If you haven’t already, check out Arun Ulag’s hero blog “Microsoft Build 2026: Building Agentic Apps with Microsoft Fabric and Microsoft Databases” for a complete look at all of our Microsoft Build announcements across our Fabric and database offerings.
__________________________________________
An end-to-end look at what's new in Fabric Analytics—where every layer is engineered for an agentic world, and every user benefits.
Every data leader asks the same question: where does AI deliver real, measurable business value? It does not come from isolated experiments. Instead, it shows up in systems where business users already work, inside the tools they open every day.
At Microsoft Build 2026, we are shipping updates across the analytics stack in Microsoft Fabric: an integrated system where every layer, from Spark to BI to Copilot, is designed for a world where AI doesn’t simply assist, but it executes across your data estate.
The agentic analytics stack
In the agentic analytics system, four layers work together:
- Data foundation (Data Engineering: Spark/ Lakehouse) that keeps data fresh, scalable, and queryable
- Serving layer (Warehouse) that delivers fast, concurrent, governed access
- Semantic layer (Power BI) that enables AI to reason on top of governed data and business logic
- Conversational analytics layer (Power BI, Fabric data agents) that brings insights into the tools people use every day
At Build, we’re advancing each layer and integrating them end-to-end so that agents and the people who rely on them get faster, more governed, and more trustworthy answers.
Data Engineering & Data Science: faster Spark, smarter exploration, elastic scale
Agents are only as effective as the data behind them. Slow pipelines lead to stale or incomplete answers. Fabric Data Engineering is shipping a set of updates that push Spark performance, lower the barrier to data exploration, and make clusters genuinely elastic.
The Native Execution Engine now handles complex data types and native UDF execution. Queries over nested structures—arrays, maps, structs—run entirely within the native engine, enabling optimizations like Z-ORDER and Liquid Clustering without fallback. Python and Scala/Java UDFs execute natively too, eliminating serialization overhead. Both structured and semi-structured workloads get faster with no code changes.
In addition, we are introducing Efficient Scaledown, which automatically offloads shuffle data to Azure Blob storage via Remote Shuffle Manager, and intelligent routing ensures data stays accessible as executors are released. Clusters scale down aggressively as soon as work completes, eliminating idle compute and cutting costs. This way, workloads stay cost-efficient and responsive, even with bursty, AI-driven demand.
Coming soon, Lakehouse Query Explorer will introduce a streamlined, in-context query experience directly within the Lakehouse. It enables data engineers and analysts to explore, visualize, and refine Spark SQL queries without switching tools. Users can write and run queries with intelligent suggestions and real-time error prevention—no notebook required. Results can be saved as Spark Views, visualized as charts, and interactively filtered and sorted, with the option to download outputs as needed. When deeper analysis is required, queries can be seamlessly promoted into a notebook.
Figure: Explore, query, and visualize right inside your Lakehouse.
Fabric Runtime 2.0 now includes Apache Spark 4.1, Delta Lake 4.1, and Python 3.13, bringing the latest advancements in performance, reliability, and developer productivity to modern data engineering workloads. In addition, Runtime 2.0 introduces key innovations such as enhancements to the Native Execution Engine like native Python/Scala/Java UDF support and expanded support for complex data types, deeper Delta Lake optimizations, and tighter integration with Fabric services—helping organizations run enterprise-grade Spark workloads more efficiently at scale.
MLflow 3 in Fabric streamlines both experiment management and development workflows. MLflow 3 enhances model and experiment tracking with tracing, logged models, and improved UX, making it easier to understand model behavior and build for GenAI applications.
Figure: MLflow 3 in Fabric: runs, models, and traces — all in a single experiment
Data Warehouse: GPU-accelerated analytics for the agentic era
When agents issue queries through NL2SQL, data agents, or automated pipelines, those queries hit the Data Warehouse. The Warehouse must respond at interactive speed, at enterprise scale, with full governance. Fabric Data Warehouse is evolving to meet this demand.
Query acceleration with GPUs
We are excited to announce GPU-accelerated query execution for Fabric Data Warehouse, purpose-built for AI-driven analytical workloads. The research paper outlining underlying technology, ‘CoddSpeed’, recently won the SIGMOD 2026 Best Industry Paper award.
What this means in practice:
- Faster analytics for most SELECT T-SQL queries, with no changes required to your data or SQL
- Simple to enable and monitor via a workspace-level setting
Based on internal testing conducted by Microsoft, GPU-accelerated Fabric Data Warehouse delivers up to 7x faster performance relative to three comparable data warehouse vendors, for reporting and application workloads. And the advantage grows as concurrency increases.
“Our experience with GPU-accelerated Fabric Data Warehouse has been excellent. The capability integrates seamlessly into our existing architecture and has delivered meaningful performance improvements across a range of queries, with complex workloads running 3.4x faster at single concurrency and more consistently.
At WTW, we operate a shared data platform supporting multiple applications and reporting workloads. GPU-accelerated Fabric Data Warehouse enables us to serve high volumes of queries more efficiently, directly improving the responsiveness of our downstream analytics and reporting. We see the potential for this capability to help unify how we deliver data across our platform.”
–Andrew Bradbrook, Director – Systems Architecture, WTW
Why does this matter for agents? Faster queries drive faster AI experiences. Data agents already translate natural language into SQL at machine speed. The warehouse needs to keep pace. Early adopters using Fabric data agents with Fabric Data Warehouse have seen end-to-end response times reduced by up to 50%. This is a significant difference turning slow interactions into truly conversational experiences. Query acceleration in Fabric Data Warehouse will be available in early access preview in the coming weeks, sign up to get access!
Cache Cooldown Configurability
Cache Cooldown Configurability gives you direct control over how long cached data is retained for warm queries. By setting the cache duration to match your workload patterns, you can eliminate cold-query latency and deliver consistently fast performance for your most critical queries.
This feature puts the performance-versus-cost tradeoff in your hands — extend the cache window to keep frequently accessed data warm for faster results or dial it back to optimize spend. Cache nodes automatically scale up and down based on demand, so you're never over-provisioned. The result is greater control and predictability over query performance, letting you tailor the warehouse experience to the SLAs and budgets that matter most to your business. Cache cooldown configurability will be available in product in the coming weeks.
Reimagined Web UI
We've dramatically upgraded the Fabric Data Warehouse web experience end to end:
- Object Explorer, Data Grid, and IntelliSense are now faster, smarter, and easier to use.
- A new table overview page lets you quickly understand table structure.
- Query management is streamlined with several new interactions like copy query and query import/export.
- Copilot chat is now embedded inline in the SQL query editor so you can stay in the flow while asking questions or iterating on SQL.
Figure: Demo of the new and improved Fabric Data Warehouse web experience.
Power BI: where agents learn to reason
AI without semantic context is a confident guesser. Power BI provides business logic that ensures trustworthy analytics. Put a language model in front of raw tables and it will produce answers that look right, until they are not. The difference is a well-curated semantic model: measures, relationships, hierarchies, and definitions that reflect how your organization works.
Agent Skills for Power BI (Preview)
Agent Skills for Power BI (Preview), announced at Build bring true end-to-end agentic development to Power BI, from raw data to semantic models to interactive reports. Describe to your AI agent what you need, even from a screenshot, and it can create models, generate reports, and iterate on visuals, all in one agentic workflow.
Figure: Demo of Agent Skills for Power BI creating a semantic model and a report from scratch.
Building apps in Fabric on top of semantic models
Rayfin is a new open-source SDK and CLI that enables developers and coding agents to build application backends, including a database, authentication, and storage. When deployed to Fabric, the app becomes a first-class Fabric artifact, inheriting the enterprise-grade features such as security and governance. For analytics teams, this means developers and their AI coding agents now have an accelerated path to building enterprise-grade data apps directly on their semantic models.
The same data foundation that allows analysts to build reports, now makes it much easier for developers to build full-blown web applications without needing to recreate business and governance logic. From financial planning to inventory management to pricing optimization, or really any app you can describe, your coding agent can build a custom-tailored app based on your specifications and semantic model in just a few prompts. And it's not just about polished UI. Coding agents can trivialize features that are used to take real engineering effort, like persona-specific views, custom calendar interfaces, bespoke business logic, and more.
Figure: Fabric data app on top of semantic model with Outlook calendar integration.
Fabric IQ: data meets every user
Everything we build across Spark, Warehouse, and Power BI serves one goal: get data to the people who need it, whether they are analysts, executives, or frontline workers. Fabric IQ is how we deliver on that.
Inside Microsoft 365 Copilot: Cowork and Copilot Chat
Fabric IQ now integrates with Microsoft 365 Copilot via Cowork and Copilot Chat (Frontier). This enables the agentic execution layer for Microsoft 365 to natively integrate with semantically rich enterprise data in Fabric and Power BI. Power BI reports and semantic models now show up directly in the productivity tools in M365 to enrich the flow of work: users can discover data, get grounded data insights, and act on those insights.
Combined with Work IQ, which personalizes Copilot with organizational context, this creates a unified intelligence layer. Data stops being something you must go find; it becomes part of the tools you already use. These experiences are currently available for customers in Frontier with a Microsoft 365 Copilot license. Join the program today.
Figure: Fabric IQ integrates with M365 Copilot Cowork.
GitHub Copilot CLI
For developers who live in the terminal, Fabric IQ is accessible through GitHub Copilot CLI via Agent Skills for Fabric. This allows developers to bring live, governed Power BI data directly into their workflow. Teams can ask questions about usage, metrics, or customer behavior in natural language and get answers grounded in their Fabric data, without leaving their development workflow. This tight integration helps reduce back-and-forth, accelerates decision-making, and ensures that product and engineering conversations are anchored in real data.
Data Agent improvements
We’re also releasing a wave of enhancements to Fabric data agents that make them more capable for analysis, easier to embed in applications, and simpler to build and tune.
- Service principal support (Preview): Data agents can now authenticate with service principals instead of only delegated user credentials. This makes it easier to use them in backend services, automated workflows, and custom applications.
- Observability in Microsoft Foundry (Preview): Fabric data agents now appear in Foundry Observability with telemetry like latency, status, and error details. This gives teams better visibility into agent behavior and makes debugging easier.
- Data agents in Microsoft 365 Copilot - General Availability: Data agent integration with Microsoft 365 Copilot is now generally available, bringing governed Fabric data into the Copilot experiences where users already work. The release also adds better support for longer-running queries.
- AI-assisted setup for SQL and Eventhouse sources (Preview): A new guided setup experience helps users generate and refine agent configurations more easily. It can suggest instructions, source guidance, and example queries to simplify onboarding.
- Preview runtime improvements: The preview runtime gives early access to foundational changes that will shape the future default behavior of data agents. It also lets teams validate updates and share feedback before broader rollout.
Figure: Preview runtime in the data agent allows teams to validate requests and share feedback.
- Improved NL2SQL and source routing (Preview): Updates improve how data agents translate natural language into SQL and how they choose the right data source. This leads to more accurate results, especially in multi-source scenarios.
- More advanced analysis with code interpreter (Preview): The Code Interpreter tool adds Python execution to Fabric data agents for richer analysis beyond standard queries. It enables scenarios like forecasting, statistical analysis, and data transformation.
- Model upgrades to GPT 5.X models, used by the data agent and its tools, resulting in~20% accuracy improvements based on internal benchmarks.
- Coming Soon - Visualizations in data agents (Preview): Visualization support will let users turn natural-language questions into charts directly in the data agent experience. This will make it easier to explore trends and compare results visually.
Figure: Visualization support will let users turn natural-language questions into charts directly in the data agent experience.
The full picture and how to get started
The agentic analytics stack is no longer a future concept. It’s here, integrated end to end in Microsoft Fabric. Every layer, from data engineering to AI-powered experiences, is designed to help your teams move faster, reason more effectively, and deliver real business impact with AI.
- Spark ensures data is fresh, scalable, and ready for AI-driven access. It runs complex workloads natively—nested types, UDFs, elastic scale down—while Lakehouse Query Explorer gives every team member a fast path from question to answer, keeping agent-facing data fresh and governed. Learn more.
- Fabric Data Warehouse delivers fast, concurrent, governed query performance. It makes data enterprise-ready with GPU-accelerated query execution, Time Travel, and procedural UDFs, delivering the response times that agentic workloads need. Learn more.
- Power BI provides the semantic layer so AI can reason over business logic: measures, relationships, and business context that make AI trustworthy, with Agent Skills for conversational report creation and Fabric Apps for governed analytics delivery. Learn more.
- Fabric IQ brings insights into every workflow. It puts that semantic knowledge in front of every user through Copilot Cowork and GitHub Copilot CLI, using natural language, governed by design. Learn more.
Every layer runs on OneLake, over open data formats. One security model. One governance framework. This is the shift from analytics as a tool to analytics as an intelligent stack that we are shipping at Build 2026. With Fabric, every layer works together so AI can reason, act, and deliver value across your data estate. Explore the agentic analytics stack in Fabric and see what you can build.
We’d love to hear what you’re building with Fabric Analytics. Share your feedback, ideas, and scenarios in the Fabric Community to help shape what comes next.