Fabric IQ vs Palantir Foundry: which ontology should a European mid-cap's agents run on?
The short answer
Choose Fabric IQ when you already run Fabric and Microsoft 365, your agents mostly need to read business entities, relationships, rules and metrics over data in OneLake, and you can accept an ontology item that Microsoft Learn still labels preview. Choose Palantir’s Ontology when agents must act as well as read, through modelled actions that write back to your systems, across many sources outside Microsoft, in production today, and you can carry a platform licence.
Both ontologies model the state of your data, and Palantir’s can also log the decisions taken through its own action types. Neither holds the procedures nobody wrote down or the handoffs between departments. That layer is the one worth owning, on whichever platform you already pay for.
Key takeaways
- Fabric IQ reads, Palantir reads and acts. Fabric’s ontology models entities, relationships, rules and metrics. Palantir’s adds actions that change objects in one transaction.
- “Generally available” covers less than it sounds. Fabric IQ was announced GA in June 2026. On 28 October 2026 Learn still labels the workload and its ontology preview, with no GA date.
- The pricing bases differ. Fabric bills capacity per second, plus Power BI licences for viewers below F64. Palantir sells licences and meters compute-seconds.
- Neither exit is a converter. Fabric imports RDF and OWL only into an empty item; Palantir exports JSON for re-import into Foundry. Actions and functions are rebuilt by hand.
- Residency depends on who reads the ontology. A Fabric data agent’s responses may leave Fabric’s region once a service outside Fabric consumes them. Palantir also lists an on-premises Foundry and customer-held keys.
Fabric IQ vs Palantir: what does each ontology actually model?
Fabric IQ models what your business is. Palantir models what it is and what people do to it.
Microsoft’s Fabric IQ overview says Fabric IQ “provides context on the state of your business”. Its ontology item defines entity types, properties, relationships, rules and metrics, and binds them to data in Fabric. Rules are business logic in natural language. Metrics are DAX measures that stay owned by their Power BI semantic model, so an agent reuses the revenue calculation finance already trusts. The ontology “isn’t itself a general-purpose data-query engine”: agents read its definitions, then query the bound source.
Palantir’s Ontology overview adds “kinetic elements (actions, functions, dynamic security)” to objects, properties and links. An action “is a single transaction that changes the properties of one or more objects, based on a user-defined logic”. Functions hold logic in TypeScript or Python, and Data Connection enables writeback to external systems through webhooks and data exports.
Fabric acts outside the ontology. Operations agents send a Teams message by default. Once a recipient approves, they can run Fabric items such as notebooks, pipelines and user-defined functions, or trigger a Power Automate flow, under the identity and permissions of whoever created the agent. Microsoft’s Build post says ontologies define actions; Learn does not, so plan on Learn. For a planner releasing a blocked order, the difference is a governed transaction in the model or a flow wired up next to it.
Palantir Foundry vs Microsoft Fabric: how do they compare as an ontology for agents?
Palantir’s Ontology models more, including actions, functions and dynamic security, and its documentation carries no preview label. Fabric’s exports more of what it models in open formats, and runs on capacity a Microsoft shop may already pay for. The table compares the two ontology layers as each vendor documents them on 28 October 2026.
| Fabric IQ ontology | Palantir Ontology | |
|---|---|---|
| What it models | Entity types, properties, relationships, natural-language rules, and metrics owned by Power BI DAX measures | Objects, properties and links, plus actions, functions and dynamic security |
| Can agents change data through it? | Not inside the ontology. Operations agents message in Teams by default and, after a recipient approves, run a notebook, pipeline or user-defined function, or trigger a Power Automate flow | Yes. Action types make transactional edits, and an action log, switched on per action type, records each one |
| Where the data must live | Fabric sources, including mirrored databases and OneLake shortcuts; the ontology holds only definitions and bindings | Foundry datasets, or virtual tables over OneLake, Databricks, Snowflake, BigQuery or S3 |
| How agents read it | Five routes, all in preview | AIP, Ontology MCP and the Ontology SDK |
| What leaves with you | Vocabulary as TTL or RDF/XML, rules and metrics as TMDL files; import works only into an empty item | A JSON file of the Ontology’s working state, for re-import into Foundry, with no stable schema; in beta, Ontology-as-code TypeScript definitions of object types, interfaces and actions in your own repository; functions as TypeScript or Python source |
| Pricing basis | Fabric capacity bought through Azure, billed per second or reserved yearly; OneLake storage and transactions; Power BI licences for viewers below F64 | Platform and capability licences; compute in Foundry compute-seconds, Ontology volume and storage in gigabyte-months |
| EU residency | Your capacity’s region; the ontology agent keeps requests from EU Data Boundary capacities inside the boundary; customer-managed keys for a list of items that excludes the ontology | Foundry SaaS regions include the European Union; Foundry On-Premises, Bring Your Own Key and Enterprise Key Management also listed |
| Status on 28 October 2026 | Fabric IQ announced GA on 2 June 2026; Learn labels the IQ workload, the ontology item, the ontology agent, import and export, agent integration and the MCP endpoint preview | Ontology, action types, SDK and Ontology MCP documented with no preview label; Ontology-as-code in beta |
The data row is not either-or. Palantir’s virtual tables query tables in other platforms “without first storing the data in a Foundry dataset”, and the OneLake connector is generally available, so a Fabric company can run Palantir’s Ontology over its OneLake data, paying for two platforms. If your ERP is SAP, Fabric’s ontology binds to SAP data only once it is in Fabric, through mirroring SAP, which runs through SAP Datasphere with SAP’s Premium Outbound Integration pricing on top, or another route into OneLake; where ERP context should live is covered in Nobody owns the handoffs.
The agents row needs more room than a cell. Learn documents five routes from an agent into Fabric’s ontology, and its page on agent integration labels them all preview: the ontology as a source for Fabric data agents, though the data agent itself is generally available, and for operations agents; as a Foundry IQ knowledge source; in Copilot Studio through the Fabric IQ MCP tool; and through the ontology MCP server, which has a known issue blocking Service Principal access. The ontology agent, also in preview, is a Copilot for building and operating the ontology itself. Palantir’s agents work inside Foundry through AIP, and from outside through Ontology MCP, with which external agents read objects and execute predefined actions, within application restrictions, or through the Ontology SDK for TypeScript, Python, Java or OpenAPI.
Is Fabric IQ generally available?
The workload is, by Microsoft’s announcement. The ontology at its centre is not. At Build on 2 June 2026, Microsoft’s Fabric and databases post said “Fabric IQ, now generally available, addresses this gap”, and two sentences later that ontologies were “expected to be generally available in the coming months”. Learn does not match the announcement: it still calls the workload itself “IQ (preview)”. Fabric IQ is delivered through two core items, the ontology (preview) and the semantic model, so the part without a preview label is the Power BI semantic model your BI team may already run.
Nearly five months on, Learn gives no GA date for ontologies. The ontology overview, updated on 16 October 2026, still says “This feature is in preview”, as do the pages for import and export, the ontology agent and the ontology MCP server. The last lists a known issue: “you can’t currently use Service Principal to access the ontology MCP”, so there is no application identity for unattended agents until it is fixed. The Fabric data agent, which can take an ontology as a source, is generally available, and operations agents went GA in the June post. Connecting either of them to the ontology still carries the label: Learn’s page on agent integration for ontology covers every route, including Foundry IQ and Copilot Studio, and says “This feature is in preview”.
Microsoft Ignite runs 17 to 20 November 2026, three weeks after this post. Every status here is dated 28 October 2026; check Learn for a GA label before you decide.
The ontology overview also says “The old experience of ontology retires on Jan 31, 2027.” Moving creates a separate copy: you reconnect agents, dashboards and integrations, and recreate rules and Activator alerts by hand. If you built on the old experience, budget that work for this winter.
In December 2025 Directions on Microsoft, an independent analyst firm, judged Palantir’s Foundry more mature and more widely adopted, particularly in operational IoT scenarios, while calling Microsoft’s IQs the likely choice for Microsoft-centric companies. The preview labels above suggest the maturity gap Directions described has not closed.
How is Fabric IQ priced compared with Palantir Foundry?
On different bases, so compare them on one workload, not on list prices.
Fabric IQ costs Fabric capacity, bought through Azure. Microsoft’s licensing page says “Azure bills you per second with no commitment, with the option of a yearly capacity reservation to save costs.” On smaller capacities viewers need a licence of their own: “On F SKUs smaller than F64, each user viewing Power BI content must have Pro, PPU, or an individual trial.” The ontology MCP server and the data agent both need a paid F2 or higher.
The ontology has its own capacity meters. Each read of its definition, by the interface, the MCP server or the ontology agent, is rated at 1,000 CU-seconds per Ontology Discovery. AI reasoning is metered dynamically, and Microsoft’s example estimate for drafting an ontology from a workspace is about 20 CU-hours. On 28 October 2026 the Discovery and AI Reasoning meters are listed as not billable and the Logic and Operations meter as billable. Ask whether that changes at GA, and size the capacity as if it does. For an evaluation, Microsoft advises: “Start with an F8. An F4 capacity might suit light testing.”
Palantir meters the platform in Foundry compute-seconds, “a unit of computational work in the platform”, including Ontology syncs and queries, and meters Ontology volume and Foundry storage in gigabyte-months, on top of platform and capability licences. What those licences cost, as far as public sources show, is set out in what Palantir AIP and Foundry cost a European mid-cap.
For a mid-cap already on Fabric, the ontology’s marginal cost is the capacity units it consumes on an F SKU you already pay for, plus any resize, standard OneLake storage and read/write charges, and SAP’s outbound integration pricing if your SAP data arrives through Datasphere. Palantir’s is a new platform licence, negotiated before the first use case runs at scale. Neither price includes the people who build and maintain the ontology. Directions on Microsoft put it plainly: “Creating an ontology, in Fabric or elsewhere, is a major effort and requires on-going maintenance”.
Fabric IQ vs Palantir in the EU: where does the data go when an agent reads it?
Both can keep the ontology’s data in an EU region. What leaves that region depends on which agent reads the ontology and where that agent runs.
Fabric data sits in your capacity’s region. Once a capacity is created, Learn’s Multi-Geo page says, “any workspaces created under it will have their content stored in that region”, though the same page notes that “Data in transit might go back and forth between multiple geographies.” The ontology agent processes requests from EU Data Boundary capacities inside the EU Data Boundary.
For a group with entities in several countries, two details matter more. A Fabric data agent cannot query a source whose capacity sits in another region, and Microsoft’s example uses two EU regions: “a lakehouse with capacity in North Europe fails if the Data Agent’s capacity is in France Central.” And once a data agent is consumed from Microsoft Foundry, Copilot Studio, Microsoft 365 Copilot or as an MCP server, the tenant settings page warns that its responses “may be sent outside of Fabric’s compliance boundary or geographic region”.
Palantir lists the European Union among its Foundry SaaS regions, on AWS, Azure, Oracle or GCP. The same page lists Foundry On-Premises, and an Enterprise Key Management option under which you can delete the master key that encrypts your data in the Foundry Filesystem, leaving that data unrecoverable. Fabric offers customer-managed keys for workspaces through Azure Key Vault, but Learn’s list of supported items does not include the ontology, and a workspace with the keys switched on accepts only supported items. Ask Palantir which region your enrollment and its model calls run in. Ask Microsoft when the ontology joins that list, and whether any part of it can run outside Microsoft’s cloud.
Residency is not jurisdiction. Both vendors are US companies, and where data sits does not decide which government can compel access to it: the Frontier Company post covers Microsoft’s side and the Palantir cost post covers Palantir’s. The EU Data Act cuts the switching fees for leaving either platform, with the exemptions the cost post sets out. Translating one vendor’s ontology into another’s stays your work.
Can you migrate a Palantir Ontology to Fabric IQ?
Partly, and by hand. Neither vendor documents a converter, and the two export formats do not meet.
Palantir’s Ontology Manager exports the Ontology’s working state as JSON for re-import into Foundry, with a warning: “You should not depend on the exported JSON schema as it may change over time.” It describes no RDF or OWL export. Palantir also documents Ontology-as-code, in beta and not on every enrollment: “TypeScript definitions of your object types, interfaces, actions, and other entities” kept in a SuperRepo, into which entities already built in Ontology Manager can be imported with the foundry import ontology command. Palantir says a SuperRepo “can be hosted, built, and deployed outside of the platform”. That is an easier source for a migration script than the JSON, but it is still not RDF or OWL, and actions still need redesigning. Fabric imports TTL, RDF/RDFS or OWL, and “You can only import into an empty ontology item.” So you re-express the Palantir model yourself:
| Palantir element | Where it goes in Fabric | Work involved |
|---|---|---|
| Object types and properties | OWL classes and datatype properties, imported as entity types and properties | Scripted from the JSON or Ontology-as-code definitions, then checked |
| Link types | OWL object properties, imported as relationship types | Scripted, then checked |
| Data behind the objects | Fabric sources bound to entity types: lakehouses, warehouses, mirrored databases, OneLake shortcuts | Rebind every entity |
| Action types | No ontology equivalent; operations agents running a notebook, pipeline, user-defined function or Power Automate flow after approval | Redesign, including the audit trail |
| Functions | Logic rewritten as notebooks, rules or DAX measures | Rewrite |
| Workshop applications | Real-Time Dashboards with the ontology as a source, or Power BI reports over the underlying semantic model | Rebuild |
Microsoft is candid about the import step. Learn says import and export “don’t provide complete or lossless support for the full RDF or OWL specification”, and warns: “A Preserved status or 100% import success doesn’t guarantee that all source semantics were retained.” Have someone who knows the business, not only the schema, review every imported class.
Leaving Fabric has its own limits. Its export writes Turtle or RDF/XML covering entity types, properties, relationship types, descriptions, labels, synonyms and single-parent inheritance. Rules and metrics leave only as TMDL in the item definition, one file per rule and one per metric. The empty-item rule bites without any vendor change, too. In the current preview, after an acquisition you cannot import the acquired company’s RDF or OWL model into your existing ontology. You import it into a second, empty item, then merge the two TMDL definitions by hand through Git or the definition API, or you run two.
Fabric IQ vs Databricks: is the Genie Ontology the same thing?
Close in purpose, different in shape, and it only matters if you already run Databricks.
Databricks calls its business layer Unity Catalog semantics: metric views for governed KPIs, domains that group assets by business purpose, Pages that define business concepts, and certification or deprecation flags. Databricks says “Together, these tools make up the human-modeled layer of the Genie Ontology”, which Genie One combines with context it infers from your assets and usage. Fabric IQ asks you to declare entity types and relationships; Databricks starts from governed metrics and lets inference fill in the rest. Neither documents an action type inside the ontology.
On Databricks, Fabric IQ needs data mirrored or shortcut into OneLake before it can bind. Fill in Unity Catalog semantics first, and model on a second platform only if your agents need explicit entities and relationships that badly.
Fabric IQ vs Snowflake: semantic views or an ontology?
Semantic views, and for agents that answer questions about numbers that is often enough. Snowflake’s semantic views are schema-level objects in which “You can define business metrics and model business entities and their relationships”. Cortex Agents read them and generate SQL against the physical tables. Nothing in them models an action.
Snowflake is ahead on the exit for metrics. Its export function, which replaces a deprecated OSI export, returns a semantic view as Apache Ossie (incubating) YAML, described as “an open standard for representing semantic models that enables interoperability across AI and BI tools”, still in preview. For metrics, that file travels further than Fabric’s TMDL files or Palantir’s re-import JSON; Fabric’s vocabulary, exported as RDF, travels as far. On Snowflake, the case for Fabric IQ on top is rules and explicit relationships across sources, not metrics.
Work IQ vs Foundry IQ vs Fabric IQ: which one competes with Palantir?
Only Fabric IQ. The Fabric IQ overview splits Microsoft IQ into four sources of context for agents, and only one of them models the business.
| Microsoft IQ part | Gives agents context on | Opposite Palantir’s Ontology? |
|---|---|---|
| Work IQ | How employees work | No |
| Foundry IQ | An organisation’s policies and authoritative documents | No |
| Web IQ | The web; added at Build in June 2026 | No |
| Fabric IQ | Business entities and data | Yes, through its ontology item |
The other three are inputs an agent may also read, not a home for the company model; the Frontier Company post covers what Work IQ and Foundry IQ let you export.
Palantir Foundry vs Microsoft Foundry: is that the same comparison?
No, the names only collide. Palantir Foundry is Palantir’s platform for data and operations, with the Ontology at its centre. Microsoft Foundry “unifies agents, models, and tools under a single management grouping”: it is where Microsoft’s agents are built and run. Comparing the two compares a company model with an agent runtime. The like-for-like pair is Palantir’s Ontology and the ontology in Fabric IQ.
When is Fabric IQ enough, and when does Palantir win?
Fabric IQ alone is enough when:
- your Power BI semantic models already hold the definitions finance trusts, since the ontology can be generated from them and keeps their DAX measures as metrics;
- the workflow stays inside Microsoft 365 and Fabric, with no handoff to a team working in another vendor’s system.
Palantir wins when:
- agents must write: an approval, a reallocation or a status change as a governed transaction on the model, logged, with writeback to the source system;
- the systems your agents must act on are mostly not Microsoft, and you want one model over them that can write back, in production today;
- you want operational applications built on the model, not only dashboards, and will pay a platform licence for a build led by Palantir’s own forward-deployed engineers, the delivery model Palantir pioneered.
Neither is enough on its own when the work crosses departments and systems, because both are incomplete in the same place.
Already chose Databricks or Fabric? The decision left is the layer on top
Most mid-caps reading this have already picked a data platform. The question left is narrower than the vendors make it: what will your agents need to know that the platform’s ontology does not hold?
Both ontologies hold the state of the business. Palantir argues that its Ontology represents the decisions in an enterprise, not simply the data, and its documentation draws the edge of that claim. The action log “is designed to capture decisions made by submitting actions”, and “Edits applied outside an action type, such as direct writes to the backing datasource or legacy Foundry Forms writeback, do not produce action log objects.” A decision taken in an email, in a Teams call between sales and credit control, or keyed straight into the ERP is not in the log.
That is where a mid-cap’s operating knowledge sits. Take a generic example, not a client case: the order in which a planner checks stock, supplier and customer before promising a date; the exception a credit controller grants a customer of twenty years; the disputed invoice waiting for sales to confirm a price. None of it is a table, so no warehouse ontology binds it. Palantir could model the credit exception as an action type, but only after someone has written the rule down and routed every exception through it. Most of the work is that writing-down. Nobody owns the handoffs describes the gap and the decision trace that closes it.
The option no platform page shows is a context engine your company owns, on the platform you already pay for. It reads whichever ontology you have and holds what none of them do: the unwritten procedures, the handoffs between departments and the trace of each decision. It sits inside your own stack, beside that ontology, as records you can read without any vendor’s tool, with the models underneath free to change. Agents run on top of it. Build vs buy vs embed compares the ways to get there, and how an engagement works sets out ours.
If you want a working use case in days and will pay a platform licence for it, a Palantir AIP Bootcamp gets you there faster than we would. If your workflow lives entirely in Microsoft 365 and Fabric and hands nothing to another system, switch on Fabric IQ and spend the money elsewhere. An owned context engine is also one more system to run: someone has to keep the procedures and decision records current, or agents read stale ones. Budget for that before the build, whoever does it. If the work crosses departments, the ontology you already pay for is the start of the company model, not the whole of it.
FAQ
Fabric IQ vs Palantir: which ontology should we choose? Choose Fabric IQ if you already run Fabric and Microsoft 365, your agents mostly read business entities, relationships, rules and metrics over OneLake data, and you can accept an ontology item still in preview. Choose Palantir’s Ontology if agents must also act through modelled actions that write back to your systems, across many sources outside Microsoft, in production now, and you can carry a platform licence. Neither holds unwritten procedures or cross-department handoffs.
Is Fabric IQ generally available? Partly. At Build on 2 June 2026 Microsoft announced Fabric IQ as generally available and said its ontologies were expected to follow in the coming months. On 28 October 2026 Microsoft Learn still calls the workload itself IQ (preview), and marks the ontology, its agent, its import and export, its MCP endpoint and its agent integrations as preview, with no GA date. Microsoft Ignite runs 17 to 20 November 2026, so check Learn for a GA label before deciding. The old ontology experience retires on 31 January 2027, and moving off it means reconnecting agents and recreating rules.
How is Fabric IQ priced compared with Palantir Foundry? On different bases. Fabric IQ runs on Fabric capacity bought through Azure, billed per second or reserved yearly, plus OneLake storage and transactions, and Power BI Pro or PPU licences for people viewing Power BI content on capacities below F64; the ontology has its own capacity meters, two of three listed as not billable in October 2026. Palantir sells platform and capability licences, meters compute in Foundry compute-seconds, including Ontology syncs and queries, and meters Ontology volume and storage in gigabyte-months.
Can you migrate a Palantir Ontology to Fabric IQ? Not with a documented converter. Palantir exports the Ontology’s working state as JSON for re-import into Foundry, with a schema it says may change. In beta, its Ontology-as-code also keeps TypeScript definitions of object types, actions and other entities in your own repository, which is easier to script from, but neither is RDF or OWL. Fabric imports TTL, RDF/RDFS or OWL only into an empty ontology item, and not losslessly. Object types and links can be re-expressed in OWL and imported; actions, functions and Workshop applications are rebuilt, with actions moving to Fabric operations agents.
Fabric IQ vs Databricks: is the Genie Ontology the same thing? Close in purpose, different in shape. Databricks calls metric views, domains, Pages and certification in Unity Catalog the human-modelled layer of its Genie Ontology, which Genie One combines with context inferred from your assets and usage. Fabric IQ asks you to declare entity types, relationships, rules and metrics. Neither documents action types inside the ontology the way Palantir does.
Fabric IQ vs Snowflake: what is the difference? Snowflake holds business meaning in semantic views, which define metrics, entities and their relationships for Cortex Agents to turn into SQL. Fabric IQ adds natural-language rules and bindings across several Fabric sources. Snowflake’s definitions export as Apache Ossie YAML, an open standard in incubation; for metrics that travels further than Fabric’s TMDL files, while Fabric’s vocabulary, exported as RDF, travels as far. Neither models actions.
Which Microsoft IQ competes with Palantir’s Ontology? Fabric IQ, specifically its ontology item. Of Microsoft IQ’s four parts, Work IQ gives agents context on how employees work, Foundry IQ on policies and authoritative documents, Web IQ on the web, and Fabric IQ on business entities and data. Only Fabric IQ models the business as entities, relationships, rules and metrics, as Palantir’s Ontology does with objects, links and actions.
Is Palantir Foundry the same as Microsoft Foundry? No. Palantir Foundry is Palantir’s data and operations platform, with the Ontology at its centre. Microsoft Foundry is Microsoft’s platform for agents, models and tools. Comparing them compares a company model with an agent runtime; the like-for-like pair is Palantir’s Ontology and the ontology in Fabric IQ.
Related reading
- What Palantir AIP and Foundry cost a European mid-cap
- Microsoft Frontier Company: whose intelligence does it protect?
- Salesforce fixed the CRM. Workday fixed HR. Nobody owns the handoffs.
- Forward-deployed engineer vs consultant vs system integrator: a decision rule for buyers
- Build vs buy vs embed: choosing an enterprise AI delivery model
- Own your intelligence: what it actually takes
Nucleo is an AI-native transformation company for European mid-caps. Our forward-deployed engineers build the context engine your company runs on, on the data platform you already pay for and model-agnostic, and put agents into production on top of it. See how an engagement works, or write to us.
Sources: Microsoft Learn on Fabric IQ, ontology in Fabric IQ, ontology import and export, the ontology item definition, ontology capacity usage, the ontology agent, the ontology MCP server, agent integration for ontology, operations agent actions, Fabric data agent concepts, Fabric data agent tenant settings, Fabric licences, customer-managed keys for Fabric workspaces, Multi-Geo, mirroring SAP and Microsoft Foundry; Microsoft Build 2026, building agentic apps with Microsoft Fabric and Microsoft Databases; Microsoft Ignite; Palantir documentation on the Ontology, why an Ontology, action types, the action log, functions, Data Connection, Ontology MCP, the Ontology SDK, Ontology export and import, SuperRepo and Ontology-as-code, virtual tables, usage types and Workshop; Palantir Foundry plans; Palantir AIP Bootcamp; Databricks, Unity Catalog semantics; Snowflake documentation on semantic views, the Apache Ossie YAML export and the deprecated OSI YAML export; Directions on Microsoft, CIO Talk: Microsoft Gets IQ.