On this page 9 sections
Databricks pricing is pay-as-you-go: you pay for the Databricks Units (DBUs) you use, billed per second, plus your cloud provider's charges for servers (on classic compute), storage and networking. In our worked example below, a small daily data pipeline and a small SQL warehouse come to about USD 184 a month before storage and network charges.
The prices in this guide are Azure Databricks list prices for South Africa North, checked on 31 October 2026. Rand costs vary with the exchange rate, and your own bill depends on how long your compute runs.
Short answer
- A DBU is a unit of processing power. The more compute you use, and the longer it runs, the more DBUs you use.
- Each DBU has a price, which depends on the type of work, the plan tier and the region.
- On classic compute you also pay the cloud provider for virtual machines. On serverless compute you don't buy machines, but some network charges are extra.
- Storage is billed by your cloud, not in DBUs.
- Budgets, alerts and sensible auto-stop settings matter more than the rate card.
What is a DBU?
A DBU is "a normalized unit of processing power" used for measurement and pricing, according to Microsoft's Azure Databricks glossary (opens in a new window). Microsoft adds that it is billed on per-second usage, and that how many DBUs you use depends on the size and type of machine running (Azure Databricks pricing (opens in a new window)).
Think of electricity. A kilowatt-hour is the unit, a bigger appliance uses more of them, and the price per unit depends on your tariff. A DBU works the same way.
What decides the price per DBU?
Three things: the type of work (called the workload), the plan tier and the region. On Azure, Microsoft lists the Premium tier only. Its Standard tier is being retired, and Databricks says Premium on Azure corresponds to Enterprise on AWS and Google Cloud (Databricks (opens in a new window)).
Here are the Premium rates for South Africa North, in US dollars per DBU-hour:
| Workload | What it is for | USD per DBU-hour |
|---|---|---|
| Jobs Light Compute | Lighter automated jobs | 0.22 |
| SQL Compute | SQL warehouse | 0.22 |
| Jobs Compute | Automated jobs and pipelines | 0.30 |
| Automated Serverless Compute | Automated jobs, serverless | 0.52 |
| All-Purpose Compute | Interactive work in notebooks | 0.55 |
| SQL Pro Compute | SQL warehouse, Pro type | 0.74 |
| Serverless SQL | SQL warehouse, serverless type | 0.95 |
| Interactive Serverless Compute | Notebooks, serverless | 1.05 |
Region matters. In the page's default region (West US), Serverless SQL is listed at USD 0.70, compared with USD 0.95 in South Africa North. Turning on Photon, Databricks' faster query engine, increases the DBU count.
Classic or serverless compute?
With classic compute, Microsoft bills you for the virtual machines in your clusters plus the DBUs. Disks, blob storage and public IP addresses are billed too.
With serverless compute, Databricks manages the servers for you. Microsoft's table lists no separate virtual machine charge for serverless, but data transfer, NAT gateway and Private Endpoint usage are billed in addition.
What you pay, and to whom
| Cost component | Who bills it | How it is measured |
|---|---|---|
| DBUs (classic compute) | Databricks, through Azure | DBU-hours, billed per second |
| Virtual machines (classic only) | Microsoft Azure | Hours the machines run |
| Serverless compute | Databricks, through Azure | DBU-hours at the serverless rate |
| Storage, disks, public IPs | Microsoft Azure | Per GB or per hour |
| Data transfer and network extras | Microsoft Azure | Per GB or per hour |
| Optional add-ons | Databricks, through Azure | 10% of list price (Enhanced Security and Compliance) or 30% (Mission Critical, with a 50% promotional discount until 30 June 2027) |
A worked example (estimate)
This example is an estimate, not a quote. It uses the South Africa North list prices above, pay-as-you-go, and the standard 730-hour month.
Daily pipeline on classic Jobs Compute. Two D4s v5 machines (one driver, one worker), each using 1 DBU per hour. The job runs 30 minutes a day for 30 days, which is 15 hours.
A small SQL warehouse on Serverless SQL. Microsoft lists a 2X-Small warehouse at 4 DBUs per hour (Microsoft Learn (opens in a new window)). It runs 2 hours a day on 22 working days, which is 44 hours.
| Item | Calculation | USD per month |
|---|---|---|
| Pipeline DBUs | 2 machines × 1 DBU × 15 h × 0.30 | 9.00 |
| Pipeline virtual machines | 2 machines × 15 h × about 0.254 | 7.62 |
| SQL warehouse DBUs | 4 DBU × 44 h × 0.95 | 167.20 |
| Total before storage and network | about 184 |
The virtual machine rate is worked out from Microsoft's table: the D4s v5 total of USD 586.92 a month, less the USD 401.50 DBU part.
Notice that the SQL warehouse is almost the entire bill. If it ran around the clock instead (about 730 hours), the same warehouse would cost about USD 2,774 a month (730 × 4 × 0.95). The lesson: how long compute runs matters more than the rate card.
To estimate rand, multiply by the exchange rate on the day. Microsoft converts non-dollar prices using month-end rates, so rand prices can differ from a simple conversion.
Can you try it for free?
There are two options, and only one is meant for business use.
- Free Edition is for non-commercial use. It is serverless only, limited to one 2X-Small SQL warehouse, and has no guaranteed reliability, support or SLA (Databricks docs (opens in a new window)). It suits learning, not testing a business workload.
- A trial gives credits valid for 14 days. If you don't terminate your compute and remove your card, the account moves to pay-as-you-go when the credits run out (Databricks docs (opens in a new window)).
How to keep costs under control
Databricks lists several cost management tools (opens in a new window):
- Usage dashboards that show where the spend is going.
- Budgets that email you when monthly spend crosses a threshold. They support up to four thresholds, alerts can take up to 24 hours, and Databricks says they are not "an absolute spend cap" (budget docs (opens in a new window)).
- Tags that attribute spend to teams or projects.
- Compute policies that limit the type and size of compute people can create.
- Overspend protection such as the default 2.5-hour time-out on serverless notebooks.
Two more habits help. Make sure warehouses and clusters stop when idle, and consider a 1-year or 3-year pre-purchase plan once your usage is steady. Microsoft advertises savings of up to 37% over pay-as-you-go DBU prices.
Not sure Databricks is the right tool at all? Read Do you need Databricks?. Planning a move? See Moving to Databricks from SQL Server or Excel.
How Mammoth Digital can help
Our team works with Databricks and includes a Databricks Certified Data Engineer Professional. We can estimate your monthly cost before you commit, set up budgets and alerts, and tell you honestly if a simpler option would do.
See our Databricks consulting and Databases & Data pages, or get in touch. We'll reply within 2 working days.
Databricks is a trademark of Databricks, Inc. Microsoft and Azure are trademarks of Microsoft Corporation. Mammoth Digital is independent and not a partner of these companies.