Microsoft Ignite 2026: What to Watch Across Fabric, Azure AI and Enterprise Integration

Justyna
PMO Manager at Multishoring

Main Topics

  • microsoft ignite 2026
  • enterprise ai platforms
  • microsoft fabric
  • finops for ai

On November 12, five days before Microsoft Ignite 2026, Microsoft put an end date on Purview Data Governance. Support stops on November 12, 2027, and the catalog and governance work moves into Microsoft Fabric. It will get little stage time this week. If your data sits in SAP, Oracle, Snowflake and a couple of regional ERPs, it will likely cost you more planning time than anything demoed.

Ignite runs November 17-20 at Moscone Center in San Francisco. Satya Nadella and Judson Althoff give the opening keynote (KEY01) on Tuesday at 10:00 a.m. PT at Chase Center, and keynotes plus many sessions stream free.

We will sort every announcement with three tests. Does it keep data governed when that data lives outside Fabric? Can it run in production at a cost finance can forecast? Does it connect to the ERP, finance and warehouse systems you already depend on? Anything that fails one of them can wait for a later release.

Why Microsoft Ignite 2026 matters now

Many CIOs already have an agent demo that impressed the board. What they lack is a version that pulls from five systems, passes an audit and fits a budget line. Ignite 2026 is Microsoft’s chance to show that its stack closes that gap.

Microsoft’s pitch: one stack for agents

Microsoft now describes enterprise AI as four connected layers:

LayerProductJob
Build and runMicrosoft Foundry (formerly Azure AI Foundry)Models, hosted agents, evaluation
Business contextMicrosoft IQ: Work IQ, Fabric IQ, Foundry IQ, Web IQFeeds company data and meaning to agents
WorkspaceCopilotWhere employees use agents
ControlAgent 365 and Microsoft Entra Agent IDInventory, identity and monitoring

That picture assumes an estate tidier than most. In large companies the blocker is rarely which model to use. It is a “customer” that means one thing in the CRM and another in the ERP, and interfaces built ten years ago that nobody wants to touch.

Why proof matters more than features this year

Gartner predicts that more than 40% of agentic AI projects will be canceled by the end of 2027 because of cost, unclear value or weak risk controls. Microsoft has noticed: the agenda has a full FinOps for AI track, and Agent 365 is in the opening keynote. In our Ignite 2025 recap the theme was building agents. This year it is running and paying for them.

Theme to watch: the enterprise AI platform

AI pilots tend to do well on curated demo data and then stall once they meet the real estate. When Microsoft walks through its enterprise AI platform this week, the two things to look for are how agent quality gets measured and how an agent gets stopped.

Microsoft Foundry and agent accuracy

Foundry is where agents are built and hosted. The Foundry sessions, starting with BRK240 on Tuesday, center on lifting agent reliability through evaluations, production traces and red-teaming. A pilot that is right most of the time looks fine in a demo and produces exceptions by the hundred in production.

Two questions to bring: is accuracy measured on your own data or on a benchmark, and who signs off on the tools an agent can call? Foundry “toolboxes,” which restrict the APIs and MCP endpoints an agent can reach, are worth checking for that second answer.

Fabric IQ still needs agreed definitions

Fabric IQ gives agents business context over data in OneLake. It starts from Power BI semantic models and adds ontologies, which tie entities such as customers or shipments, their rules and their metrics to live data. Microsoft’s documentation marks both the IQ workload and Ontology as preview.

Test this harder than anything else here. If three business units calculate margin three different ways, an ontology built from their reports will not pick a winner. It hands the disagreement to an agent that acts on it at once. Agreeing on shared definitions is slow work, and it decides whether Fabric IQ gives you a single version of truth or faster confusion. We made the same case about the semantic layer as AI infrastructure.

Agent 365, Entra Agent ID and the AI bill

According to Microsoft’s Agent 365 guide, the opening keynote presents Agent 365 as the monitoring layer for agents, models and assistants, Microsoft-built or not. BRK214 (Wednesday) and BRK345 (Thursday) go deeper, and THR877 covers Entra Agent ID, which gives each agent its own least-privilege identity.

CFOs should look at the FinOps sessions first. BRK308 starts from the premise that enterprise AI has three cost models: people, consumption and capacity. In practice that means per-user licenses, metered credits and tokens, and reserved Fabric or Azure capacity, each on a different invoice. If nobody maps those three to business units now, the AI line in next year’s budget will be a guess.

Theme to watch: data governance tools

For anyone running Purview Data Governance, a migration just landed on the 2027 plan, approved or not. It is also the clearest signal of where Microsoft’s data governance tools are heading.

Who does what after the change

Microsoft has divided governance and security between two products. The Fabric OneLake catalog takes policies, lineage, stewardship and trust signals, and it comes with Fabric capacity instead of separate meters. Purview keeps classification, sensitivity labels, data loss prevention and compliance.

If most of your analytics already runs in Fabric, this simplifies things. If it does not, the gaps below matter more.

Gaps Microsoft lists in its own transition guide

The transition guide is unusually direct about what the OneLake catalog cannot do yet:

NeedOneLake catalog todayBusiness impact
Catalog sources outside FabricOnly data brought in through shortcuts, mirroring or connectorsOracle, SAP or on-prem systems may need a separate catalog
Business glossary and critical data elementsPartial fit with tags, descriptions and endorsementRegulated glossary workflows have no direct replacement
Data quality rulesGovernance Health is not a rules or scoring engineQuality checks need their own tool
Central policiesFabric Policies (preview): allow rules only, three policy types, 100 rules per tenant and 50 per capacityNot available in West Europe, North Europe or West US capacities

Microsoft points customers with wider needs to Atlan, Alation, Collibra and Informatica, and tells regulated companies to talk to their account team early.

Zero-copy data, but not zero-copy security

Mirroring and shortcuts let Fabric read outside data without classic ETL. The limits are in the documentation. When you mirror an Azure SQL database, row-level security, dynamic data masking, object-level permissions and Purview sensitivity labels stay behind in the source. The cap is 1,000 tables, and “mirror all data” on a bigger ERP schema copies the first 1,000 alphabetically and skips the rest.

Mirroring still saves pipeline work. Budget for rebuilding access rules in Fabric, and for proving to an auditor that they match, before an agent queries that data.

The enterprise questions announcements must answer

A keynote shows what works. Before an announcement goes into a roadmap, it should hold up against these questions. A vague answer means pilot, not plan.

TestQuestion for Microsoft
Governed dataWill the OneLake catalog add scanning, glossary or data quality features before November 2027?
Governed dataWhen do Fabric Policies reach Europe and West US, and will deny rules come?
Governed dataWill mirroring carry row-level security, masking and labels into OneLake?
Production AICan seats, credits, tokens and capacity be seen in one view per business unit, with hard limits?
Production AICan Agent 365 and Entra Agent ID require approval before an agent starts writing data, and stop it?
Production AIDoes Fabric IQ Ontology leave preview, and how does it handle conflicting definitions?
IntegrationWhen will Fabric Data Factory support Azure-SSIS runtimes, managed virtual networks and private endpoints, which Microsoft’s ADF comparison still lists as “To be determined”?
IntegrationWill standalone AI Gateway, now in preview only in East US 2 and Sweden Central, expand, and can MCP calls pass the end user’s identity to an ERP?

Sessions we will follow

SessionTestWhat we want to learn
KEY01 (Tue)All threeWhat reaches GA and what stays in preview
BRK214 / BRK345Production AIHow agent inventory, identity and lifecycle work in practice
BRK348 / BRK362Governed dataTools for the Purview-to-OneLake move
BRK262 / BRK308 / BRK537Production AICost attribution across all three billing models

What Multishoring will evaluate after the event

We are not at Moscone this year. We will work from the keynotes, recordings and updated Microsoft Learn pages, and sort each announcement with five checks.

First, release status by region. A feature in preview, or missing from the Azure regions you use, stays out of critical systems until GA and a rollback path are documented.

Second, the Purview move. If the gaps stay open, mixed estates do better with a split: native controls where each dataset lives, curated data products in Fabric, Purview for labels and DLP.

Third, pipelines. For SSIS-heavy estates and private-network ETL, Azure Data Factory with Logic Apps stays the safer backbone until Fabric Data Factory closes those gaps.

Fourth, agents that change records. Writing to ERP or finance systems needs per-user identity, retries that cannot double-post, and recovery when a legacy system times out. Where a fixed rule works, use a workflow.

Fifth, cost and upkeep two years out. Each new layer adds a meter, and your own team should be able to run and budget the setup after handover.

Our take on the announcement with the biggest enterprise impact follows on November 19.

Key takeaways

  1. Microsoft Ignite 2026 is mostly about running agents in production. Judge each announcement by the cost or risk it takes away.
  2. Purview Data Governance retirement is the pre-Ignite change with the widest reach for mixed estates. Map what you use before November 2027.
  3. Fabric IQ works only as well as your shared definitions. Settle them before agents start using them.
  4. Mirrored data arrives without its source security rules. Plan to rebuild and audit them.
  5. AI spend comes in on three meters. Assign seats, consumption and capacity to owners before usage grows.

Next step: list the data platforms and integration tools your AI plans rely on and mark which ones run inside Fabric. That one page tells you how much of Ignite applies to you today and how much depends on roadmap promises.

Not sure which Ignite changes affect your estate?

In a 30-minute call we review your Azure, Fabric and integration setup and tell you what needs action in 2027 and what can wait.

BOOK A 30-MINUTE MICROSOFT BRIEFING

One call, no commitment.

Justyna - PMO Manager
Justyna PMO Manager

One call, no commitment.

BOOK A 30-MINUTE MICROSOFT BRIEFING
Justyna - PMO Manager
Justyna PMO Manager

FAQ

When and where is Microsoft Ignite 2026?

Microsoft Ignite 2026 runs November 17-20, 2026, at Moscone Center in San Francisco. The opening keynote is at Chase Center on November 17, and an optional pre-day takes place on November 16.

Can I watch Microsoft Ignite 2026 online?

Yes. The keynotes and many sessions stream free online, and recordings usually appear in the session catalog after the event.

What is happening to Microsoft Purview Data Governance?

Microsoft announced on November 12, 2026, that it will retire Purview Data Governance, with support ending November 12, 2027. Governance moves to the Microsoft Fabric OneLake catalog. Purview keeps data security and compliance, including sensitivity labels and data loss prevention.

Is Azure AI Foundry now Microsoft Foundry?

Yes. Microsoft Foundry is the current name for Azure AI Foundry, Microsoft’s platform for building, hosting and evaluating AI agents and models on Azure.

Talk to expert →
contact

Thank you for your interest in Multishoring.

We’d like to ask you a few questions to better understand your IT needs.

Justyna PMO Manager

    * - fields are mandatory

    Signed, sealed, delivered!

    Await our messenger pigeon with possible dates for the meet-up.

    Justyna PMO Manager

    Let me be your single point of contact and lead you through the cooperation process.