A few weeks ago at Build, Microsoft pulled the sheet off Microsoft Fabric, and the data corner of the internet has been a little breathless ever since. The pitch is genuinely big: take Power BI, Synapse, Data Factory, and a pile of adjacent services, fold them into a single software-as-a-service product with one storage layer underneath — OneLake — and sell it as an end-to-end analytics platform where everything already talks to everything. It's the most ambitious thing Microsoft has done in data in years, and I've spent the weeks since it landed being asked the same question by clients and former colleagues: should we move? So let me give the answer I've been giving, which is more measured than the excitement around it and, I think, more useful: understand it now, plan for it, and — for almost everyone — don't touch production with it yet.

Let me be clear that I'm not being a cynic here. I think the idea behind Fabric is right, and I've believed for a while that the industry was drifting toward exactly this kind of consolidation. The modern data stack, for all its power, had become a sprawl of tools that each did one thing well and left you to pay the integration tax — the connectors, the duplicated governance, the seams where data got lost between systems. Fabric's bet is that most organisations would rather have one coherent platform than the best-of-breed menagerie, and that bet is a sound reading of where a lot of tired data teams actually are. When I wrote a while back about what was hype and what was real in the modern data stack, the unglamorous truth was that integration cost was eating the benefits. Fabric is Microsoft's answer to that, and it's a serious one.

What Fabric actually is, without the gloss

Strip the marketing and Fabric is three ideas stacked together. First, one storage layer: OneLake, a single logical lake for the whole tenant, so data lives in one place in an open format instead of being copied between six services. Second, one product surface: the things you used to buy and wire together separately — the warehouse, the lakehouse, the pipelines, the reporting, the real-time bits — presented as workloads inside a single SaaS experience. Third, one commercial model: capacity you buy once and spend across all of it, rather than a bill per service. Put together, the promise is that you stop being a systems integrator of your own data platform and start just using one.

That's the promise. The reality, today, is a preview — and the word matters.

Why 'wait' is the right answer right now

Here's the part the hype skips. Preview means exactly what it says: the product is early, features are incomplete, behaviour will change, and the things you'd need to run it seriously in production — the governance maturity, the cost predictability, the operational documentation, the hard-won knowledge of how it breaks — do not yet exist because nobody has run it long enough to find out. Betting a production workload on a preview platform is not being an early adopter; it's volunteering to be the person who discovers the sharp edges the hard way, on data that matters.

So the guidance I'm giving, tailored by who's asking:

  • If you're on Power BI already (which is most people): keep doing what you're doing. Fabric builds on the foundation you already have, and you lose nothing by letting it mature while you watch.
  • If you're mid-decision on a new platform: factor Fabric into the plan, but don't pin the delivery to a preview. Design so you can adopt it when it's ready without having bet the project on it being ready on your timeline.
  • If you're technically curious: absolutely get into the preview — in a sandbox, with non-critical data, to build real understanding. That's how you'll be ready to advise well when GA comes. Learning is free; production risk is not.
  • If a vendor or an internal enthusiast is pushing you to migrate now: ask them who's going to own the incidents on a preview platform, and watch the enthusiasm recalibrate.

What I'd actually do with the wait

"Wait" is not the same as "ignore," and this is where the measured position earns its keep. The organisations that will get the most out of Fabric when it's genuinely ready are the ones using this preview period to do the unglamorous groundwork that Fabric will reward but not provide: getting your data into clean, open formats; sorting out the governance and cataloguing you've been deferring; understanding your real workloads and costs so you can size capacity sanely later. Fabric concentrates whatever you bring to it — good practice and bad — so the preview window is a gift of time to make sure what you bring is good.

The counter-view deserves a hearing, because it's not silly. There are organisations for whom getting in early is genuinely worth the risk — greenfield builds with no legacy to protect, teams with the depth to absorb a preview's rough edges, situations where the strategic value of being early outweighs the operational danger. If that's you, and you go in clear-eyed about what preview means, good. But that's a small minority, and even they should keep anything that truly matters off it until GA. For the broad middle — the organisations just trying to run their reporting and not set money on fire — the boring answer is the right one. Understand it, prepare for it, and let someone else find the bugs.

Fabric is coming, and I think it'll matter. But "it'll matter" and "you should move today" are different sentences, and the space between them is where a lot of expensive mistakes are about to be made. Stand in that space calmly. The platform will still be there, more finished, when you're actually ready for it.