Don’t Let Microsoft Be Your Warden

Microsoft just made the case for AtScale by playing Warden Norton, the prison boss from “The Shawshank Redemption.”

On Tuesday, Sara Vredevoogd from Microsoft’s Power BI team announced that Microsoft 365 Copilot can now query Power BI semantic models directly. She explained why:

“If you’re investing in AI-powered analytics experiences, it’s worth investing in the quality of the semantic layer that sits underneath them.”

That’s the AtScale pitch. From Redmond. (We’re based in Boston.)

Look closer, though. Microsoft’s “semantic layer” runs inside Power BI, against Power BI models, designed to work with Copilot. Give Microsoft its due: Azure is growing 40% a year, and its AI revenue is real. But the seat where this new feature lives, Microsoft 365 Copilot, is the part that isn’t landing. Despite bundling it into the world’s largest enterprise software suite, fewer than 4.5% of Microsoft’s 450 million commercial Microsoft 365 customers pay for it. In an internal memo this summer, Microsoft told its own 11,000-person Copilot team the product must “earn the right to exist.”

The AI model layer is competitive; Anthropic, OpenAI, Cursor, Databricks, and Snowflake are making bold moves, and Microsoft itself now routes Copilot to OpenAI, Anthropic, or its own models. Locking your semantic layer, the definitions every one of those tools depends on, inside a single vendor’s product is the bet that doesn’t survive the next reshuffle. It’s the age-old lock-in play, and it’s the losing move no matter how strong that vendor’s platform is.

That’s not a semantic layer. It’s a semantic jail cell.

Here’s the good news: there’s a readily available, Shawshank-style tunnel to daylight. It has three sections.

1. An independent semantic layer, not a single-platform one. A universal semantic layer runs on any warehouse (including, but not limited to, Microsoft), any coding agent like Claude Code or Codex, any knowledge-work agent like Claude Cowork, and any BI tool, from Power BI to Tableau, through a single standards-based, performant engine. That’s AtScale’s MCP-wrapped universal semantic layer.
2. Open standards for semantic independence. Apache Ossie is the vendor-neutral standard for moving semantic definitions across the stack. Snowflake, Databricks, Salesforce, and AtScale have all signed on. Microsoft, tellingly, has not.
3. Open standards (MDX, DAX) to reach back to legacy Microsoft tools. These protocols connect Power BI and Excel to any open backend you choose, from Snowflake to Google to Microsoft. You stay free to choose.

That’s the path out of the Microsoft semantic jail.

Redmond is right about the semantic layer: it’s worth investing in one. But don’t let Microsoft become your Warden Norton, running the yard and pocketing the profits while you do the work. Microsoft wants to own your stack. AtScale sets you free.