Recently, Microsoft introduced composite Direct Lake models which I demonstrated as a part of a pilot project, mainly to preserve the Excel report experience for financial users.

Direct Lake Only vs Composite Direct Lake

I view the composite Direct Lake models as the best of both worlds and this table summarizes their characteristics.

Therefore, composite Direct Lake models could be a good compromise between real-time BI and flexibility. For example, now you can implement the following configuration:
1. Dimensions in Import mode and refresh them overnight since they probably don’t change frequently anyway.
2. Large fact tables or tables requiring real-time BI in Direct Lake without refresh.

Lessons Learned

If composite models sound appealing, you might be eager to convert an existing Direct Lake model to composite. Here are some issues/gotchas that I ran into doing so:

1. The web modeling experience (currently, the only way to add imported tables using Microsoft tooling) would show the Get Data buttons disabled. After some reverse-engineering of a brand new model, I fixed it my changing the connection expression in the *.bim file to use onelake (previously, it was pointing to a Fabric warehouse).

"expression": [
"let",
" Source = AzureStorage.DataLake(\"https://onelake.dfs.fabric.microsoft.com/44744d84-...\", [HierarchicalNavigation=true])",

1. If you’re switching many tables from Direct Lake, you can automate this in two ways:

• Semantic Links lab notebook

import sempy_labs as sl
from sempy_labs import migration
migration.migrate_direct_lake_to_import(
  dataset="Your_Direct_Lake_Model_Name",
  workspace="Your_Workspace_Name",
  mode="import" # or "directquery"
)

• C# script in the Tabular Editor

1. In my case, since there weren’t that many tables, I converted the dimension table partitions manually to “M” partitions, as in this example (change bolded text for each table):

   "partitions": [

{
"name": "Account",
"mode": "import",
"source": {
"expression": [
"let",
" Source = Sql.Database(\"<guid>.datawarehouse.fabric.microsoft.com\", \"<warehouse name>\"),",
" #\"Navigation 1\" = Source{[Schema = \"dbo\", Item = \"DimAccount\"]}[Data]",
"in",
" #\"Navigation 1\""
],
"type": "m"
}
} ]

1. After switching a partition from Direct Lake to Import in PBI Desktop project and synchronizing to the connected published model, Fabric rejected the change complaining that existing DirectLake table can’t be switched to imported storage. As a workaround, I dropped the connected model.
2. Being in public review, Composite Direct Lake is rough around the edges. I’ve got various complaints about credentials missing which I fixed in the dataset settings.
3. Although the documentation says that web modeling is the only tooling experience, Power BI Desktop worked for me as with the Direct Lake only counterpart. However, currently Power Query and Get Data (unless you add the tables directly in *.bim file) is only available on the web.

A “Limited” Performance Note

I know everyone is interested in performance. I did some limited performance tests by tracing a massive query against equivalent Direct Lake Only and Composite Direct Lake models. On a cold cache, Composite outperformed Direct Lake Only by some 20%. On a warm cache, surprisingly I saw the reverse, Direct Lake Only outperforming Composite five to six times. Please don’t take this verbatim. More than likely, your results will vary. For example, in that blog I said that I saw much better performance with SWITCH…CASE in Imported vs Direct Lake. Test!

Performance issues with calculated accounts

Financial MD cubes are notoriously difficult to convert to Tabular/Power BI because of advanced features that aren’t supported in the new world, such as Account Intelligence, scope assignments, parent-child hierarchies, and calculated dimension members. The latter presented a performance challenge. Consider the following MDX construct:

CREATE MEMBER CURRENTCUBE.[Account].[Accounts].&[1].[Calculations List].[ROI %]
AS IIF([Account].[Accounts].&[1].[Calculations List].[Average Invested Capital] = 0, NULL, ..,
FORMAT_STRING = "#,##0.0 %;-#,##0.0 %",

This construct adds an expression-based account as though the account is physically present in the chart of accounts. MD evaluates the MDX expression only for that account.

No such a construct exists in Tabular. To provide a similar reporting experience, I attempted to overwrite the Value measure conditionally based on the “current” account, such as:

VAR _Level03 = SELECTEDVALUE ('Account'[Level 03]) 
RETURN 
IF (_Level03 <> "Calculations List", 
       [Value],
       VAR _Level04 = SELECTEDVALUE ('Account'[Level 04])
       RETURN
            SWITCH (
               Level04,
               "ROI %", [ROI],
               "Economic Profit", [Economic Profit],
…

However, no matter what I tried, the report performance got a big dent (from milliseconds to 10+ seconds) even when the Calculations List account was excluded. Interestingly, report performance in Direct Lake fared 2-3 times worse than an equivalent Power BI imported model.

So, we had to scrap this approach in favor of one of these workarounds:

1. Pre-calculating the calculated accounts values (materializing)
   1. Pros: same reporting behavior as MD, faster performance compared to MDX expressions
   2. Cons: effort shifted to ETL, potentially impacting real-time forecasting if calculations must be recomputed with each change.
2. Separate DAX measures
   1. Pros: formulas applied at runtime as MD, no impact on ETL
   2. Cons: different report experience

Excel dropping user-defined hierarchies

Excel never fails to disappoint me. Sad, considering its potential as an alternative reporting client, especially for financial users.

This time Excel pivots decided not to show user-defined hierarchies, which turns out to be a document limitation for DirectQuery and Direct Lake. Microsoft provides no explanation and I’m sure the Excel team has no plans to fix it, as well as to finally embrace DAX and Power BI semantic models.

Luckily, the client uses a third-party Excel-based tool, which provides better report experience and supports user-defined hierarchies. If the Excel limitation becomes an issue, Fabric Direct Lake is expected soon to support composite models. This will let you implement models with hybrid storage, such as importing dimensions, which don’t change frequently, but leave fact tables in Direct Lake. Luckily, Excel supports user-defined hierarchies with imported tables.

Understanding the challenge

Power BI has long supported custom tooltip pages. Passing member context, such as the selected product category, is not an issue. However, passing the selected measure won’t work without some blackbelt magic because measures are not field members.

And so, we will have to turn the measures used in the matrix in a dimension. Then, instead of adding measures to the matrix as you would typically do, you would use the dimension. You can probably follow a similar implementation approach using Power BI field parameters, but since I’m a purist, I went with a regular table.

As a disclaimer, I’d rather show the trends in separate graphs side by side on a real-life dashboard, so the end user can see everything at a glance instead popping up tooltips and because I’ll have more control and flexibility over the visualization setup. But, just in case you must use tooltips…

Adding a measure “dimension”

Start by adding a new table to the model either way you want. In this case, I used the Power BI Enter Data feature to create a MatrixMeasures table. This table lists the measures that will be added to the matrix. The format column will be used for dynamic formatting later on.

Then, I added a DAX measure to the MatrixMeasures table as follows:

SelectedMeasure =
VAR _measure = SELECTEDVALUE(MatrixMeasures[Measure])
RETURN
SWITCH(TRUE(),
_measure = "Sales", [Sales],
_measure = "PY Sales", [PY Sales],
_measure = "% Growth", [% Growth],
[Sales]
)

This formula switches dynamically the DAX measure based on the “current” matrix column. For example, the Sales column will use the existing [Sales] measure. If nothing is selected, it piggy backs on the [Sales] measure. This default provides data to work with when you’re configuring the tooltip chart (otherwise, nothing will be shown because the matrix doesn’t pass the context).

While we are in the DAX rabbit hole, let’s add one more measure for dynamic tooltip chart title:

SelectedTitle = SELECTEDVALUE(MatrixMeasures[Measure])

Making report changes

Next, remove all the individual measures from the matrix and add only the SelectedMeasure measure to the matrix Values well. At this point, the matrix should look like before, except that the measure format settings would be lost.

Tip: If the measures need to be sorted in a specific way, add a new column to the MatrixMeasures table that defines the custom sort order. Then, sort the Measure column by this new column.

To fix formatting:

1. Select the SelectedMeasure field in the Data pane.
2. In the Measure Tools ribbon, change the Format dropdown to Dynamic.
3. In the formula bar, enter this formula to dynamically change the measure format using the format specified in the Format column of the MatrixMeasures table:

SELECTEDVALUE(MatrixMeasures[Format])

Then, implement the tooltip page and hook it up to the page with the matrix, as explained here. To change the chart title dynamically in the tooltip page, bind it to the SelectedTitle measure.

There you have it. With some wizardry and DAX, you have a measure context-aware tooltip.

The Telegraph Road, Dire Straits

At Ignite in November, 2025, Microsoft introduced Fabric IQ. I noted to go beyond the marketing hype and check if Fabric IQ makes any sense. The next thing I know, around the holidays I’m talking to an enterprise strategy manager from an airline company and McKinsey consultant about ontologies.

Ontology – A branch of philosophy, ontology is the study of being that investigates the nature of existence, the features all entities have in common, and how they are divided into basic categories of being. In computer science and AI, ontology refers to a set of concepts and categories in a subject area or domain that shows their properties and the relations between them.

So, what better way to spend the holidays than to play with new shaky software?

What is Fabric IQ?

According to Microsoft, Fabric IQ is “a unified intelligence platform developed by Microsoft that enhances data management and decision-making through semantic understanding and AI capabilities.” Clear enough? If not, if you view Fabric as Microsoft’s answer to Palantir’s Foundry, then Fabric IQ is the Microsoft equivalent of Palantir’s Foundry Ontology, whose success apparently inspired Microsoft.

Therefore, my unassuming layman definition of Fabric IQ is a metadata layer on top of data in Fabric that defines entities and their relationships so that AI can make sense of and relate the underlying data.

For example, you may have an organizational semantic model built on top of an enterprise data warehouse (EDW) that spans several subject areas. And then you might have some data that isn’t in EDW and therefore outside the semantic model, such as HR file extracts in a lakehouse. You can use Fabric IQ as a glue that bridges that data together. And so, when the user asks the agent “correlate revenue by employee with hours they worked”, the agent knows where to go for answers.

Following this line of thinking, Microsoft BI practitioners may view Fabric IQ as a Power BI composite semantic model on steroids. The big difference is that a composite model can only reference other semantic models while Fabric IQ can span data in multiple formats.

The Good

Palantir had a head start of a decade or so compared to Microsoft Fabric, but yet even in its preview stage, I like a thing or two about Fabric IQ from what I’ve seen so far:

• Its oncology can span Power BI semantic models (with caveats explained in the next section), powered by best-in-class technology. As I mentioned before, this allows you to bridge all the business logic and calculations you carefully crafted in a semantic model to the rest of your Fabric data estate.
• Fabric IQ integrates with other Microsoft technologies, such as real-time intelligence (eventhouses), Copilot Studio, Graph. This tight integration turns Fabric into a true “intelligence platform,” reducing duplicated logic, one-off models, and maintenance while enabling multi-hop reasoning and real-time operational agents.
• Democratized and no-code friendly – Visual tools allow business users to build and evolve the ontology, lowering barriers compared to more engineering-heavy alternatives. Making it easy to use has always been a Microsoft strength.
• Groundbreaking semantics for AI Agents: Fabric IQ elevates AI from pattern-matching to true business understanding, allowing agents to reason over cascading effects, constraints, and objectives—leading to more reliable, auditable decisions and automation.
• Compared to Palantir, I also like that Fabric OneLake has standardized on an open Delta Parquet format and embraced data movements tools Microsoft BI pros and business users are already familiar with, such as Dataflows and pipelines, to bring data in Fabric and therefore Fabric IQ.

The Bad

I hope some of these limitations will be lifted after the preview but:

• Only DirectLake semantic models are accessible to AI agents. Import and DirectQuery models are not currently supported for entity and relationships binding. Not only does this limitation rule out pretty much 99.9% of the existing semantic models, but it also prevents useful business scenarios, such as accessing the data where it is with DirectQuery instead of duplicating the data in OneLake.
• No automatic ontology building – It requires cross-functional agreement on business definitions, workshops, and governance—labor-intensive for organizations without mature semantic models. I hope Microsoft will simplify this process like how Purview has automated scans.
• Risk of overhype vs. delivery gap – We’ve seen this before when new products got unveiled with a lot of fanfare, only to be abandoned later.

The Ugly

OneLake-centric dependency. Except for shortcuts to Delta Parquet files which can be kept external, your data must be in OneLake. What about these enterprises with investments in Google BigQuery, Teradata, Snowflake, and even SQL Server or Azure SQL DB? Gotta bring that data over to OneLake. Even shortcut transformations to CSV, Parquet, JSON files in OneLake, S3, Google Cloud Storage, will copy the data to OneLake. By contrast, Palantir has limited support for virtual tables to some popular file formats, such as Parquet, Iceberg, Delta, etc.

What happened to all the investments in data virtualization and logical warehouses that Microsoft has made over years, such as PolyBase and the deprecated Polaris in Synapse Serverless? What’s this fascination with copying data and having all the data in OneLake? Why can’t we build Fabric IQ on top of true data virtualization?

Which is where I was thinking that semantic models with DirectQuery can be used as a workaround to avoid copying data over from supported data sources, but alas Fabric IQ doesn’t like them yet.

Summary

Microsoft Fabric IQ is a metadata layer on top of Fabric data to build ontologies and expose relevant data to AI reasoning. It will be undoubtedly appealing to enterprise customers with complex data estates and existing investments in Power BI and Fabric. However, as it stands, Fabric IQ is OneLake-centric. Expect Microsoft to invest heavily in Fabric and Fabric IQ to compete better with Palantir.

Performance

Performance is difficult to translate because Power BI Pro/PPU run in a shared capacity, meaning compute resources (v‑cores) are pooled across many tenants and dynamically allocated, whereas Fabric capacities are dedicated, meaning that Microsoft grants specific resources expressed as number of cores and memory. Therefore, Fabric performance is predicable while Pro/PPU might not be, although I’m yet to hear from client complaining about unpredictable performance.

Also, keep in mind that Power BI Pro limits you to a quota of 1 GB per dataset, PPU to 100 GB per dataset, and Fabric starts at 3 GB per dataset with F2 and doubles the grant up the chain. This is important for semantic models with imported data.

Although the tool wasn’t designed for estimating upgrade scenarios, you could start with the Fabric Capacity Estimator (preview) to get an initial ballpark estimate for the Fabric capacity. Start low, then monitor the capacity performance using the Microsoft Fabric Capacity Metrics app and be prepared to upgrade if necessary, such as when more parallelism is needed. 

Cost

This is easier. Here are the advertised, undiscounted and unreserved prices:

• Power BI Pro: $14/user/month (free with M365 E5 plan)
• PPU: $24/user/month ($14 discount with M365 E5 plan)
• Fabric: Starts at $262.80 per month with F2 and doubles the price up the chain. Finding what capacity you need requires evaluating what workloads you will be running to ensure you have enough resources.

It’s important to note that Fabric capacities lower than F64 require a Power BI Pro license for every user who accesses shared content, regardless of viewing or creating content. Microsoft Copilot and ChatGPT got this wrong by adamantly claiming that viewers don’t require Pro license, while Grok got it right, so be careful which agent you use when researching. The Fabric Capacity estimator also correctly identifies the required Pro licenses.

Of course, Fabric gives you features unfortunately not available in the pay per user licensing plans, so the actual decision in favor of Fabric will probably transcend just performance and cost. When evaluating the performance of the lower Fabric SKUs, you might find the following blogs I wrote on this subject helpful:

Notes on Fabric F2 Performance: Warehouse ETL

Notes on Fabric F2 Performance: Report Load

Somewhere along the way without me noticing, Microsoft added a better authentication option for refreshing Power BI semantic models: workspace identity. This option lets the Power BI workspace using its own managed identity to authenticate to the data source. And it’s available in all Power BI and Fabric SKUs!

What’s not clear from the documentation is how to grant permissions to the workspace identity to read data from Azure SQL SKUs but it’s no different that granting access to the Azure Data Factory managed identity.

1. Create the workspace identity as explained in the documentation. It has to be done for each workspace that has your published model(s).
2. In SSMS, connect to your Azure SQL Database using Entra credentials that has permissions to manage security (SQL login won’t work).
3. Open a new query connected to your database.
4. Execute the following script assuming you want to grant read permissions to the workspace identity:

CREATE USER [<workspace name>] FROM EXTERNAL PROVIDER;
ALTER ROLE db_datareader ADD MEMBER [<workspace name>];

Then back to Power BI, configure your semantic model for workspace identity authentication:

1. Navigate to the semantic model settings and click “Edit credentials”.
   
2. Select “Workspace identity” as the authentication method.

That’s it. Using the workspace identity to read data during model refresh is more secure and easier to manage.