Share with your CDO
Every major data platform is now positioning its semantic layer, a shared dictionary that tells AI and analysts what “revenue” or “customer” actually means across systems, as the foundation for enterprise AI. Microsoft, Databricks, Snowflake, and Salesforce have all shipped or rebranded semantic layer products in the past 18 months. The strategic logic is sound: without a common business vocabulary, AI agents give confidently wrong answers. But as this analysis of universal semantic layers makes clear, the vendor race is outpacing enterprise readiness.
What this means for your business
The organizations most exposed here aren’t the ones behind on tooling. They’re the ones that have bought a semantic layer product and mistaken the purchase for the work. Cross-functional alignment on definitions, what counts as an active customer, how churn is measured, which revenue line includes refunds, is the actual bottleneck, and no vendor ships a solution to that. If your data governance program hasn’t forced those definitional fights yet, your semantic layer is a well-labeled container holding conflicting answers.
The vendor consolidation pressure is real and worth watching on its own terms. When Salesforce wires Tableau Semantics directly into Agentforce, and Databricks routes its Genie agent through Unity Catalog Metric Views, they’re not offering a neutral infrastructure play. They’re building toll roads: the semantic layer becomes the switching cost. A CDO who standardizes on one platform’s semantic model will find it increasingly painful to route queries, agents, or analytics workloads anywhere else. The architecture decision is also a vendor lock-in decision, and that’s rarely surfaced in the procurement conversation.
The open-source counter, dbt Labs releasing MetricFlow at Coalesce 2025, is the most important hedge available to enterprises that want semantic portability. If MetricFlow gains enough adoption to become a de facto standard, the platform vendors lose their definitional monopoly. Watch whether the hyperscalers treat MetricFlow as a first-class integration or quietly friction it out. That signal, not the product announcements, tells you whether the semantic layer becomes infrastructure or inventory.
Concept deep-dive: Semantic layer
A semantic layer sits between raw data and the people or AI systems querying it, translating technical table names and fields into business terms everyone agrees on. Think of it as a company-wide glossary that also does the math: when an AI agent asks for “monthly recurring revenue,” the semantic layer defines the formula, not just the label. Without it, the same question asked of two systems can return two different numbers, both technically correct, neither trustworthy.
Based on reporting from Universal semantic layers: critical infrastructure or the next data fabric?, originally published 2026-06-15 06:04:00.
