A reader says to me yesterday very directly during a linkedin exchange: "All of the data leadership is focused on Fabric." Nine words. He is right and I will not argue the point. The evidence is everywhere. But 'where leadership is focused' and 'what works for the small shops' are two different things, and only one of them gets answered in the keynote. Fabric is the destination. The question nobody on stage is asking is who is actually ready to go there -- and what the costs may be for those who are not.
The Tell Isn't a Keynote. It's a Feature.
You do not have to read Microsoft's strategy in a press release. You can read it in what they built into the flagship database release. SQL Server 2025's headline analytics capability is not a faster engine or a smarter optimizer. It is Mirroring to Microsoft Fabric, and it went generally available alongside the release.
Here is what it does. You point Fabric at your SQL Server, pick the tables you want, and it continuously replicates them into OneLake -- Fabric's storage layer -- with no ETL, no pipeline, no orchestration. On SQL Server 2025, the engine itself scans the transaction log at high frequency and publishes the changes; Fabric merges them in near-real-time, as fast as every 15 seconds. On 2016 through 2022, it rides on Change Data Capture instead. Either way the data lands in OneLake as read-only Parquet, ready for analytics, BI -- and AI.
Read that again, because the direction is the whole story. The marquee analytics feature of the newest SQL Server is a one-click, zero-ETL pipe whose entire job is to copy your data out of SQL Server and into Fabric. When the headline of your flagship database is 'we made it trivial to send your data somewhere else', that somewhere else is where the value is being built.
One SQL, One Direction
The positioning also backs the plumbing. March 2026 brought the first-ever SQLCon -- run not on its own, but inside FabCon, framed as unifying databases and Fabric on a single platform. The database and the platform now share one stage and one story. And the momentum behind that story is not subtle. By Microsoft's own numbers, Fabric has passed a 2 billion dollar annual revenue run rate, serves more than 31,000 customers, and is growing about 60% year over year -- the fastest-growing analytics product they currently have.
Set that next to the other half of the picture. Fabric is now the named successor to Azure Synapse. Synapse is still supported, but Microsoft has been clear that all new analytics innovation is going into Fabric, and specific Synapse pieces are already being retired in its favor. Line the signals up, and they all point the same way.
| The signal | What it tells you |
|---|---|
| Mirroring SQL Server to Fabric went GA, on-prem included |
Microsoft built the on-ramp from your estate into theirs, zero-ETL. |
| SQL Server 2025 added a native log-scan publisher for Fabric |
The flagship engine release's analytics headline is, in effect, 'feed Fabric.' |
| The first SQLCon ran inside FabCon |
The database and the platform now share one stage and one roadmap. |
| Fabric named Synapse's successor; new analytics work goes there |
The investment is pointed at the SaaS platform, not the box engine. |
The data flows one way. Operational SQL Server feeds analytical Fabric. Your engine is being recast as the trusted source of record -- the place where your data is born and kept honest -- but it is also no longer the place where the analysis happens.
What This Is Not
Now the part the panic merchants will skip. Fabric is not replacing your transactional SQL Server, and nobody is migrating your order-entry system to a lakehouse. Mirroring is read-only and analytical by design. The copy to OneLake is for reporting and AI, not for inserts or writing. The OLTP engine that runs your business is going nowhere.
So this is not a funeral, and it is not a reason to let a vendor sell you a frightened migration. The engine is healthy. What has changed is not whether SQL Server survives. It is where the new value gets built on top of it.
Who's Ready? The Part Nobody's Costing
Here is where the keynote goes quiet. Fabric is the destination, fine. But the people who own a SQL Server, a rack of SSIS packages, and a budget they have to defend are not asking 'is Fabric the future.' They are asking 'does this work for my shop', and 'can I afford it'. Those answers are a lot less tidy than the slide deck.
Start with the money, because the shape of it is the opposite of what you own today. A SQL Server license is a capital cost on hardware you control. You buy it, you run it, it is yours. Fabric is a meter. You rent capacity by the hour, and the meter runs every hour the capacity is on. Here is the pay-as-you-go list, US region, per month.
| Fabric capacity | Pay-as-you-go (list, per month) |
|---|---|
| F2 (entry) | ~$263 |
| F8 | ~$1,051 |
| F32 | ~$4,205 |
| F64 | ~$8,410 |
F2 at about $263 a month looks approachable, and that is the number that gets quoted. Two things it does not tell you. First, the way you make pay-as-you-go cheap is to pause the capacity nights and weekends. But, the near-real-time mirroring needs the capacity running to stay near-real-time, so the live feed and the pause-to-save trick will counter each other. Second, and this is the one that makes people blink, below F64 every single report viewer still needs a Power BI Pro license, roughly $10 per user per month. The capacity that lets viewers in for free is F64, at about $8,410 a month -- roughly thirty times the F2. So the small shop that picks F2 to be frugal pays the entry rate plus a per-seat tax on everyone who opens a report, and the next rung up that removes the tax is a canyon away. Reserved pricing trims the capacity bill by 30 to 40 percent, but it commits you to paying around the clock, idle hours included. You have traded a server you owned for a subscription you cannot turn off.
None of that makes Fabric a bad platform. It makes it a different cost model, billed monthly, forever, and pointed at organizations big enough to keep a capacity busy. For the small and mid-size shop, 'free of a license' is not the same as 'cheap,' and nobody selling the destination is doing this arithmetic out loud.
And SSIS.
This is the part that costs me something to write. SSIS has been a workhorse in my career and a lot of yours, and it still ships with SQL Server 2025 and is fully supported. But Fabric has no SSIS runtime. The data-movement story over there is Data Factory pipelines and Dataflows Gen2 -- which means an SSIS-heavy shop does not migrate to Fabric, it rewrites. Every package is a re-implementation in a different tool, and the investment, the tooling, and the momentum are all on the new side. You can keep running SSIS for years. You just cannot bring it with you. Pretending otherwise does not change the rewrite waiting at the other end, and that rewrite is exactly the kind of cost that turns 'we are moving to Fabric' into a multi-year line item nobody scoped.
What It Means for the On-Prem DBA
If the gravity has moved, the practical question is what that does to your job. A few things, and none of them are panic.
The investment is moving up the stack.
This is the same pattern I wrote about in the Postgres piece. Microsoft feeds the layer where its customers are spending, and right now that layer is the SaaS analytics platform, not the box engine. The engine gets hardening and careful maintenance. Fabric gets the new features and the 60% growth. Plan your attention accordingly.
Mirroring makes your data Fabric's concern, which makes the plumbing yours.
A mirror is not free of consequences for the DBA. It means a dedicated Fabric login reading your transaction log, an on-premises data gateway sitting inside your network, and a real decision about which tables leave the building. That is a new security surface, and it lands on your desk. Scope the login down, watch the gateway, and treat 'what gets mirrored' as a data-governance question, not a checkbox. Least privilege, again, same as it ever was.
The skills hedge is cheap.
You do not have to become a Fabric engineer. But knowing how your data gets mirrored, and who can reach the copy in OneLake, is now part of the job description whether you asked for it or not. Learning the mirroring path on a test database is a few hours. Call it insurance.
The Bottom Line
So yes -- follow the headcount and you find the roadmap, and it all says Fabric. I am not arguing with that. My reader was right. The data leadership is focused on Fabric and the feature list proves they mean it. But focus does not also mean fit. The destination is set for the large, capacity-filling, analytics-hungry shop. For the small one, the SSIS-heavy one, the one still scrambling to clear the 2016 finish line, the honest answer to 'who is ready' is 'not yet', and the road may be longer and more expensive than anyone on stage is admitting. That is not a reason to refuse the trip. It is a reason to cost it out properly, before someone who has never met your SSIS estate puts a date on the calendar for you.
More to Read
Microsoft Fabric Blog: Mirroring for SQL Server in Microsoft Fabric (Generally Available)
Microsoft Learn: Mirroring in Microsoft Fabric (overview)
Microsoft Azure: Microsoft Fabric pricing (capacity F-SKUs)
Microsoft Fabric Blog: From Azure Synapse and Azure Data Factory to Microsoft Fabric
sqlfingers inc: Why Are We Still Paying for SQL Server?
No comments:
Post a Comment