A brand new in-memory engine for planning in Microsoft Fabric, with 10x+ improvements in both performance and scale. No separate system database is required anymore.
This is Part 1 of our three-part Native Planning Engine series. Part 2 covers writeback to OneLake, and Part 3 covers the performance improvements across Fabric Plan.

The scale figure compares the Native Planning Engine’s 12M-cell aggregate limit with 1.2M cells on the previous planning engine. The calculation figure is compared with the March preview; for writeback, the full benchmark in Part 2 shows up to 13× at 12M cells. Performance varies with semantic model design and capacity. For every user-facing improvement, see Part 3.
You can now enable the Native Planning Engine in Tenant settings as a Public Preview capability, as announced on our roadmap at FabCon Barcelona. Planning inputs are stored in OneLake as Apache Arrow files and loaded in memory for planning, so Plan items no longer need a separate system database.
We are also excited to share that you can now write back directly to OneLake as a new writeback destination, in Delta Parquet format. We cover that in Part 2: Writeback to OneLake.
In this article:
Turn On the Native Planning Engine
The Native Planning Engine is turned on by the Store new Plan items in OneLake setting. Once it is on, planning inputs are persisted in OneLake, and business output tables can write directly to OneLake as Delta Parquet. Turn the setting on at two levels.
1. Tenant Level
Your Fabric administrator opens the Admin portal, goes to Tenant settings, turns on Store new Plan items in OneLake, and selects Apply.

2. Workspace Level
In the workspace that holds your Plan items, open Workspace settings, go to Plan settings under Delegated Settings, and turn on Store new Plan items in OneLake.

What to Expect After You Turn On the Setting
- New items use the new engine. All new Plan items are created in Native Planning Engine mode.
- Start with new items. Create new Plan items first and experience the better performance and scale.
- Move existing items when you’re ready. Follow the migration plan below.
At general availability, every remaining Plan item moves automatically. GA is planned in the next few weeks. Plan items still on the previous planning engine will be migrated to the Native Planning Engine, so we recommend moving and checking your key items during the preview.
Massive Scale and Performance with the In-Memory, File-Based Native Planning Engine
Fabric Plan now runs on a brand new native, file-based, in-memory planning engine that raises the planning limit tenfold, from 1.2 million to 12 million cells in aggregate. Apache Arrow supports efficient planning data access, while Delta Parquet makes business writeback outputs available across Fabric.
Plan items no longer need their own system database. Planning inputs live in OneLake, so there’s one less item to provision and manage, and storage is billed as part of your Fabric capacity. Fabric SQL remains a fully supported writeback destination, and PowerTable uses it as its writeback database.
Power BI semantic models can be massive, from millions to billions of rows or more. Because Fabric Plan runs directly on top of those semantic models, it can support any number of dimensions, measures and scenarios. That sets it apart from other planning tools, where performance degrades as the number of dimensions, measures or scenarios grows.
Flexible Enough for Every Team, Built for Enterprise Scale
Any planning method, with the controls to run it across every line of business.

Go deeper: tabular modelling and choosing a planning method, measure models, row models, driver-based planning, operational planning with PowerTable, writeback and automation, and governance with Audit Cell History.

The Inventory Grid and Calculation Paths
The illustration shows 12 million model cells, with 2.4 million cells affected by the bulk input operation.

Match Planning Scope and Capacity
Data fetched through XMLA affects capacity and responsiveness. The size of an interactive planning model is distinct from the volume published in a writeback instance. For scaling techniques such as context filtering, dimension partitioning, cubes and PowerTable, see Key Considerations for Planning at Scale with Microsoft Fabric Plan.
- Use filters and RLS to focus each user’s authorized planning scope. With these set up properly, most customers and users won’t need more than 2.4 million cells in a planning grid.
- If your use cases still need more than 12 million cells, consider multidimensional cube modeling, or dimension partitioning with consolidation through InfoBridge or PowerTable–Planning integration.
Caution: watch your Power BI XMLA query and concurrency limits. Large plans can reach the XMLA query and concurrency limits of your Fabric capacity, which can cause throttling. Limits depend on your capacity size, so consider a larger capacity when you need to fetch more Actuals data per query from the semantic model.
Roadmap: Up to 120 Million Cells in the Planning Grid
With this release, the Native Planning Engine supports up to 12 million cells in aggregate, 10x the 1.2 million cells of the previous engine. That total covers Actuals fetched from your semantic model plus visual forecast and formula measures. Next on our roadmap is support for up to 120 million cells inside the planning grid, 100x the previous limit. At that point, the only limit will be how much data you can fetch efficiently from your semantic models without throttling. Data fetched through XMLA alone will be capped at around 12 to 24 million cells.
This is a forward-looking roadmap item; release timing may change.
Make the Native Planning Engine Your Default
The Native Planning Engine is our default recommendation going forward.
New Capabilities Require the File-Based Engine
Advanced modeling features require OneLake storage. Audit cell history and driver-based modeling are available now, and snapshots (frozen versions) arrive on December 8, 2026. These capabilities are supported only by the Native Planning Engine: they require its file-based architecture and will not be supported by the previous planning engine.
Fabric SQL remains a fully supported writeback destination.
Move Existing Items When Your Team Is Ready
- Test with deployment pipelines. To try the Native Planning Engine before migrating, use deployment pipelines (CI/CD) to deploy an existing Plan artifact that uses the previous planning engine under a new name. Use Native Planning Engine mode for the new item and test its performance capabilities before migrating the original.
- During preview. After one or two weeks of getting comfortable with the native engine, migrate your existing Plan items gradually rather than all at once. Migration is one-way, so smoke test a copy first. Open Item settings and migrate each item to Native Planning Engine mode, or use deployment pipelines to copy an item into a brand-new one if you don’t need its data. The Migration to OneLake guide walks through each step.
- After migrating. Open your key sheets and Intelligence Sheets and check that results and reports look as expected.
- At general availability. GA is planned in the next few weeks. At GA, all customers and existing Plan items that have not yet moved will be automatically migrated to the Native Planning Engine.
Frequently Asked Questions
Who can turn on the Native Planning Engine?
A Fabric administrator turns on Store new Plan items in OneLake in Tenant settings. The setting can be delegated so that capacity and workspace admins can enable or disable it, and it is then turned on for each workspace in Workspace settings → Plan settings. See Turn on the Native Planning Engine.
Do existing Plan items change when I turn it on?
No. The setting applies to new Plan items. Existing items keep running on the previous planning engine until you migrate them, or until the automatic migration at general availability.
Is Fabric SQL still needed?
Yes, but not as the system database. Planning inputs are now stored in OneLake, which removes the need for a second database. PowerTable continues to use Fabric SQL as its writeback database, and Planning sheets continue to support Fabric SQL as a writeback destination, alongside OneLake writeback.
Does my data come with the item when I migrate?
Migrating an item from Item settings moves that item to the new engine. Copying an item with deployment pipelines creates a new item without its data, which is useful for testing. The recommended path is to smoke test first: copy the item with deployment pipelines, check it on the Native Planning Engine, then migrate the original from Item settings.
Can I move an item back to the previous planning engine?
No. Migration is one-way, so a migrated item can’t be moved back to the previous planning engine. Smoke test a copy first, as described above, before migrating the original.
What happens to the Fabric SQL system database?
Once every Plan item in the workspace has been migrated to OneLake, the Fabric SQL system database (the application database) can be safely deleted. If any Plan items in the workspace haven’t been migrated yet, leave the database in place so their metadata isn’t lost. Keeping it does no harm.
Does the new engine change what I pay?
The pricing model doesn’t change: Fabric Plan is still billed by session type on your Fabric capacity. What changes is the other resources part of the bill. Instead of charges for a Fabric SQL system database, you now incur OneLake storage. See how Fabric Plan pricing works.
Are there limitations during the preview?
No. There are no known limitations during the preview.
Resources and Feedback
- Migration to OneLake guide: step-by-step migration of existing Plan items
- Key Considerations for Planning at Scale with Microsoft Fabric Plan: scaling options and fetch-size guidance
- What is plan in Fabric IQ? (Microsoft Learn)
- Fabric Plan documentation
- Watch: the Native Planning Engine in action (LinkedIn)
- Your first plan in a few clicks: the other admin settings a plan needs
- Share feedback: tell us what works and what doesn’t in the Fabric Plan community forums.
The inventory illustration uses fictional data. Performance figures are examples, not performance guarantees.

Comments
0Please log in to comment.