Databricks

Overview

Databricks is the platform that coined the lakehouse, and it owns more of the stack than any of its rivals: the engine (Spark), the table format (Delta Lake), the catalogue (Unity Catalog) and the ML tooling (MLflow) all began here and are all open source, with the commercial platform built on top.

Architecturally it splits in two. A control plane that Databricks runs holds the workspace and its services; a compute plane does the work, either classic, where clusters run in your own cloud account, or serverless, where they run in Databricks' account in the same region.

That split is the thing to understand when comparing cost and control. Classic keeps data and compute in your account and bills the cloud separately; serverless is simpler and includes the underlying compute in the price.

Workloads span ETL, warehousing and BI, streaming, machine learning and, more recently, transactional workloads through Lakebase. Delta Lake is the default format, with Iceberg supported for the wider ecosystem, and query federation reaches out to Snowflake, BigQuery, SQL Server, Postgres and others without moving data.

Governance runs through Unity Catalog: a three-level namespace, automatic lineage from source to dashboard, row filters and column masks, and audit as system tables. Compliance features such as HIPAA and FedRAMP profiles are a paid add-on.

Current direction is agents: Agent Bricks for building them, model serving for deploying them and a vector search product underneath. The company reports a $7bn revenue run rate and more than 20,000 customers.

Key features and capabilities

The same headings are used for every lakehouse platforms entry, so two tools can be read side by side.

Architecture and storage

• A Databricks-run control plane, separate from the compute plane where data is processed
• Classic compute runs in your own cloud account; serverless runs in theirs, in the same region
• Classic workspaces need your own object storage bucket for system data and catalogue metadata
• Serverless available for SQL, jobs, pipelines and notebooks

Engines and workloads

• SQL warehouses in classic, pro and serverless flavours, plus a low-latency beta
• Spark for ETL and data science, with the Photon engine for query speed
• Structured Streaming, declarative pipelines and job orchestration built in
• Dashboards, notebooks and a natural-language interface for business users

Table formats and catalogues

• Delta Lake is the default format; Iceberg is supported for the wider ecosystem
• Hudi support is not published
• Catalogue federation to Hive, AWS Glue, Snowflake and others
• Query federation to MySQL, Postgres, Oracle, Redshift, Snowflake, SQL Server, Synapse and BigQuery

Security and governance

• Unity Catalog enforces access at query time across a three-level namespace
• Lineage is automatic, from source data through to models, services and dashboards
• Attribute-based policies, row filters and column masks
• Audit as system tables; HIPAA, PCI-DSS and FedRAMP profiles are a paid add-on

AI and machine learning

• Managed MLflow for experiments, models and evaluation, with a governed feature store
• Model serving for classical ML, GenAI and agents, with GPU support
• Agent Bricks for building agents grounded in your own data
• Vector search with real-time syncing from source tables

Ecosystem and integrations

• Partner Connect creates trial accounts and wires up partner tools from the interface
• A marketplace for datasets, models, notebooks and dashboards, built on open sharing
• REST APIs and ANSI SQL across the SQL tiers

Where it runs and what it costs

• AWS, Azure and Google Cloud, plus a first-party SAP offering
• Azure Databricks is billed and supported by Microsoft
• Classic compute is effectively bring-your-own-cloud; serverless is hosted
• Prices vary by region, plan and cloud

Pricing

Usage-basedSQL classic from $0.22 per DBU

Consumption-based, billed per second in Databricks Units. As listed in September 2026: SQL classic $0.22 a DBU, SQL pro $0.55, SQL serverless $0.70; jobs $0.15 classic and $0.35 serverless; all-purpose clusters $0.55 classic and $0.75 serverless. Classic prices exclude the cloud compute underneath, serverless includes it, and Photon raises the DBU rate. Figures change with plan, cloud and region. There is a 14-day trial and a free edition for personal use; committed-use discounts are quoted.

Vendor pricing page →

Demos and videos

• Databricks demo centre Product demo · databricks.com
• Databricks channel Channel or playlist · YouTube

About Databricks

Databricks was founded in 2013 by the creators of Apache Spark, and is based in San Francisco. It maintains Spark, Delta Lake, MLflow and Unity Catalog as open-source projects while selling the proprietary platform around them. It has raised repeatedly at rising valuations, most recently $5bn at a $190bn valuation in August 2026 led by Coatue, and reports a $7bn revenue run rate with over 20,000 customers, including 70% of the Fortune 500.

Founded 2013 · San Francisco, California · databricks.com