Microsoft Fabric vs Databricks: how they differ and how each bills

Microsoft Fabric is a software-as-a-service analytics suite that you buy as one fixed capacity shared by every workload, Power BI included; Azure Databricks is a data and AI platform where you pay for the compute each job and query uses. Both store tables as Delta LakeDelta LakeAn open table format that turns Parquet files in cloud storage into a table with transactions, a version history, schema checks, and updates and deletes, so a data lake behaves like a database. and both run Spark, so the choice is less about features than about how your organisation works and how your bill is shaped.

What each is built for

Fabric puts Data Factory pipelines, Spark notebooks, a SQL data warehouse, real-time analytics, databases and Power BI in one place, over one store called OneLake. There are no clusters to size and no Azure resources to manage; you do not even need an Azure account. It suits organisations that already live in Power BI and Microsoft 365 and want one bill.

Databricks is a LakehouseLakehouseOne store for all of an organisation's data: cheap files in cloud storage, as in a data lake, with the tables, transactions and fast queries of a data warehouse on top. platform for engineers and data scientists: Spark and Delta Lake, Lakeflow pipelines and jobs, MLflow and SQL warehouses, governed by Unity CatalogUnity CatalogDatabricks' governance layer: one place that names every table, file volume, model and function as catalog.schema.object, decides who may use each, and records lineage and access.. You choose the compute for each workload, or use serverless. It runs on Azure, AWS and Google Cloud.

How Fabric bills: capacity units

You buy an F SKU: F2 has 2 capacity units (CUs), F64 has 64, up to F8192. In East US a CU costs $0.18 an hour pay as you go, billed per second, so an F64 is $8,409.60 a month if left running. A 1 or 3-year reservation cuts that by about 41%, to $5,002.67, but bills whether you use it or not. A pay-as-you-go capacity can be paused at night.

Every workload draws on the same CUs. For Spark, 1 CU is 2 Spark vCores, so an F64 has 128 and can burst to 384. To stop a big job from failing, Fabric smooths its cost: background jobs over 24 hours, interactive ones over 5 to 64 minutes. If smoothed use still exceeds the capacity, Fabric allows 10 minutes of overage, then delays interactive work by 20 seconds, then rejects interactive requests, then rejects everything. So one heavy Spark job can slow everyone's Power BI reports on the same capacity.

The licence detail that often decides it: on F64 and above, people with a free licence can view Power BI reports. Below F64, every viewer needs a Power BI Pro or Premium Per User licence.

How Databricks bills: per use

You pay Databricks for DBUs and, on classic compute, Azure for the VMs. In East US on Premium, jobs compute is $0.30 a DBU, so a Standard_D8ds_v5 worker (8 vCPUs, 32 GiB, 2 DBUs an hour, $0.452 for the VM) costs about $1.05 an hour. A Small serverless SQL warehouse uses 12 DBUs an hour at $0.70, $8.40 an hour. Nothing bills while nothing runs. The Databricks cost calculator prices every worker type and region.

A worked month

A team runs a nightly Spark job on 1 driver and 8 workers for 3 hours (27 node-hours a day), and analysts query a SQL warehouse 10 hours each working day (22 days).

• Databricks: 27 × 30 × $1.052 = $852 for Spark, plus 10 × 22 × $8.40 = $1,848 for SQL: $2,700 a month.
• Fabric F64 reserved: $5,003 a month, covering Spark, the warehouse and Power BI viewers with free licences.
• Fabric F16 reserved: $1,251 a month, with 32 Spark vCores (96 bursting), but every report viewer then needs a Pro licence.

Whether F16 is enough depends on how many CUs the work actually uses; Microsoft's Fabric Capacity Metrics app shows that after a trial. Storage is extra on both: OneLake hot storage is $0.026 per GB a month in East US.

Try itA fixed capacity, or pay for what runs

Fabric capacity

·F2·F4·F8·F16·F32·F64·F128·F256

ReservedPay as you go

Databricks Spark node-hours a day

27

SQL warehouse hours a working day

10

Fabric F64 reserved$5,003

Azure Databricks per use$2,700

F64 and up: report viewers need no Power BI Pro licenceReserved bills whether or not it runs

Which fits

• Choose Fabric if Power BI is your main way of consuming data, you have many report viewers (F64 removes their licences), your team prefers low-code pipelines and SQL, and your load is steady enough to reserve.
• Choose Databricks if engineering and machine learning are the core, you need fine control over compute, your load is bursty, or you run on more than one cloud. Our Snowflake vs Databricks comparison covers the other common choice.
• Use both if engineering lives in Databricks and reporting in Power BI: Fabric can mirror Azure Databricks tables into OneLake, and both read Delta Lake.

What to do

1. List your workloads: Spark hours, SQL hours, number of report viewers.
2. Price Databricks for that list, and the Fabric SKU that covers it.
3. Run a Fabric trial (an F64-sized capacity) on real work and read the Capacity Metrics app before reserving.
4. If you choose Databricks, decide between serverless and classic compute job by job.