Microsoft Fabric is generally available: what GA covers, and what is still preview

Six months after the preview at Build, Microsoft Fabric is now generally available. Microsoft announced it at Ignite today, and Fabric can now be bought as a product, not just tried. For anyone who has been waiting for the "is it production ready?" signal before planning a real project, this is the signal.

But GA is a label on a platform, not on every feature in it. Plenty of what you saw in demos today is still in preview, and the difference matters when you sign off on an architecture. So here is what is GA, what is not, and how I would decide whether to start.

What reached GA today

According to the announcement from Arun Ulag, Microsoft Fabric is "now generally available for purchase". Alongside the platform itself, these items are called out as generally available:

• OneLake shortcuts to OneLake, Azure Data Lake Storage Gen2, Amazon S3 and Microsoft Dataverse.
• Direct Lake mode in Power BI, including Direct Lake on Fabric Warehouses.
• Microsoft Purview sensitivity labels and audit integration for Fabric.

The November 2023 update adds a few more GA items in the same release cycle: semantic model scale-out, shareable cloud connections for semantic models and paginated reports, the Lakehouse, Warehouse and KQL Database connectors in Dataflow Gen2, and Event Streams.

Microsoft also says around 25,000 organisations used Fabric during the preview. That is a useful number. It means the core engines have had real workloads running on them for months.

What is still in preview

This is the list I would print and keep next to the architecture diagram:

• Copilot in Fabric, starting with Power BI, Data Factory, Data Engineering and Data Science. It is rolling out in stages and needs F64 or higher, or Power BI Premium P1 or higher.
• Git integration and deployment pipelines for Lakehouse and notebooks.
• The Fabric User APIs (the new Fabric REST API).
• Row-level and object-level security plus stored credentials for Direct Lake semantic models.
• OneLake integration for import-mode semantic models.
• The Purview Hub.
• The environment item for Spark settings.
• VNet data gateway support for Dataflow Gen2.
• Data Activator, which is in public preview and enabled through a tenant setting.

Some of these are not side features. Lifecycle management, automation APIs and security on Direct Lake models are things most enterprise teams expect from day one. If your design depends on them, you are building on preview, whatever the platform badge says.

What GA means in practice

For me, GA changes three things.

First, it is now a purchasing decision. Pay-as-you-go capacity was announced in June, and Microsoft now adds reservation pricing in one-year commitments, smaller SKUs below the entry-level P SKU, and eligibility for Azure consumption commitments (MACC). Finance teams can now plan for it.

Second, the GA core is a reasonable foundation. Lakehouse and Warehouse storage in OneLake, shortcuts and Direct Lake reporting are the backbone of most Fabric designs, and those are the parts Microsoft is standing behind today.

Third, it does not change how you should treat the preview list. Preview features can still change shape before they land. Build around them only if you can tolerate rework.

A checklist before you start

If you are deciding whether to start a Fabric project now, work through this:

1. Map your design to the GA list. Write down every Fabric feature your solution needs. Mark each one GA or preview. If the critical path is mostly preview, plan a pilot, not a production go-live.
2. Decide on CI/CD. Git integration and deployment pipelines for Lakehouse and notebooks are preview. Agree on how you will promote changes between dev, test and production until that settles.
3. Check your security model. If you rely on row-level security in Direct Lake semantic models, remember that it is still preview. Test it with your real roles before committing.
4. Settle governance early. Sensitivity labels and audit integration are GA, so there is no reason to skip them. Set them up before data starts flowing.
5. Check Copilot expectations. If stakeholders expect Copilot, be clear that it is preview, rolls out in stages and needs F64 or P1 and above. Smaller capacities do not get it.
6. Use the trial properly. New customers can get a free 60-day trial without a credit card. Use it to run one real workload end to end, not a demo.
7. Size capacity on evidence. Run your pilot on a capacity close to what you expect to buy. Commit to a reservation only once you have seen real usage.

What this means for you

If you already run Power BI Premium, the step is small. Direct Lake on top of a Lakehouse or Warehouse is GA, and it is the most immediate reason to look at Fabric for reporting teams.

If you run a classic stack with Azure Data Factory, Synapse and Power BI, I would start with one bounded workload, such as a single subject area moved into a Lakehouse with Direct Lake reporting on top. Keep the preview features at arm's length until your team has a feel for capacity usage and lifecycle management.

If you are starting from scratch, Fabric is now a valid default to evaluate. Just be honest in your planning about which parts of the story are still preview.

Takeaway

After two decades in data, I read GA announcements with one question in mind: which features would I bet a production deadline on? With Fabric, the answer today is the core storage, Direct Lake and governance basics. The rest is promising, but still moving.

Are you starting a Fabric pilot now, or waiting for Git integration and Direct Lake security to reach GA first?

Sources

• Prepare your data for AI innovation with Microsoft Fabric, now generally available (Microsoft Fabric blog)
• Microsoft Fabric November 2023 Update (Fabric Updates Blog)
• Announcing the Data Activator public preview (Fabric Updates Blog)