Microsoft touts scalability of its new PostgreSQL-compatible managed database

Microsoft is previewing a third PostgreSQL-compatible managed database for its cloud, Azure HorizonDB, that it hopes will prove attractive for modern workloads and for legacy application modernization.

The company already offers two databases compatible with PostgreSQL, the database interface proving popular with developers of AI applications: Azure Cosmos DB for PostgreSQL and Azure Database for PostgreSQL.

Azure HorizonDB differs from both the existing offerings in that it’s a shared-storage, scale-out compute architecture explicitly engineered for AI-era application patterns, said Stephanie Walter, practice lead of the AI stack at HyperFrame Research.

“In contrast, Azure Database and Cosmos DB for PostgreSQL remains the general-purpose managed Postgres for traditional apps and the distributed, sharded option for horizontally scaled multi-tenant workloads respectively,” she said.

“HorizonDB is the AI-era Postgres tier, where transactional modernization and vector-driven applications finally land in the same engine. It’s meant to be the place where you both modernize legacy transactional systems and bolt on RAG, agents, and embeddings-heavy features without stitching together separate databases and ML endpoints,” said Walter.

One of the things that has made PostgreSQL so popular with AI developers is its support for vector indexes, said Microsoft, adding that in Azure HorizonDB it is extending that support with advanced filtering capabilities in the DiskANN vector index that “enable query predicate pushdowns directly into the vector similarity search”. This common optimization technique filters data as early as possible in the pipeline to avoid unnecessary data processing and transmission.

“That early pruning reduces engine work and can improve long-tail latency on vector-plus-metadata queries,” said IDC research director Devin Pratt.

Microsoft also said HorizonDB will make it easier for developers to integrate and manage generative, embedding, and reranking models from Microsoft Foundry with zero configuration.

That, said Pratt, will help reduce glue code, simplify audits, and shorten the path from prototype to production by keeping the data and the models close under one governance plane.

PostgreSQL elsewhere

Microsoft is not the only cloud provider offering PostgreSQL-compatible database services, but it is seeking to differentiate itself.

“Similar approaches exist in offerings such as AlloyDB AI, Snowflake Cortex, Aurora PostgreSQL with Bedrock/SageMaker, but HorizonDB emphasizes model management and vector search together inside the Postgres service,” Pratt said.

Another key differentiator for Microsoft, said Walter, is that Azure HorizonDB’s model layer is plugged into AI Foundry and the broader Azure AI ecosystem, so the same models and governance primitives can be reused across apps, agents, and data services.

Microsoft plans to add mirroring for Azure HorizonDB to its unified data and analytics platform, Fabric.

This will provide Fabric with a more solid, operational database foundation, according to Bradley Shimmin, practice leader of data, analytics, and infrastructure at The Futurum Group.

Fabric already supports mirroring of Azure Database for PostgreSQL.

It looks like you’re migrating to HorizonDB. Would you like help with that?

Although a big part of Microsoft’s story for Azure HorizonDB is that it’s ideal for serving up data to modern AI applications, it’s also pitching it as a way to modernize legacy workloads currently running on Oracle.

To help with that, it’s integrating GitHub Copilot into the PostgreSQL Extension for Visual Studio Code.

“Teams of engineers can work with GitHub Copilot to automate the end-to-end conversion of complex database code using rich code editing, version control, text,” Microsoft wrote in the blog post announcing Azure HorizonDB.

This feature, said Futurum’s Shimmin, “can help Microsoft displace established rivals, most notably Oracle, giving those customers another reason to make the move to Microsoft’s platform.”