Microsoft spins up Azure HorizonDB

Microsoft has announced a distributed PostgreSQL database service designed to rival other hyperscaler systems and third-party RDBMSes such as CockroachDB and YugabyteDB.

Claiming 100 percent compatibility with open source PostgreSQL, Microsoft said the database service available on Azure would implement a new storage layer that significantly improves performance, scalability, and availability of the database compared to its other PostgreSQL services. Microsoft already offers Azure Database for PostgreSQL and Cosmos DB for PostgreSQL, which handles distributed tables, but the vendor said the new service would be fully distributed.

Shireesh Thota, corporate vice president for databases at Microsoft, said: "Developers would choose HorizonDB for its massive read scale without the need to replicate the database between instances, and unique AI features including advanced DiskANN vector indexes that push filters down to improve performance, and AI model management that provides one-click integration into AI Foundry."

He did not answer questions about pricing for the various Azure PostgreSQL services.

Microsoft said Azure HorizonDB had been re-engineered for cloud platforms to offer scale and performance "far beyond" that of open source Postgres. It has autoscaling storage up to 128 TB, scale-out compute up to 3,072 vCores, with <1 millisecond multi-zone commit latency, and enterprise security and compliance, the vendor said.

The database service launch comes while PostgreSQL is booming. Stack Overflow found it is used by 58 percent of professional developers, making it by far the most popular database. However, it joins a crowded market for distributed PostgreSQL services, which have varying degrees of compatibility with the open source RDBMS. Third-party systems are available in the form of CockroachDB, YugabyteDB, and pgEdge, which promises a multi-master distributed database built on open source PostgreSQL. PlanetScale offers a service built on open source PostgreSQL and uses a proprietary operator the company developed for MySQL/Vitess. Meanwhile, Google (AlloyDB) and AWS (Aurora DSQL) both offer distributed PostgreSQL services.

Aurora DSQL, CockroachDB, and YugabyteDB offer serverless SKUs, meaning developers avoid provisioning tasks. HorizonDB does not yet.

"Initially Azure HorizonDB is not serverless," Thota said. "While storage is auto-scaling, customers configure the compute that they require and add or remove replicas themselves depending on the throughput needed for their app."

Devin Pratt, research director at IDC, said: "The big clouds are moving toward the same pattern of PostgreSQL compatible services with cloud native storage and compute plus AI features.

• Google previews Code Wiki: Can you trust AI to document your repository?
• Microsoft ships .NET 10 LTS and Visual Studio 2026, Copilot everywhere
• Azure stumbles in Western Europe, Microsoft blames 'thermal event'
• Snowflake goes all out to woo PostgreSQL developers with lakehouse extensions

"HorizonDB fits that arc and folds vector search with predicate pushdown and model management into the service. For developers, that means fewer components to wire up and a more direct path to AI features next to transactional data. Teams should still validate latency, cost, and extension support on their own workloads."

While Google and AWS have similar services, Pratt said it was a case of alignment rather than catch-up as the Azure service has "fewer moving parts and a straighter path to AI features."

Holger Mueller, principal analyst at Constellation Research, said there was a path to an interoperable database standard because PostgreSQL services are becoming so prevalent among the leading cloud providers. "PostgreSQL is really the hope of potential code compatibility between the different clouds... and it's the potential bet for something other than Oracle, which has clearly won the battle for transactional databases," he said.

The move is perhaps another sign of Microsoft increasing commitment to the open source database system, despite also promoting the general availability of SQL Server 2025, its proprietary RDBMS. For example, in January, it announced two PostgreSQL extensions: pg_documentdb_core is designed to optimize for Binary JavaScript Object Notation, or BSON, a binary-encoded serialization of JSON documents, while pg_documentdb_api offers the data plane implementing operations to create, read, update, and delete data, as well as query functionality and index management. Both are designed to create a document database platform from a PostgreSQL back end, with Microsoft suggesting FerretDB as a front end.

The resulting MongoDB-compatible "multi-cloud and hybrid NoSQL" services are available now in Azure, Microsoft said. ®