approach the major cloud providers use for raw infrastructure, applied one layer up to a data and AI platform. The practical consequence is that your Databricks cost is a function of three things you control: how much compute you provision, how long it runs, and which type of workload it is. A nightly pipeline that finishes in twenty minutes and shuts down costs a fraction of an always-on interactive cluster that a team leaves running through the workday.
Practitioner note: The biggest budgeting mistake teams make is treating Databricks like a fixed subscription. It is not. Your invoice tracks behavior, so idle clusters, oversized compute, and the wrong workload type are where money leaks. Auto-termination and right-sizing are your two most important cost controls, and both are configuration choices, not contract terms.
What Is a DBU?
A DBU, or Databricks Unit, is the abstraction that makes per-second metering possible across very different hardware. Databricks defines it as a normalized unit of processing power driven by processing metrics, which means the compute you use combined with the data you process. Rather than billing you directly against raw machine types, Databricks translates your activity into DBUs and charges a rate per DBU.
The reason this matters for budgeting is that two clusters can consume DBUs at very different speeds. A larger cluster burns DBUs faster per second than a small one, so doubling the cluster size to finish a job in half the time does not automatically halve the cost. You need to think in terms of total DBUs consumed, not just wall-clock runtime.
Databricks also references two related units in its pricing materials: the DSU, a storage unit, and the CU, a compute unit used for the operational database (Lakebase). For most analytics and AI workloads the DBU is the number that drives your bill, but if you run Lakebase you will see CU-based pricing as well.
Per-DBU Rates by Workload
The per-DBU rate is not flat. It depends on the type of workload, because different workloads run on different runtimes with different value. The table below lists the vendor-reported starting rates. These are entry points: your effective rate varies by cloud provider, by region, by whether you use serverless or classic compute, and by any contract discount you hold.
| Workload | What It Covers | Starting Rate |
|---|---|---|
| Data Engineering | Lakeflow Jobs and automated pipelines | $0.15 / DBU |
| Data Warehousing | SQL, both Classic and Serverless | $0.22 / DBU |
| Interactive | Data science and ML, notebook-driven work | $0.40 / DBU |
| Artificial Intelligence | Model serving, AI search, agents | $0.07 / DBU |
| Genie | AI assistant usage beyond the free allowance | $0.07 / DBU |
| Operational DB (Lakebase) | OLTP Postgres for AI agents, priced per CU | $0.069 / CU |
Vendor-reported starting rates. Verified 2026-06-09. Rates vary by cloud provider and region. Confirm current rates at databricks.com/product/pricing.
The spread is worth internalizing. Interactive data science at $0.40/DBU is the most expensive line item per unit, while AI serving and Genie sit at $0.07/DBU, roughly a sixth of the cost. That difference is not arbitrary: interactive clusters keep humans in the loop and tend to run longer with lower utilization, whereas AI serving is optimized for throughput. When you map a project's budget, the workload mix matters as much as the volume.
What the DBU Rate Does Not Include
The per-DBU rate is the Databricks platform fee. It is not your total bill. Two large cost categories sit outside the DBU charge, and missing them is the second most common budgeting error after ignoring idle clusters.
The upside of separating compute and storage this way is real: because your data lives in your own cloud account in open formats, you avoid being locked into a proprietary store, and you sidestep the egress fees you would pay to extract data from a closed warehouse. The tradeoff is that your true cost of ownership is the Databricks DBU charge plus your cloud provider's storage, networking, and compute lines, which you have to read together rather than on one invoice.
Committed-Use Discounts
The starting per-DBU rates are the on-demand prices. If your usage is predictable, Databricks offers Committed-Use Contracts that lower the effective rate in exchange for a spending commitment over a term. The more you commit, the larger the discount. This is the standard lever for organizations that have moved past experimentation into steady production volume.
The decision is the same one you face with reserved cloud capacity. A commitment only saves money if you actually consume what you reserved, so the right time to sign one is after you have a few months of consumption data showing a stable baseline. Committing early, before you understand your workload mix, risks paying for capacity you do not use.
Why Azure Databricks Is Billed by Microsoft
Databricks runs natively on AWS, Azure, and GCP, and the platform is functionally the same on all three. Billing, however, is not. On AWS and GCP, Databricks bills you directly on a pay-as-you-go basis against your DBU consumption. On Azure it works differently, and this trips up a lot of first-time buyers.
Azure Databricks is a first-party Microsoft service. Microsoft has offered it as an integrated Azure service since 2017, which means its pricing is set and billed by Microsoft and governed by your Azure subscription terms. The Databricks charges appear on your Azure invoice alongside the rest of your Microsoft cloud spend, not as a separate Databricks bill.
Practitioner note: If you are on Azure, do not budget from the Databricks pricing page alone. The DBU concept is the same, but the actual rates and any enterprise agreement discounts come through Microsoft and your Azure commitment. Check azure.com for Azure Databricks pricing, and factor it into your existing Azure consumption rather than treating it as a standalone vendor.
The practical takeaway is that your procurement path determines your pricing path. AWS and GCP customers contract with Databricks. Azure customers effectively contract with Microsoft for the same platform. Neither is cheaper by default, but the invoice, the discount mechanism, and the support relationship differ.
Controlling Your Spend
Because the bill follows behavior, the levers that move it are operational, not contractual. Here is where practitioners get the most savings, roughly in order of impact.
None of these require a contract negotiation. They are configuration and scheduling choices you control from day one, and together they typically matter more to your invoice than the headline per-DBU rates.
Frequently Asked Questions
How does Databricks pricing work?
Databricks bills pay-as-you-go with no up-front cost and per-second metering. Usage is priced in DBUs, a normalized unit of processing power driven by the compute used and the data processed. Storage and networking are billed separately by your cloud provider. On AWS and GCP, Databricks bills you directly on DBU consumption; on Azure, Microsoft sets and bills the price.
What is a DBU in Databricks?
A DBU, or Databricks Unit, is a normalized unit of processing power. Databricks describes it as driven by processing metrics, meaning the compute you consume plus the data you process. You are charged a per-DBU rate that varies by workload type, cloud provider, and region. The DBU is the platform fee; the underlying cloud compute, storage, and networking are billed separately.
What are the starting per-DBU rates?
Vendor-reported starting rates are: Data Engineering at $0.15/DBU, Data Warehousing (SQL) at $0.22/DBU, Interactive at $0.40/DBU, Artificial Intelligence at $0.07/DBU, Genie at $0.07/DBU, and Operational DB (Lakebase) at $0.069/CU. These are starting points and vary by cloud and region. Confirm current rates on the Databricks pricing page.
Why is Azure Databricks billed differently?
Azure Databricks is a first-party Microsoft service, so its pricing is set and billed by Microsoft and governed by your Azure subscription terms. The charges land on your Azure invoice rather than a separate Databricks bill. On AWS and GCP, Databricks bills directly on pay-as-you-go DBU consumption. The platform itself is the same across all three clouds.
Can a contract reduce Databricks costs?
Yes. Databricks offers Committed-Use Contracts that provide discounts at higher commitment levels in exchange for a spending commitment over a term. Beyond contracts, the largest cost levers are workload type, serverless versus classic compute, and right-sizing clusters so you are not paying for idle capacity.