Databricks pricing explained: DBUs, tiers, and the VM bill the calculator leaves out

Short answer: Databricks charges per DBU (Databricks Unit), billed per second, at a $/DBU rate that depends on the product, tier, cloud and sometimes region. On classic compute that's only part of the bill: your cloud provider separately charges for the VMs, disks and network underneath. On serverless compute the DBU rate already includes the VMs. For a typical classic jobs cluster, the VM bill was 43% to 69% of the total in our example below.

What is a DBU?

Databricks defines a DBU as "a normalized unit of processing power on the Databricks Lakehouse Platform used for measurement and pricing purposes" (Databricks pricing FAQ). In practice:

• Every classic instance type has a DBU rating per hour. A bigger VM burns more DBUs per hour. On AWS, an m5d.2xlarge (8 vCPU, 32 GB) is rated at 1.37 DBU/hour for Jobs Compute.
• Each product (Jobs, All-Purpose, SQL, serverless, pipelines) has its own $/DBU price. The same DBU costs very different amounts depending on where you spend it.
• Billing is per second, pay as you go, with discounts available through committed-use contracts (Databricks).

So the formula for classic compute is:

cost = hours × nodes × (DBU/hour per node × $/DBU + VM $/hour per node)

For serverless it's just DBUs consumed × $/DBU, because, in Databricks' words, "for serverless services, the DBU cost already includes the virtual machine costs" (Databricks cost optimization docs).

Current list prices per DBU

Here's what the main products cost on the Premium tier today, in a US East region on each cloud. These come straight from DBU Watch's daily pull of the Azure Retail Prices API and Databricks' published pricing data.

Classic AWS and Google Cloud rates are the same in every region. Serverless and SQL rates vary by region, so check the full price table for yours.

Tiers: Premium, Enterprise, and the end of Standard

The Standard tier is gone on every cloud now:

• AWS and Google Cloud: Standard workspaces had until October 1, 2025 to upgrade. Any left were moved to Premium automatically on that date (AWS docs, GCP docs). The tiers are now Premium and Enterprise.
• Azure: new workspaces had to be Premium from April 1, 2026, and remaining Standard workspaces were upgraded automatically on October 1, 2026 (Microsoft Learn). See our Standard-to-Premium cost guide.

One trap when comparing clouds: tier names don't line up. Databricks notes that "the Premium tier on Azure Databricks corresponds to the Enterprise tier on AWS and GCP" (Databricks pricing page). That's why Azure Premium Jobs Compute lists at $0.30/DBU while AWS Premium is $0.15. The like-for-like AWS number is Enterprise at $0.20/DBU (All-Purpose: $0.65 on AWS Enterprise vs. $0.55 on Azure Premium).

The VM bill the official calculator leaves out

This is where most estimates go wrong. Databricks' own pricing calculator says so in its fine print: "Serverless estimates include compute infrastructure costs. Non-serverless estimates do not include cost for any required AWS services (e.g., EC2 instances)" (Databricks pricing calculator). Databricks' engineering blog describes total cost of ownership the same way: platform costs, plus "cloud infrastructure costs, such as virtual machines, storage, and networking charges" (Databricks blog).

Azure is the partial exception. Microsoft's Azure Databricks pricing page shows per-instance totals that "include both Virtual Machines compute price and DBU price." But the headline $/DBU table on the same page is DBUs only, and that's the number people quote.

Example (one hour): Five m5d.2xlarge nodes on AWS Jobs Compute (Premium) consume 6.85 DBUs per hour: 6.85 × $0.15 = $1.03/hour to Databricks. AWS then bills the five EC2 instances at $0.452/hour each, $2.26/hour. Real cost: $3.29/hour.

Worked example: the same cluster on three clouds

Example, not a quote. Assumptions: 1 driver + 4 workers of a comparable 8-vCPU general-purpose instance, Jobs Compute, no Photon, Premium tier, 8 hours a day for 22 days (176 hours a month), on-demand VM list prices in each cloud's US East/Central region.

Two things jump out. First, on AWS and Google Cloud the VM line is bigger than the Databricks line for plain jobs compute, so a DBU-only estimate misses more than half the bill. Second, Azure looks pricier per DBU, but remember the tier mapping above and that Azure instances carry different DBU ratings. Compare total cost, not $/DBU.

What else isn't in the $/DBU price

• Storage and disks. Microsoft lists managed disks, blob storage and public IP addresses as billed in addition to VMs for Azure Databricks (Azure pricing page). On AWS and Google Cloud the equivalents are EBS/Persistent Disk and S3/GCS, billed by them.
• Networking. For serverless, Microsoft notes that data transfer, NAT gateway and Private Endpoint hours are billed on top of the DBU (Azure pricing page).
• Idle pool instances. Databricks doesn't charge DBUs for instances idling in a pool, but "instance provider billing does apply" (Databricks docs).
• Discounts. List prices are the ceiling. Databricks offers committed-use discounts (Databricks), Azure offers up to 37% off DBUs through 1- or 3-year pre-purchase (Azure), and the VM side can be cut with spot, reservations or savings plans. Our calculators let you type in your effective VM price.

Which compute type costs what

Which product a workload runs on is one of the biggest levers you have. On AWS Premium, the same DBU costs $0.15 on Jobs Compute, $0.55 on All-Purpose Compute and $0.35 on serverless jobs (VMs included). Databricks' own advice: scheduled, non-interactive work "will cost significantly less" on job compute than on all-purpose compute (Databricks docs). The two decisions worth modeling next:

• Serverless vs. classic compute: which is cheaper?
• Is Photon worth it? The break-even math

How to see what you actually pay

For your real numbers, query the billing system tables. system.billing.usage has every DBU you consumed, by SKU, job, cluster and tag (docs). system.billing.list_prices has the list price history per SKU so you can turn DBUs into dollars (docs). Neither includes your cloud provider's VM charges for classic compute. For those, you need your AWS Cost and Usage Report or Azure Cost Management export, as the Databricks blog explains. Then compare against list prices here and track changes on the price history page.