Azure HorizonDB: Microsoft goes big with PostgreSQL

Enterprises need data, and data needs to be stored, with a flexible, portable environment that scales from developers’ laptops to global clouds. That storage also needs to be able to run on any OS and any cloud without breaking the bank.

There aren’t many options. You might be able to use MySQL or any of its forks for most purposes, but it struggles to support large databases that run over multiple data centers. Plus there are licensing issues with proprietary tools like Microsoft’s SQL Server, with its hyperscale Azure SQL variant only available on Microsoft’s own cloud.

PostgreSQL is everywhere

It’s not surprising that developers have become increasingly dependent on the open source PostgreSQL, the nearly 30-year-old successor to the University of California Berkeley’s Ingres (hence its name). It’s a flexible tool that works well across a wide range of platforms and, thanks to an extensible architecture, can support most workloads.

Microsoft has been supporting PostgreSQL on Azure since 2017, with its 2019 acquisition of Citus Data bringing significant experience with scaling and performance. Since then, Microsoft has begun to build out a family of PostgreSQL platform-as-a-service implementations, with a hyperscale version as part of its Cosmos DB platform and a managed flexible server for most day-to-day operations. It even supports you running your own PostgreSQL instances on Azure VMs.

Microsoft has adopted PostgreSQL as a key part of its growing data platform, and the company has been a major contributor and sponsor of the open source project. There are 19 contributors who work for Microsoft, and code is already being delivered for the next major release in 2026.

Introducing Azure HorizonDB

This week at Ignite 2025, Microsoft announced the latest member of its PostgreSQL family: Azure HorizonDB. Designed to be a scale-out, high-performance database, it’s intended to be a place for a new generation of PostgreSQL workloads, for when you need an operational database that’s fast and can scale automatically without requiring complex sharding operations.

In advance of Ignite, I spoke to Shireesh Thota, CVP Databases at Microsoft, about the new service. He described the rationale for a new PostgreSQL variant:

I think increasingly what we notice is that people either go into the bucket of, “I want to lift and shift my PostgreSQL that’s working in the community version on-premises, or maybe another cloud.” They want to move it to Azure. They want 100% Postgres. They want all extensions working. They just want something that really has the flexibility of performance and speed. Then Azure Database for PostgreSQL, the existing version is perfect. Somebody who wants to build an AI-native, cloud-native kind of a workload that may need a lot of storage, wants really fast latencies, significantly higher IOPS. Then you go to HorizonDB.

Certainly, the published performance data for Azure HorizonDB is impressive: Microsoft is claiming a three-times increase in throughput over the open source release when running transactional workloads. You can scale up to 3072 cores, with 128TB of storage and sub-millisecond commits. HorizonDB builds on Azure’s multiregion architecture with data replicated in multiple availability zones and automated maintenance and backups with minimal impact on operations. Such performance is needed for AI applications and for large-scale Kubernetes. As Thota notes, “These cloud-native workloads can really succeed on HorizonDB.”

Key to the performance boost are changes to the architecture of the database, separating compute and storage and allowing them to scale independently. If you need more compute, Horizon DB will give it to you. If you need more read replicas, it’ll provision them.

Using Azure HorizonDB for AI

On top of compatibility with most standard PostgreSQL features, Microsoft has added its own features that support modern AI applications, with fast DiskANN-based vector search as part of retrieval-augmented generation (RAG) applications and model tuning. Using DiskANN’s new advanced filtering will give you a significant performance boost over PostgreSQL’s standard vector search, and its hybrid in-memory and disk search allow you to work with the largest vector indexes without significant performance impairments. Also, using the new filtered graph traversals makes queries up to three times faster.

Microsoft’s tools help you bring AI models inside your queries, using Microsoft Foundry to conduct AI operations as part of a SQL query. Managed models let you pick and choose from a list of default models in the Azure Portal, or you can bring your own Microsoft Foundry models. This allows you to do things like generating embeddings for query results as you write them to a vector index table without leaving the database. Other options let you use AI-based semantic searches or summarize results and provide insights into customer comments. Thota describes the process as a simple one: “You keep your SQL structure and invoke our semantic operators in the right places.”

It has built-in integration with Azure’s enterprise tools, adding support for encryption, Entra ID, and private endpoints so that cloud-hosted data can only be accessed by your own systems and applications. Added security comes from support in Azure Defender for Cloud to keep sensitive data protected. “Our core cohort of our customers are enterprises, and I want to make sure that we build something for both enterprises as well as developers,” Thota says. HorizonDB will initially be available in a small number of Azure regions, with limited access to the preview release.

Managing PostgreSQL in VS Code

Outside the database, Microsoft has released a general availability version of its Visual Studio Code PostgreSQL extension. This adds database development and management tools to your development environment, connecting to on-premises and in-cloud PostgreSQL instances, including HorizonDB. It’s important to note that this is a tool for any and all PostgreSQL implementations. You’re not limited to working only in Azure; you can use it with any database that implements the PostgreSQL APIs.

Tools in the extension allow you to visualize database schema, drilling into tables and displaying joins. You can display complex data structures and zoom in to specific tables as needed. Another set of visualizations delivers a server dashboard that drills down into various metrics to help you understand how your database is running and where you can improve performance.

With Microsoft positioning Visual Studio Code as its AI development tool, both for creating AI applications and using AI, the PostgreSQL tool provides an agent for its GitHub Copilot tools. Natural language queries help you refine queries and design databases, and they use the same metrics as the server dashboard to help improve operations.

Bringing Oracle to PostgreSQL

HorizonDB’s performance improvements make it a good target for migration from existing relational databases, which can reduce licensing costs—especially for databases that have a per-core licensing model. Tools in the Visual Studio Code PostgreSQL extension help migrate Oracle schemas to Azure-hosted PostgreSQL, using AI tools to handle transformations based on best practices. To avoid problems, it allows you to validate its output in a Scratch database before you deploy the resulting database structure.

The tool works with more than just databases; it also helps you update application code to work with the new schema. Not everything will be updated automatically. To reduce the risks of hallucinations, it flags elements and code that can’t be migrated so you can perform manual updates. Not all Oracle features will migrate, as proprietary SQL extensions may not map to PostgreSQL’s standards-based approach.

Mirroring in Fabric

Data is increasingly important for businesses, with the growing capabilities of analytical platforms like Microsoft’s Fabric. HorizonDB and other operational databases are part of this approach, as they will mirror their tables into Fabric without affecting your applications. This brings near-real-time business data into an analytical platform for use in dashboards and AI applications. There’s no need for complex ETL to go from a row-based store to a column-based one, as it’s all handled by the platform. Microsoft won’t detail a timeline for bringing HorizonDB to Fabric, but it is part of the road map.

PostgreSQL is an important part of Microsoft’s data platform. Its open source foundations make it easy to develop outside Azure and then configure as part of deploying an application. HorizonDB takes it further, with support for at-scale cloud-native applications and for embedded AI. At the same time, mirroring operational, transactional data from PostgreSQL into Fabric ensures that your analytic applications have access to up-to-date information, making it easier to make business decisions without having to wait for data.