Kicking the tires on Microsoft Fabric

My daily stack is Snowflake and dbt with Power BI on top, which makes me exactly the person Microsoft is courting with Fabric: already inside the Power BI universe, warehouse loyalty theoretically up for grabs. So I spent a stretch of evenings actually building in it rather than forming opinions from keynotes. Mixed, specific feelings follow.

The genuinely compelling part is what OneLake does to the BI seam. Direct Lake mode reads delta tables without the import-refresh dance or DirectQuery’s performance apology. If you live in Power BI, and statistically your executives do, the gap between “data landed” and “dashboard current” nearly closing is a real, felt improvement. I got a working lakehouse-to-report path going in an evening, which surprised me. Onboarding is smoother than Azure Synapse ever was, a bar admittedly located underground.

The part that gave me pause is the all-or-nothing gravity. Fabric wants to be your ingestion, your lake, your warehouse, your notebooks, your BI, your orchestration. Adopt the bundle and the integration is lovely. Want to swap one piece, say keep your dbt project as the transform layer with everything versioned in git the way civilized people work? You can, mostly, but you’re visibly swimming against the current. The product’s happiest path assumes clicks over code, and the git integration, while it exists, feels like it was added for people like me rather than designed for people like me.

Capacity-based pricing also deserves a squint. One pool of compute units shared across workloads sounds simple until a heavy notebook job and month-end reporting want the same capacity at the same hour. Snowflake’s per-warehouse isolation, whatever its costs, makes noisy-neighbor problems somebody’s specific fault. Blended capacity makes them everybody’s mystery.

My scorecard, then. For a Microsoft-native org without an entrenched data platform, Fabric is a legitimately strong default and I’d say so out loud. For teams with a working Snowflake-and-dbt spine, the switching math doesn’t close yet; the BI-side wins are real but not rip-out-your-warehouse real. I’m keeping my hand in, because the Direct Lake idea clearly points somewhere, and because “working knowledge of the thing your CFO saw at a conference” has never once been wasted skill.