Databricks Billing Explained: DBUs, Cloud Costs and Invoices

Databricks billing confuses finance teams for one simple reason: there are usually two bills for the same workload. Databricks charges for processing in DBUs, and on classic compute your cloud provider charges for the machines underneath. Add a dozen compute types, each with its own DBU price, and it's easy to see why "what does Databricks cost us?" rarely gets a straight answer. This guide gives you one: how DBUs are priced, what runs up the cloud side, why bills don't drop when usage does, and where to find the numbers.

It's written for the people who pay the bill. If you're still deciding whether Databricks is the right platform, start with our Databricks practice page. If you also run Snowflake, its credits work quite differently, as our Snowflake billing guide explains.

How does Databricks billing work?

Five rules cover most of it:

1. Usage is measured in Databricks Units (DBUs), a normalised unit of processing, billed per second.
2. Each kind of compute has its own DBU price: jobs, interactive notebooks, SQL warehouses and serverless all differ, and so do tiers and regions.
3. Classic compute is two bills. Databricks charges DBUs; your cloud provider charges for the virtual machines, disks and IPs in your account.
4. Serverless compute runs in Databricks' account, so there are no VMs for you to pay for separately.
5. Pay-as-you-go is billed monthly to a card, and commitments trade an upfront promise for a discount.

For pay-as-you-go accounts, Databricks bills your card monthly at list prices, and may charge more often if fees build up, with the balance billed at month end. (Source)

What does a DBU cost?

Prices vary by cloud, so here's one concrete table: Azure Databricks Premium, pay-as-you-go, per DBU-hour, as published on 9 November 2026.

Compute type	East US	South Africa North
Jobs Compute (scheduled pipelines)	USD 0.30	USD 0.30
All-Purpose Compute (interactive notebooks)	USD 0.55	USD 0.55
Automated Serverless Compute	USD 0.45	USD 0.52
SQL Pro Compute	USD 0.55	USD 0.74
Serverless SQL	USD 0.70	USD 0.95

The same work costs nearly twice as much as an interactive job (USD 0.55) as it does as a scheduled job (USD 0.30), before you count the VMs. Moving production pipelines off notebooks and onto jobs is one of the cheapest savings available.

Azure's Standard tier was retired on 1 October 2026, so Premium prices are now the ones to plan with. (Source)

Why is classic compute two bills?

On classic compute, clusters run as virtual machines inside your cloud account. So on top of the DBUs, you pay your cloud provider for:

• the virtual machines in the cluster, priced by instance type;
• managed disks and blob storage;
• public IP addresses.

Turning on Photon, Databricks' faster query engine, also increases the DBU count on classic compute, so check it pays for itself in shorter run times.

On Azure, both halves land on your Azure bill as separate lines, and Microsoft's own pricing note says the total price "includes both Virtual Machines compute price and DBU price". (Source) On AWS and Google Cloud, the VM side comes from the cloud provider and the DBUs from Databricks. Databricks even warns that during a free trial, your cloud provider still charges for the compute instances in your account. (Source)

The practical consequence: the Databricks console only shows half the cost of classic compute. Any cost review that ignores the cloud bill understates it.

How are serverless and SQL warehouses billed?

Serverless compute runs in a compute plane managed by Databricks, without provisioning anything in your cloud account, and it starts faster and spends less time idle. (Source) There are no VMs to pay for: on Azure, the only charges on top of serverless DBUs are data transfer, NAT gateway and Private Endpoint usage. (Source)

SQL warehouses are billed by size in DBUs per hour. On Azure, the smallest Serverless SQL warehouse (2X-Small) uses 4 DBUs an hour, which is USD 2.80 an hour in East US and USD 3.80 in South Africa North. (Source)

The setting that matters most is Auto Stop:

Warehouse type	Default auto-stop	Minimum
Pro and Classic	45 minutes	10 minutes
Serverless	10 minutes	5 minutes in the UI, 1 minute via the API

Databricks is explicit that idle SQL warehouses keep accumulating DBU and cloud instance charges until they stop. A Pro warehouse left on the 45-minute default bills for three-quarters of an hour after every last query. (Source)

What do commitments and pre-purchase save?

On Azure, you can pre-purchase Databricks Commit Units (DBCUs) for one or three years and save up to 37% on pay-as-you-go DBU prices. DBU usage across workloads and tiers draws down from the commitment until it's used up. (Source)

Two cautions. The discount applies to DBUs, not to the VMs on your cloud bill, which need their own reservations or savings plans. And a commitment sized on a busy pilot month can leave you paying for units you never use. Size it on a few months of real usage.

Why didn't our Databricks bill go down?

The usual culprits:

• Idle compute. Interactive clusters without an aggressive auto-termination, and SQL warehouses on the 45-minute default, keep billing after people go home.
• Notebooks doing production work. All-Purpose DBUs cost more than Jobs DBUs for the same processing, and Photon raises the DBU count on both.
• The cloud side. You cut DBUs, but the VMs, disks and IPs on your cloud bill didn't move, or a cluster's instance type got bigger.
• Budgets at list price. Budgets are measured in USD at the SKU list price and don't factor in credits or negotiated discounts, so they won't match your invoice exactly. (Source)
• Late and corrected records. Usage records typically arrive within 12 hours, and corrections are posted as a retraction that cancels the original plus a restatement. A query that ignores retractions double-counts. (Source)

Where do I find Databricks invoices and usage?

What	Where	Who
Invoice (Azure Databricks)	Azure Cost Management + Billing, alongside your other Azure charges	Azure billing roles
Payment method (pay-as-you-go on AWS)	Account console » Subscription & billing (credit card or AWS Marketplace)	Account admins
Detailed usage	The system.billing.usage table	Anyone granted access
Budgets and alerts	Account console » Budgets	Account admins

If you have a contract account, billing and subscription changes go through your Databricks account team, not the console. (Source)

Budgets send email alerts when spending crosses a threshold, and can be scoped by workspace, product or custom tag, so tag clusters by team or project from day one. (Source)

Is tax included, and what should South African buyers watch?

The prices above exclude tax. For South African companies, four more things:

• Azure Databricks is available locally. It runs in South Africa North (Johannesburg). (Source) Some DBU rates there are higher, notably SQL Pro and Serverless SQL. Databricks on AWS doesn't list AWS's Cape Town region.
• It's a US-dollar bill. Azure prices South African Microsoft Customer Agreement usage in USD, and Microsoft says that usage "will have to be paid in US dollars", so the rand cost moves with the exchange rate. (Source)
• Card fees. Pay-as-you-go accounts paying by card can pick up cross-border and forex fees on every charge.
• VAT treatment. How VAT on a foreign cloud service lands in your books is a question for your accountant. Ask before the first invoice.

A one-page checklist for finance

• Get both halves of the cost: DBUs and the cloud bill for classic compute.
• Move scheduled production work from All-Purpose to Jobs compute.
• Set Auto Stop on every SQL warehouse, well below the 45-minute default where usage is bursty.
• Set auto-termination on every interactive cluster.
• Tag compute by team and project, and create budgets with email alerts per workspace.
• Size any commitment on several months of real usage, not a pilot.
• Budget the bill in USD, and agree the VAT treatment with your accountant.

Get the bill under control, then get the value

Billing is the admin. The return comes from what your lakehouse does for the business: pipelines that run on time, governed data people trust, and models and agents built on it. Our Databricks partner practice designs, builds and runs Databricks platforms through our cloud data platforms service, from compute policies and cost guardrails to the pipelines and dashboards on top.

Want a second opinion on what your Databricks bill should look like? Book a Free AI Assessment and we'll size it with you.