Power BI & Excel on Snowflake Without Data Movement | Semantic Layer for AI

Most enterprises have already standardized where their data lives. What they haven’t standardized is how that data is interpreted.

Power BI and Excel users still rely on a workaround: moving data out of Snowflake into extracts, cubes, or Microsoft Fabric just to make it usable. That workaround introduces latency, duplicates storage, and fragments governance. And according to Gartner, poor data quality, much of it caused by this kind of architectural fragmentation, costs organizations an average of $12.9 million per year.

Another challenge with data movement is inconsistency: the same data can mean different things depending on which tool is querying it. 

Data movement breaks analytics because it:

• Creates duplicate data copies
• Introduces latency and stale data
• Breaks governance and access controls

A universal semantic layer is how you solve it. A semantic layer is a logical data layer that standardizes business definitions, metrics, dimensions, and relationships, enabling every tool and AI system to query data consistently.

I recently walked through this architecture live with Josh Klahr, Director of Product Management at Snowflake, in a webinar covering how Power BI and Excel can connect directly to Snowflake without data movement, extracts, or Microsoft Fabric. What follows draws on that conversation.

Why the Semantic Layer Is Becoming Critical AI Infrastructure

Gartner’s recent research makes the trajectory clear: by 2030, universal semantic layers will be treated as critical infrastructure, alongside data platforms and cybersecurity. No longer a BI enhancement. Infrastructure.

The implication is straightforward: a semantic layer is no longer optional. It is required to deliver accuracy, manage AI costs, reduce technical debt, align multi-agent systems, and prevent inconsistencies at scale. Semantic capabilities now need to be budgeted as core infrastructure, not layered on after problems emerge.

The market is responding accordingly. According to Futurum Research, semantic layer adoption is accelerating from 16% in 2026 to 30% by 2031, outpacing growth in core data categories. 

Taken together, Gartner signals inevitability, and Futurum confirms acceleration. Enterprises that treat semantics as optional architecture are already falling behind.

Why AI Requires a Semantic Layer

For years, semantic layers were primarily a BI concern. They helped business users get consistent answers without writing SQL. It was a convenience story.

AI changes the requirements entirely.

Enterprises are moving from chatbots that answer questions to autonomous agents that take actions. Large language models do not operate reliably without a consistent business dictionary, deterministic definitions of metrics, hierarchies, and relationships. Without that foundation, AI systems produce conflicting answers, erode trust, and drive up compute costs.

The $12.9 million annual cost of data inconsistency compounds when AI is involved. Every additional agent, workflow, and decision that runs on inconsistently defined data multiplies the problem. The semantic layer prevents inconsistency from becoming an exponential liability.

Why Data Movement Breaks Semantic Consistency

“Moving data is the curse of all things analytics,” said Josh. “When you move data, you kill your RBAC. You have to re-implement all of your access controls. Moving data introduces latency, costs, and complexity.”

He’s exactly right.

Every time data moves:

• Security is re-implemented
• Metrics are redefined
• Governance is rebuilt, imperfectly

At scale, that leads to diverging definitions, brittle pipelines, and systems that are harder to maintain than the data they were meant to simplify.

Every copy of data creates another version of the truth.

The Shift: Stop Moving Data. Start Standardizing Meaning.

The alternative is straightforward in principle, though technically demanding in practice: query the data where it lives, and let the semantic layer handle the translation.

AtScale’s semantic layer sits between the consumption tools, Excel, Power BI, AI agents, and the underlying data platform. It translates DAX, MDX, and SQL into native Snowflake SQL in real time. It standardizes metrics and dimensions so that every consumer, human or agent, uses the same definitions. It enforces governance automatically, without requiring each tool to re-implement its own security logic.

The semantic layer is not a data store. It’s a translation and governance layer that gives every downstream system access to a shared, governed understanding of the business.

Power BI and Excel on Snowflake Without Data Movement

Architectures that depend on Microsoft Fabric, extracts, or cubes solve for performance by moving data. A semantic layer solves it without moving data.

In practice, this means Power BI and Excel connect directly to Snowflake through live DAX and MDX queries. There are no imports, row limits, or extract pipelines to maintain.

For Excel users, this means live pivot tables that operate on the full Snowflake dataset. Cell-level queries are backed by live data. There’s no latency from stale exports and no data duplication to manage.

For Power BI users, this means full-feature support for semantic models inherited from the AtScale semantic layer—not rebuilt from scratch in Microsoft Fabric. Business logic defined once in the semantic layer flows to Power BI without duplication or fragmentation.

The practical impact is that data governance no longer breaks at the consumption layer. Snowflake’s security model, row-level permissions, column masking, and role-based access work seamlessly because the data never leaves Snowflake.

Performance and Cost: Optimization Without Tradeoffs

One concern that consistently arises is performance. Live connectivity sounds good architecturally, but enterprises run thousands of concurrent queries, and sub-second response time is not optional for adoption. This stands in contrast to architectures that depend on pre-loading or duplicating data to achieve performance.

AtScale addresses this with continuous query intelligence. The system observes query patterns, automatically builds aggregates where they’re needed, rewrites queries for efficiency, and applies caching where appropriate. Performance improves over time as the system learns actual usage patterns.

In benchmarking against Microsoft Fabric’s DirectLake, AtScale delivers 14x better user concurrency and 100% query success at enterprise scale. Total cost of ownership is 30–70% lower than for architectures that require data movement to a secondary platform.

Beyond compute efficiency, organizations also achieve cost savings by eliminating entire categories of infrastructure: ETL pipelines, duplicate storage, secondary compute charges, and the engineering labor required to keep those systems synchronized.

One Model and One Answer Across BI and AI

When Excel users, Power BI users, and AI agents all query the same semantic model, optimizations are shared, and definitions are consistent.

“When you have high-quality semantics, and you’re using those same semantics across AI and BI, you can satisfy the needs and address the fears of traditional BI and analytics teams and allow them to unlock AI while getting a governed, consistent experience,” explained Josh.

A universal semantic layer is what allows “revenue” to mean the same thing in an executive dashboard, a regional manager’s Excel forecast model, and a Cortex Analyst natural language query. That consistency is a requirement, especially for regulated industries, AI-driven automation, and any organization making consequential decisions from data.

Why Semantics Becomes Mandatory for Agentic AI

The transition from answering to acting changes what AI systems need from the data layer. In this environment, the semantic layer functions as the control plane for multi-agent systems: enforcing a shared understanding of the business, preventing metric drift across agents, and reducing AI technical debt before it accumulates.

Agentic AI requires deterministic metrics and governed access. It cannot navigate raw schemas reliably. LLMs are inherently non-deterministic; they guess when context is ambiguous, and in complex data environments, ambiguity is everywhere. The semantic layer removes that ambiguity, so AI queries govern metrics rather than raw tables. Hallucinated joins and schema misinterpretations are eliminated because the agent never sees the raw schema.

With native support for Model Context Protocol (MCP), AtScale ensures that AI applications—including Claude, Cortex Analyst, and other agent frameworks—operate on consistent, trusted definitions.

The Semantic Layer Is Now Non-Negotiable

Most enterprises have solved data democratization. Now they have a consistency problem.

As AI increases the number of decisions made from data, the cost of inconsistency compounds. A universal semantic layer is what ensures those decisions are grounded in a shared, governed understanding of the business.

That’s why it’s no longer optional architecture. It’s required infrastructure.

See how AtScale enables Power BI and Excel to connect live to Snowflake without Fabric or compromise. Watch on demand now.

---

FAQ: Semantic Layer and Snowflake