Incident sheet · /data
- System
- Data & Analytics
- Filed
- Read
- 4 min · 758 words
- Severity
- Routine
Microsoft previews serverless HorizonDB for agent applications
Microsoft’s public-preview Fabric HorizonDB gives developers and AI agents serverless PostgreSQL and translytical queries, but introduces cold starts, data lag and governance needs.

Event log
- Fabric HorizonDB entered public preview at Ignite after Microsoft unveiled HorizonDB for Azure in November 2025.
- Applications use one PostgreSQL interface to query live transactional data and OneLake analytical data.
- Supported analytical query portions run on CPU and GPU resources, while PostgreSQL retains transactional processing.
- Fabric HorizonDB pauses compute after five consecutive minutes of eligible inactivity.
- Cost visibility must cover database compute, storage, analytical queries and charges that continue while compute is paused.
At Microsoft’s ongoing Ignite conference, Fabric HorizonDB entered public preview, giving developers, business teams and AI agents a serverless PostgreSQL-compatible database for applications that automate business workflows or support agent-driven tasks. The launch followed Microsoft’s unveiling of HorizonDB for Azure at Ignite in November 2025.
That Azure product targets demanding, mission-critical workloads. Microsoft says its shared-storage architecture and elastic, scale-out compute support large databases and transaction-heavy applications. Fabric HorizonDB instead targets smaller, intermittent applications that do not need database infrastructure running continuously.
A managed backend provisions and scales databases with demand, pauses them when idle and resumes them when needed. Applications can query live transactional data and analytical data in OneLake through a single PostgreSQL interface, reducing the infrastructure and data-integration work involved in building applications.
For supported analytical portions of PostgreSQL queries, Fabric HorizonDB uses distributed compute from Fabric Data Warehouse across CPU- and GPU-powered resources. Its PostgreSQL engine retains transactional processing, creating a translytical path between operations and analytics.
What changes for developers and CIOs?
David Linthicum, an independent technology consultant, said: “Agent-driven applications often need to spin up databases on demand to support specific tasks and workflows. Fabric HorizonDB makes this easier by allowing developers and AI agents to provision databases quickly without having to manually set up separate backend services for workloads that may be short-lived.”
He added: “That reduces the friction between an idea, its implementation as an application, and the data needed to make it work.”
Ashish Chaturvedi, executive research leader at HFS Research, said: “Previously, querying operational and transactional data together required engineering effort, usually involving a pipeline or connector that teams had to build, secure and maintain. Microsoft has turned that into a built-in capability, allowing an application to retrieve a customer’s current order and purchase history in a single request.”
“As developers spend less time building and maintaining data integration pipelines and more time developing applications, it shortens development cycles, improves engineering productivity and makes it easier for enterprises to scale agent-driven applications across business functions,” Chaturvedi said.
Linthicum said: “The new database can reduce the number of platforms, pipelines, and teams required to connect applications with enterprise data. If Microsoft executes well, CIOs get less integration debt, better use of Fabric investments, and a clearer path from operational systems to analytics and AI.”
Chaturvedi said: “Mainly because they can draw on live operational data and historical business information together, rather than relying on either in isolation. This allows applications and AI agents to factor in current business activity as well as historical patterns when evaluating situations and recommending actions.”
Chaturvedi cited sales applications that show a customer’s open issues alongside lifetime value, and AI agents that combine current business information with historical trends to make recommendations.
For CIOs, the serverless model can also restrain spending as the number of AI-agent applications grows. Chaturvedi said: “Every new application needs a database to store and manage its data, and provisioning dedicated infrastructure for each one can quickly add to costs and operational complexity.”
Fabric HorizonDB automatically pauses compute after five consecutive minutes of eligible inactivity. Chaturvedi said a prototype or rarely used internal tool costs very little while idle, though storage and analytical-query charges remain a separate governance issue.
The latency and freshness trade-offs
“Serverless is attractive, but it often introduces cold-start latency, scaling lag, and less predictable performance than provisioned infrastructure,” Linthicum said.
Developers should test pause-and-resume operations, concurrent requests and transaction-heavy workloads, he added.
Chaturvedi said: “The brief wait when a database resumes after being idle is fine for internal tools and occasional workloads, but less suitable for customer-facing systems that need consistent response times.”
Data freshness is another constraint. The analytical copy in OneLake updates with a slight delay, making translytical queries better suited to applications that tolerate data a few moments old than systems requiring strict real-time consistency.
Portability remains a concern. PostgreSQL compatibility offers a familiar interface, but applications using Fabric-specific services, OneLake or translytical capabilities may be harder to move elsewhere, Linthicum cautioned.
The governance test
Stephanie Walter, practice leader of AI stack at HyperFrame Research, said: “CIOs will need visibility into costs across database compute, storage and analytical queries, including which charges continue when database compute is paused.”
For CIOs, the trade is clear: fewer integration assets and shorter development cycles, set against cold starts, variable performance, delayed analytical freshness and deeper dependence on Microsoft’s stack.
via infoworld.com/feed/ (Source)
More from Olivia Hart
Show full bio
News editor covering money, funding and market shifts around enterprise information technology and the leaders who run at CIO Monitor.
207 articles
The newsroom digest
One email a day with the stories that matter. No noise.
Cross-references
- Microsoft puts HorizonDB Postgres into general availability at Ignite
- Dell Rolls Out AI Data Platform to Unify Enterprise Data Silos
- Snowflake and Palantir Strike Enterprise AI and Analytics Partnership
- CloudOps, FinOps and AIOps: Building a Framework for IT Autonomy
- Software Consolidation Offers Benefits, Not Automatic Savings