Databricks develops a cloud data platform referred to as a 'lakehouse', combining features of data warehouses and data lakes.
Unlike traditional proprietary data warehouses, the Databricks architecture processes data while allowing organizations to retain the underlying files in their own cloud storage using open-source formats like Apache Iceberg and Delta Lake. Industry analysts note this separation of compute and storage mitigates vendor lock-in and avoids proprietary data egress fees.
The platform is built on the open-source Apache Spark framework, enabling analytical queries on semi-structured data without requiring a traditional database schema. While the platform's classic architecture provided robust distributed computing capabilities, it inherently required data teams to manually configure, size, and manage the underlying cluster infrastructure. To eliminate this operational complexity, Databricks introduced a serverless architecture across its analytics and AI workloads. This serverless tier fully abstracts the cloud environment, automatically provisioning, scaling, and terminating compute resources without requiring manual tuning or cluster management.
In October 2022, Lakehouse received FedRAMP authorization for use with the U.S. federal government and contractors.
Read the full text
The company has also created Delta Lake, MLflow and Koalas, open source projects that span data engineering, data science and machine learning.
In June 2020, Databricks launched Delta Engine, a fast query engine for Delta Lake, compatible with Apache Spark and MLflow.
In November 2020, Databricks introduced Databricks SQL (previously called SQL Analytics) for running business intelligence and analytics reporting on top of data lakes. Analysts can query data sets with standard SQL or use connectors to integrate with business intelligence tools like Holistics, Tableau, Qlik, Sigma, Looker, and ThoughtSpot.
Databricks offers a platform for other workloads, including machine learning, data storage and processing, streaming analytics, and business intelligence.
In early 2024, Databricks released the Mosaic set of tools for customizing, fine-tuning and building AI systems. It includes AI Vector Search for building RAG models; AI Model Serving, a service for deploying, governing, querying and monitoring models fine-tuned or pre-deployed by Databricks; and AI Pretraining, a platform for enterprises to create their own LLMs.
In March 2024, Databricks released its DBRX foundation model under the Databricks Open Model License. It has a mixture-of-experts architecture and is built on the MegaBlocks open-source project. DBRX cost $10 million to create. According to the company, DBRX performed competitively on industry benchmarks against other open-source models available at the time. It beat other models like Llama 2 at solving logic puzzles and answering general knowledge questions, among other tasks. While it has 136 billion parameters, it only uses 36 billion, on average, to generate outputs. According to Databricks, DBRX can be used as a foundation for companies to build customized AI models using their proprietary data.
In June 2024, Databricks open sourced Unity Catalog, a unified governance offering, under the Apache Spark 2.0 license.
In addition to building the Databricks platform, the company has co-organized massive open online courses about Spark and a conference for the Spark community called the Data + AI Summit, formerly known as Spark Summit.
At the 2025 Data + AI Summit, Databricks introduced Agent Bricks, a development platform for AI agents, Lakebase, a transactional database, and Databricks One, a no-code AI business intelligence platform. The company also disclosed its Databricks SQL product would grow to a $1 billion revenue run rate.
In March 2026, Databricks released Genie Code, an AI agent for data science and engineering tasks. That same month, the company announced Lakewatch, an AI-powered agentic security platform to automate threat detection and response.
In June 2026, Databricks released Omnigent, which it called an open source “meta-harness” for AI agents, like Anthropic's Claude Code and OpenAI's Codex, which are classified as harnesses, or the wrapper that turns AI models into agents. Omnigent is a common interface layer above command-line agents that allows users to build, control and collaborate on AI agents across models. During the company's "Data + AI Summit" conference, also in June 2026, Databricks announced a real-time analytics engine called Lakehouse//RT; an architecture to collapse online transaction processing and online analytical processing on a single copy of data, called LTAP (which is an initialism for "lake transactional-analytics process"); and Genie One, an agentic "co-worker." At the same conference, the company also released Genie ZeroOps, which automates monitoring, investigation and remediation of issues across data and AI workloads.
