Databricks Lakebase Reaches General Availability on Azure, Completing Multi-Cloud Rollout of Its Serverless PostgreSQL Database
Databricks has made Lakebase generally available on Azure across 14 regions, completing a multi-cloud rollout of its serverless PostgreSQL database built on technology from its Neon and Mooncake Labs acquisitions.
Overview
Databricks has announced the general availability of Lakebase on Microsoft Azure, making its serverless PostgreSQL database available for production workloads across 14 Azure regions worldwide. The March 2, 2026 release follows Lakebase’s AWS general availability on February 22, completing a rapid multi-cloud rollout that began less than a year after Databricks acquired PostgreSQL startup Neon in June 2025.
Lakebase is a fully managed, cloud-native PostgreSQL database that brings online transaction processing capabilities directly into the Databricks Lakehouse platform. By separating compute from storage and placing data directly in the data lake, the service allows organizations to run transactional workloads alongside their existing analytical and machine learning pipelines without maintaining separate database infrastructure.
Architecture and Key Features
The Azure release introduces Lakebase Autoscaling as the default for all new instances, a shift that took effect on March 12, 2026, according to the Azure Databricks release notes. The autoscaling version represents a significant upgrade over the earlier provisioned offering, adding several capabilities designed for modern application development.
Compute resources automatically adjust based on workload demand, and a scale-to-zero feature suspends instances after inactivity, waking them in seconds when needed. This approach eliminates the need for teams to provision and right-size database infrastructure manually. The service supports PostgreSQL 17, the latest major release of the open source database, and includes pgvector for AI-driven vector search use cases.
Instant data branching, built on copy-on-write technology, allows developers to create isolated database copies in seconds for testing and experimentation without duplicating storage. Point-in-time restore enables recovery to any moment within a configurable window of up to 30 days. High availability, which launched alongside the Azure GA on March 2, pairs primary compute with secondary instances distributed across availability zones for automatic failover.
From Neon Acquisition to Production Service
Lakebase traces its origins to Databricks’ acquisition of Neon, a serverless PostgreSQL provider, in June 2025. Development accelerated further with the October 2025 acquisition of Mooncake Labs, which brought technology to bridge PostgreSQL with lakehouse data formats, according to InfoQ. Together, the two acquisitions provided the architectural foundation for separating compute from storage while maintaining full PostgreSQL compatibility.
The result is what Databricks describes as lightweight, ephemeral compute on top of durable, data lake storage. This design prevents resource-intensive queries from affecting live transactional operations, a long-standing challenge in systems that attempt to serve both analytical and operational workloads.
Databricks CTO and co-founder Matei Zaharia has highlighted that Lakebase enables database branching, snapshots, and point-in-time rollbacks while maintaining standard PostgreSQL interfaces, as reported by InfoQ. The emphasis on PostgreSQL compatibility is strategic: organizations can adopt Lakebase without rewriting applications or abandoning existing tooling.
Target Use Cases and Platform Integration
Databricks positions Lakebase primarily for three use cases: real-time feature serving for machine learning models, persistent memory for AI agents, and embedded analytics. These scenarios share a common pattern — they require low-latency transactional reads and writes that feed into or are informed by analytical workloads running on the broader lakehouse platform.
The service integrates with Databricks’ Unity Catalog for unified governance, providing consistent access controls, auditing, and compliance across both transactional and analytical data. Synced tables allow Lakebase to serve lakehouse data with low latency, while Lakehouse Sync, currently in beta on AWS, pushes Lakebase table changes back to Delta and Iceberg formats for downstream analytics.
On March 16, Lakebase Autoscaling databases became available as resources for Databricks Apps, according to the Azure Databricks March release notes, tightening the integration between application development and database management within the platform.
Regional Availability and Compliance
The Azure GA extends Lakebase beyond the initial three regions — East US 2, West Europe, and West US — to 11 additional regions including Australia East, Brazil South, Canada Central, Central India, and UK South, as documented in the March 2026 release notes. On March 4, Lakebase reached general availability for workspaces operating under compliance security profiles, supporting HIPAA, C5, and TISAX standards.
Google Cloud availability has not been announced but is expected later in 2026, according to InfoQ’s reporting.
Competitive Context
The launch places Databricks in direct competition with managed PostgreSQL providers including AWS Aurora, Google Cloud SQL, and Azure Database for PostgreSQL, as well as serverless database startups. The differentiator Databricks is betting on is the elimination of the traditional boundary between transactional and analytical systems. Rather than maintaining separate OLTP and OLAP databases connected by ETL pipelines, Lakebase aims to consolidate both on a single governed platform.
The move also signals a broader industry trend. Snowflake has been expanding its real-time data capabilities, while traditional database vendors are adding analytical features. Databricks is approaching the convergence from the opposite direction, bringing transactional capabilities to a platform originally built for analytics and machine learning.
Lakebase uses usage-based billing in Databricks Units for compute, with storage billed separately, following the consumption pricing model that has become standard across the data platform industry.