Video Summary - Microsoft Fabric: Synapse Lessons, Open Semantic Models, and Databricks Competition w/ Microsoft CVP

Key speakers / source

Bogdan (Microsoft CVP, “Ashure data analytics” / Azure data & analytics context)

Synapse’s core positive: First attempt to unify many Microsoft data/analytics assets under one product, letting customers choose the right engine for the right workload:
- SQL server components
- serverless SQL
- dedicated SQL pools
- Spark
- telemetry / KQL
Big architectural shortcoming: Even though multiple engines were offered together, they didn’t fully run on the same underlying data format/standard at the time.
Operational trade-off in Synapse:
- Serverless SQL: strong elastic/scalable architecture, but described as lacking a strong query optimizer and missing warehouse-like features such as indexes.
- Dedicated SQL pools (data warehouse Gen2): powerful data-warehouse capabilities, but scaled only up to a capacity ceiling and couldn’t operate as elastically as serverless.
Why Fabric was built (as a solution): Synapse succeeded with many customers, but Fabric was created to solve the core cross-engine architectural problems, especially around unified format and engine integration.

Fabric is described as “starting from scratch,” not optimizing for Synapse backward compatibility.
No backward compatibility at the storage/data format level: the data format changes completely.
Compatibility at the “dialect/IP” level: SQL dialect compatibility helps items like:
- stored procedures
- views …“just work,” implying syntactic compatibility.
Recommendation for Synapse customers today: migrate to Fabric at their own convenience.
Migration tooling is “ready”: customers can reuse investments in:
- stored procedures / views
- ecosystem assets such as pipelines
- Spark
- Power BI integration
Support stance: Synapse is still fully supported, with innovation primarily going into Fabric.

Microsoft’s broader strategy: keep data in Azure for integration, security, hardware, and global distribution.
Goal: reduce “lock-in” by providing customers access to multiple engines and options.
Fabric’s starting point: unifying Microsoft data technologies so customers don’t need to move data across formats between engines.
Unified open-ish data format focus: Delta was discussed as a choice over alternatives like Iceberg / Hudi, aligned with customer needs and ecosystem reality.
Competition framing: engines from multiple teams/vendors compete on performance within an open standard, benefiting customers.
Where they differ / where security matters:
- The big challenge is cross-engine security (not just row-level security inside a single engine).
- Microsoft claims Fabric provides a cross-engine security layer so BI tools (e.g., Power BI) can apply RLS across the same storage used by other engines.
Databricks customers are supported: relational integration remains a priority, including:
- Direct Query
- Direct Lake patterns Databricks customers are “welcome.”

The speaker addresses concerns about openness at the semantic model layer.
Specific change: Power BI stops supporting metric views (presented as a deliberate call by the speaker).
Reason: metric views led to cases where Power BI produced wrong numbers, so reliability couldn’t be guaranteed.
Operational/UX rationale: unsupported or unreliable semantic-layer behavior would cause support calls (“numbers are wrong”) that Microsoft engineers can’t reliably resolve.
Future stance: Microsoft wants to explore better ways to integrate semantic layers from other vendors, potentially via:
- standardized semantic model approaches (the speaker references “metric use for DataBricks” and other options like OSI/open approaches)
- making the Microsoft semantic model more accessible via open approaches
No objection to openness: Power BI / Analysis Services is described as open long-standing (XMLA / open protocol context referenced).

Timeline:
- Public preview: May 2023
- General availability: November 2023
Biggest challenge described: unifying end-to-end analytics across multiple engines so it feels like a single product, not separate experiences (notebook vs serverless vs warehouse).
Interface unification problem:
- A single unified UI risks alienating users who prefer specialized UIs (notebooks, Synapse serverless).
- Fully separate UIs risk looking like an incoherent “many tools in a trench coat.”
Noted technical complexity: security/private connectivity causes different behavior across Spark, warehouse, and Power BI consumption.
What improved / “wins” claimed:
- uniform access and same format across engines
- progress toward one security layer (described as shipped after ~3-year promise)
- better balance of common suite experience while respecting different user workflows
- engine performance improved significantly after GA
What still needs work: monitoring and governance are flagged as areas with less progress than desired.

Prediction: increasing agentic development/consumption of data.
LLMs/agents shouldn’t “do joins” directly; instead they should generate queries, while specialized query engines execute efficiently.
Because agents can create a wider spectrum of query shapes than typical human BI workflows, engines must handle:
- high concurrency
- heterogeneous query patterns
- high volumes
Documentation shift: from HTML toward markdown, with agent-oriented “API-like” interfaces and skills.
Core data modeling expertise still matters:
- star schema concepts
- normalization
- preparation for efficient execution …but it’s used differently alongside agentic technologies.
Convergence claim: AI and BI consumption patterns will become less distinguishable.

Differentiator is described less as raw infrastructure and more as Microsoft bridging analytics into business productivity tools:
- Office / M365 integration
- business-user-centric workflows
- security controls like information protection labels (example: preventing forwarding of Excel derived from Power BI reports outside the company)
Philosophical end-to-end framing: analytics is a means to business decisions, and Microsoft aims to connect data insights directly into the tools where decisions happen.

Bogdan — Microsoft CVP (Azure data & analytics / Microsoft Fabric discussion)