Microsoft Fabric IQ is Microsoft’s business-context layer for Fabric and OneLake data: an ontology (still preview), Power BI semantic models, a Graph, and data and operations agents that ground Copilot and Foundry in business meaning. Fabric IQ sits in an ontology camp with Palantir Foundry and Timbr; Snowflake and dbt anchor a metric-governance camp; and Atlan and other cross-platform context layers form a third, complementary camp.
Atlan reads Fabric’s own semantic models and OneLake tables as context sources, without requiring a rebuild, then carries that meaning to whatever agent runtime a team uses. For how Fabric IQ relates to Microsoft’s other two IQs, see Fabric IQ vs. Work IQ vs. Foundry IQ; what follows goes deeper on Fabric IQ alone, including where the enterprise context layer picks up where OneLake stops.
| What it is | Core components | GA status | Licensing note | Best for |
|---|---|---|---|---|
| Microsoft’s business-context layer over Fabric and OneLake data | Ontology, Power BI semantic model, Graph, data agent, operations agent | Mixed: data agent and Copilot Chat/Cowork integration GA; IQ workload and ontology preview | No separate Fabric IQ fee; data agents need F2+ Fabric capacity or Power BI Premium P1+ | Teams whose business data and agents live mostly inside Microsoft Fabric |
What is Microsoft Fabric IQ?
Microsoft Fabric IQ is the Fabric workload that models business entities, relationships, and rules over data already living in OneLake, part of Microsoft Fabric itself, so agents and people reason in business terms, not raw tables. Per Microsoft’s own Fabric IQ overview (Microsoft Learn, 2026), the workload’s two core items are the ontology, still labeled preview, and the Power BI semantic model. An ontology here is what ontology in AI generally means: a shared, machine-understandable model of the entities, properties, and rules an organization uses to describe its business.
Microsoft’s June 2026 announcement of a Gartner Magic Quadrant Leader placement for Analytics and BI Platforms said Fabric IQ “extends the value of Power BI semantic models so that AI experiences can reason over trusted business context” (Bismart’s recap of the Microsoft announcement, 2026-06-29), and Forrester frames the broader category the same way (more on that below). Futurum Group’s analysts write that Fabric IQ “connects business data to how the business is described” (Futurum Group, 2025).
One scope note, stated plainly once: Fabric IQ’s business meaning is scoped to data Fabric can reach through OneLake, including external sources pulled in via shortcuts or mirroring. That’s a strength for a Microsoft-resident estate and a boundary once data lives somewhere OneLake doesn’t reach.
How does Fabric IQ work?
Fabric IQ has five workload items: ontology, semantic model, Graph, data agent, and operations agent.
Ontology (preview)
Fabric IQ’s ontology turns business concepts like Customer or Shipment into entity types with properties, relationships, and rules, binding them to live OneLake data for one shared vocabulary. Microsoft calls the item explicitly preview; its newest version adds entity hierarchies, inheritance, and keyless entity types the earlier build lacked (Microsoft Learn, 2026). See ontology-first AI architecture and how to design an active ontology for AI without boiling the ocean.
Power BI semantic model
A Power BI semantic model is the curated layer of measures, hierarchies, and dimensions most Fabric customers already run in production (Microsoft Learn, 2026) and Fabric IQ’s common starting point: generate an ontology from an existing model instead of rebuilding business logic. See what is a semantic layer and ontology vs. semantic layer.
Fabric Graph
Graph is Fabric’s queryable relationship layer: nodes and edges over OneLake data, queried in GQL or natural language. It carries no preview label, but grounding an agent’s reasoning on it through the Fabric data agent is still preview, the distinction context graph vs. ontology and what is a context graph draw.
Fabric data agent
A Fabric data agent is a generally available, read-only Q&A layer over OneLake data: lakehouses, warehouses, semantic models, KQL databases, and ontologies. It caps responses at 25 rows and 25 columns by design, needs F2-or-higher Fabric capacity or Power BI Premium P1-or-higher, and respects Purview DLP and access-restriction policies (Microsoft Learn, 2026). See what Fabric data agents actually do and Fabric database agent vs. data agent for the difference.
Operations agent
An operations agent is the action-taking counterpart: it monitors live data against team-defined rules and, on approval, acts. Each gets its own Microsoft Entra Agent ID for auditing, but runs in delegated mode with its creator’s permissions, not a separate standing grant. See what Fabric’s own MCP servers actually let an agent do, who’s accountable for what an agent creates, and governing Fabric AI agents on governance.
Five items doing five jobs is modularity, correctly marketed as a strength inside Fabric. What it doesn’t do alone is tell an agent running outside Fabric which definition to trust.
Is Fabric IQ generally available, or still in preview?
Five items doing five jobs only works as modularity if a team knows which of those five it can rely on today. Some of Fabric IQ has shipped to general availability, but the ontology, the piece doing the real business-meaning work, is still labeled preview.
| Fabric IQ component | Status (per Microsoft’s current doc label) | Source / date | What this means in practice |
|---|---|---|---|
| IQ workload | Preview | Fabric IQ overview, Microsoft Learn | Still a preview workload in Fabric |
| Ontology | Preview | Ontology overview, Microsoft Learn | The business-meaning modeling itself is the least mature part |
| Power BI semantic model | GA (pre-existing) | Power BI semantic models, Microsoft Learn | Already production infrastructure at most Fabric customers |
| Graph | GA as a structure; graph-powered AI reasoning is preview | Graph overview, Microsoft Learn | Build and query a graph today; grounding agent reasoning on it is newer |
| Fabric data agent | GA | Fabric data agent concept page, Microsoft Learn | Read-only, capped at 25 rows/25 columns by design |
| Copilot Chat / Cowork integration | GA since September 28, 2026 | Arun Ulag, Azure blog, 2026-09-28 | Semantic model content surfaces in Microsoft 365 Copilot at no extra token cost |
| IQ Sharing | Preview since September 28, 2026 | Arun Ulag, Azure blog, 2026-09-28 | Brand new; expect rough edges |
| Foundry agent connection | Preview, “without a service-level agreement” | Fabric IQ + Foundry how-to, Microsoft Learn | Not yet the path for a production Foundry agent’s grounding |
Arun Ulag, EVP of Azure Data at Microsoft, announced “Fabric IQ in Copilot Chat and Cowork is generally available today,” with IQ sharing now in preview too (Azure blog, 2026-09-28). Fast-shipping progress, but Microsoft’s own overview pages still label the IQ workload and ontology preview: GA in places, preview where it matters most. See Microsoft Fabric Copilot for the broader Copilot-in-Fabric picture this integration sits inside.
Connecting a Foundry agent to Fabric IQ, listed above without an SLA, isn’t yet production-ready. For where Fabric IQ and Foundry meet at FabCon, see the FabCon Europe 2026 guide and sessions by role.
Treat “is Fabric IQ ready” as a per-item question, with the ontology the item to watch before a production agent leans on it.
What are Fabric IQ’s ontology limitations right now?
Practitioner reports through mid-2026 described the ontology item as functional but rough: community signal, not an audited test, so treated here as a starting point rather than confirmed fact.
Reddit discussions, generalized without usernames, described the ontology’s earlier preview build as missing cardinality, inheritance, hierarchies, and editable relationship properties, and unreliable at inferring joins from raw OneLake tables rather than an imported semantic model. The same discussions reported bugs: decimal columns returning null, external tables failing to bind, and Delta column mapping not working.
The newer default ontology experience explicitly adds “keyless entity types, relationships without dedicated join tables, namespaces, entity hierarchies, and reuse through inheritance” (Microsoft Learn, 2026), matching several gaps practitioners flagged earlier this year. The documentation says nothing about join inference from raw tables or cardinality by name, so whether those two gaps are fixed remains unconfirmed. Real progress on structure; an open question on data binding.
Setup itself belongs to Microsoft’s own how-to docs; confirm licensing at the primary source too: Fabric IQ adds no separate line-item license on top of Fabric capacity, but a data agent needs F2-or-higher capacity or Power BI Premium P1-or-higher, and ordinary Power BI or Copilot licensing still applies to the data touched (Microsoft Learn, 2026). See extending your ontology beyond Snowflake, Genie ontology and the Atlan context layer, and Microsoft Fabric governance.
An ontology that fixed structural gaps in one cycle and left data-binding gaps open is evolving fast enough that a status check belongs in every production decision.
When does Fabric IQ make sense, and when does it hit a boundary?
Fabric IQ’s business-meaning layer stops at OneLake, either irrelevant or decisive depending on where the rest of the estate lives. The signals below are a starting checklist, not an exhaustive test.
| Signal | Fabric IQ alone may be enough | You likely need a cross-platform layer too |
|---|---|---|
| Where the data lives | Entirely in Fabric and OneLake | Also in Snowflake, Databricks, Salesforce, SAP, or similar |
| Reporting surface | Power BI-centric | BI spread across multiple tools |
| Agent strategy | Copilot-first | Multiple runtimes: Copilot Studio, Claude, Cortex, Genie, custom agents |
| Definition ownership | One team, one platform | Several teams need one certified definition across platforms |
Microsoft’s own cold-chain logistics example, from Navyasree Potluri, Senior Solution Engineer at Microsoft, traces Order to Shipment to Temperature Sensor entirely inside Fabric IQ’s ontology because the whole chain lives in OneLake (ESPC, 2026-07-13). The same questions in evaluating a governance platform for Fabric apply just as well to whether a layer belongs alongside it.
Analyst coverage treats the second column as the common enterprise condition, not an edge case. How to implement an enterprise context layer for AI and AI readiness share the same premise: a business entity defined once must mean the same thing everywhere, and multi-agent memory silos is what happens when it doesn’t.
For a Microsoft-only estate, the OneLake boundary is a sensible scope, not a gap. For the more common multi-platform estate, it’s exactly what a cross-platform context layer exists to sit alongside.
Fabric IQ vs. a cross-platform context layer: which camp are you in?
A market-research scan of competitor positioning describes three distinct camps, and Fabric IQ, Atlan, and Snowflake’s semantic views each sit in a different one.
| Camp | Example platforms | What it optimizes for | Scope |
|---|---|---|---|
| Ontology camp | Fabric IQ, Palantir Foundry, Timbr | Full business semantic structure: entities, relationships, hierarchies, rules | Typically single-platform |
| Context-layer camp | Atlan, Promethium, Collibra | Wraps lineage, quality, and usage signals around existing definitions | Cross-platform |
| Metric-governance camp | Snowflake, dbt, Cube, AtScale | Define a metric once, query it everywhere | Usually cross-platform, metrics-scoped |
These are different jobs: Fabric IQ’s ontology models complete business semantics inside one platform, a context layer like Atlan carries lineage, quality, and usage signals across many, and a metric-governance layer solves one consistent number. Ontology vs. knowledge graph, context graph vs. knowledge graph, and why AI agents need an enterprise context layer go deeper on the gap.
Single-platform scope is a legitimate buyer question, not only a competitor’s talking point. At least one ontology-camp vendor already runs a “locked to single platforms” critique against the whole category, Fabric IQ and Atlan’s camp included, reason to state the boundary plainly rather than call it Atlan-invented; see Timbr vs. Atlan: ontology vs. context layer compared. Forrester’s Boris Evelson, VP and Principal Analyst, and Indranil Bandyopadhyay put the broader shift this way: “a context layer is the next evolution of semantic layers and knowledge graphs” (Forrester, 2026).
Being the strongest platform in the ontology camp doesn’t settle whether an enterprise also needs a context layer for AI agents; the market’s own taxonomy says most will need both.
How Atlan approaches Microsoft Fabric IQ
Atlan extends the same kind of trusted business context across Fabric and the rest of the enterprise estate, complementing Fabric IQ’s Microsoft-native job rather than competing for it. The moment an enterprise’s data also lives in Snowflake, Databricks, Salesforce, or SAP, Fabric IQ’s business meaning stops at the OneLake boundary, no flaw in what Microsoft built.
Atlan’s generally available Fabric connector catalogs Fabric workspaces, lakehouses, warehouses, semantic models, and pipelines, with table- and column-level lineage, treating an existing Power BI semantic model as an input. The Enterprise Data Graph binds those definitions to lineage, ownership, quality, and certification, then serves that context, over MCP, SQL, and APIs, to whichever agent runtime a team runs: Copilot Studio, Claude, Cortex, Genie, or custom. The Context Lakehouse is Iceberg-native, reading OneLake as open tables without a copy step.
In Atlan’s AI Labs benchmark, adding this kind of context improved AI’s text-to-SQL accuracy by 38%, a general proof point, not Fabric-IQ-specific. The split holds regardless of vendor: Fabric IQ for grounding agents inside Fabric, a cross-platform context layer for everything it was never scoped to reach. How to build an AI agent harness and “Gartner named three components, production needs seven” sequence that split in practice, running alongside Fabric IQ, not replacing it, the same split memory layer vs. context layer draws elsewhere.
Why Fabric IQ and a cross-platform context layer are different jobs
Fabric IQ is real, shipping fast, and useful for grounding Microsoft-native agents in business meaning. Microsoft announced general availability for Fabric IQ in Copilot Chat and Cowork on September 28, 2026, backed by capacity and licensing requirements. At the same time, Microsoft’s own documentation still labels the ontology, the piece doing the actual business-meaning work, preview, though a newer experience closed some structural gaps and left others, like join inference from raw tables, unconfirmed. That’s the honest status of a fast-moving feature, not a damning one. For a Microsoft-only estate, the OneLake boundary isn’t a gap at all. For the more common multi-platform estate, it’s exactly the problem a cross-platform context layer exists to sit alongside, not replace.
FAQs about Microsoft Fabric IQ
1. What is Microsoft Fabric IQ?
Fabric IQ is a Microsoft Fabric workload that models business entities, relationships, and rules over OneLake data through an ontology and Power BI semantic models, grounding agents like Copilot and Foundry in that meaning. It also includes a Graph for relationship queries and data and operations agents for Q&A and action.
2. Is Fabric IQ generally available or still in preview?
Both, depending on the piece. The Fabric data agent and the Copilot Chat/Cowork integration are GA, the latter since September 28, 2026. The IQ workload and ontology item are still preview, and connecting a Foundry agent to Fabric IQ is preview too, shipped without a service-level agreement.
3. What is the difference between Fabric IQ and Power BI?
Power BI semantic models are curated measures, hierarchies, and dimensions that already exist in most Fabric estates. Fabric IQ’s ontology builds a broader business-entity layer on top, often generated directly from those same models.
4. What is an ontology in Microsoft Fabric?
An ontology in Fabric IQ is a shared, machine-understandable model of entity types such as Customer or Shipment, their properties, relationships, rules, and metrics, bound to live OneLake data so people and agents share one vocabulary.
5. Does Fabric IQ require a separate license?
No separate Fabric IQ fee exists on top of Fabric capacity. A Fabric data agent specifically needs F2-or-higher capacity, or Power BI Premium P1-or-higher, and ordinary Power BI or Microsoft 365 Copilot licensing still applies to the data an agent touches.
6. How does Fabric IQ work with Microsoft Copilot and Foundry agents?
Fabric IQ has been GA inside Copilot Chat and Cowork since September 28, 2026, surfacing semantic model content at no extra AI token cost. Connecting a Foundry agent to Fabric IQ exists but remains preview, documented without a service-level agreement.
7. Is Microsoft IQ the same thing as Fabric IQ?
No. Microsoft IQ is the umbrella name covering Fabric IQ alongside Work IQ, Foundry IQ, and Web IQ, each contributing a different kind of context (business data, how employees work, organizational policies, and the web) rather than being one product.
8. What is a Fabric data agent?
A Fabric data agent is a generally available, read-only Q&A layer over OneLake data: lakehouses, warehouses, semantic models, KQL databases, and ontologies. It caps responses at 25 rows and 25 columns by design and needs F2-or-higher capacity or Power BI Premium P1-or-higher.
9. Can Fabric IQ work with data outside OneLake or Microsoft Fabric?
Fabric IQ’s business meaning is scoped to data reachable through OneLake, including shortcuts or mirroring. For data and agents outside that boundary, a cross-platform context layer is the complementary answer, not a replacement.
Sources
- What is Fabric IQ?, Microsoft Learn (Microsoft, 2026)
- What is ontology (preview)?, Microsoft Learn (Microsoft, 2026)
- Fabric data agent creation, Microsoft Learn (Microsoft, 2026)
- What is graph in Microsoft Fabric?, Microsoft Learn (Microsoft, 2026)
- Power BI semantic models in Microsoft Fabric, Microsoft Learn (Microsoft, 2026)
- Connect agents to Microsoft Fabric with Fabric IQ (preview), Microsoft Learn (Microsoft, 2026)
- FabCon and SQLCon 2026 in Barcelona: Building the Data Foundation for Microsoft Copilot and Agents, Azure Blog (Arun Ulag, Microsoft, 2026-09-28)
- If Fabric Unifies Where Your Data Lives, Fabric IQ Unifies What It Means, ESPC (Navyasree Potluri, Microsoft, 2026-07-13)
- The Next Evolution Of AI Will Rely On Context Layers, Forrester (Boris Evelson & Indranil Bandyopadhyay, Forrester, 2026)
- Microsoft Ignite 2025: AI Agent 365, Anthropic on Azure, Security Advances, Futurum Group (Nick Patience, Mitch Ashley, Fernando Montenegro & Keith Kirkpatrick, Futurum Group, 2025)
- Making Sense of Microsoft’s AI Strategy: Work IQ, Fabric IQ, Foundry IQ, James Serra (James Serra, 2026)
- Microsoft Named a Leader in the 2026 Gartner Magic Quadrant for Analytics and BI Platforms, Bismart (Bismart, 2026-06-29)
