AI Fuels Battle for Semantic Layer Supremacy

The convergence of artificial intelligence and business intelligence is exposing a critical dependency: the semantic layer. As AI models increasingly generate SQL queries to interact with databases, their ability to understand precisely what data resides within a table becomes paramount. This is where the semantic layer steps in, acting as an essential map that translates natural language queries into accurate data retrieval, making its control a burgeoning battleground in the tech world.

Prior to the AI revolution, semantic layers were primarily a concern for specialists implementing BI tools and modeling data within data warehouses. Organizations adopting platforms like Tableau or Looker on top of analytical databases from Oracle or Teradata would typically leverage the semantic capabilities embedded within those BI products. Their function was clear: to define and enforce rules about data, including source validation, necessary transformations, and integrity checks. A Chief Financial Officer, for instance, relied on the semantic layer to guarantee that “operating revenue” figures on a dashboard were calculated consistently and correctly every single time, making it a crucial component of the analytical infrastructure.

The advent of large language models, particularly since ChatGPT’s emergence in 2022, has propelled the semantic layer from a niche data modeling exercise into a cornerstone of the AI-BI landscape. Unleashing an AI model on a database without this contextual map is a recipe for misinformation, frustration, and potential business setbacks. Without a guiding semantic layer, language models are almost guaranteed to misinterpret data, leading to errors that could have significant business implications.

For the past couple of years, the humble semantic layer has steadily gained prominence. Vendors that developed independent semantic layers, such as AtScale and Cube, which standardize how analysts interact with underlying databases regardless of the BI tool, have significantly ramped up their development and marketing efforts. Even dbt Labs, known for its popular data transformation tool, launched its own semantic layer in 2023.

Now, the industry’s titans are staking their claims. Snowflake introduced its “semantic views” feature at its recent Summit 2025. Not to be outdone, Databricks unveiled its “Unity Catalog metric views” at its AI & Data Summit 2025.

Snowflake’s semantic views are described by its engineers as a new schema-level object that natively stores all semantic model information directly within the database. This innovation replaces previous metadata files and establishes a standardized metadata definition for a wide array of Snowflake experiences, encompassing AI-powered analytics, BI clients, and custom applications. Currently in beta, Snowflake offers various methods for creating these views, including a user interface, a database object explorer, or direct DDL statements. The company has adopted a model definition language allowing users to define core attributes like physical model objects (tables or views), relationships between them, dimensions (business-friendly attributes for grouping and filtering), and metrics (business-friendly calculations representing KPIs). Snowflake engineers emphasized that semantic views address a common customer concern: the desire for AI-powered conversational analytics without the risks of ungoverned data access or inconsistent results.

Databricks is pursuing a similar path with its Unity Catalog metric views, building upon its centralized data catalog and governance offering. Databricks asserts that defining metrics at the data layer, rather than solely within the BI layer, ensures reusability and integration across all workloads, from dashboards to AI models and data engineering jobs. These metric views, defined in YAML and registered in Unity Catalog, are fully accessible via SQL, promoting a consistent view of metrics across an organization regardless of the tool used. Databricks highlights that metric views are governed and auditable by default, providing certified metrics with built-in auditing and lineage for trusted insights. Expected to reach general availability this summer, these views can be created once in Unity Catalog and applied across various Databricks tools. In the future, Databricks plans to extend support to external BI tools like Tableau, Hex, Sigma, ThoughtSpot, and Omni, as well as observability tools such as Anomalo and Monte Carlo.

The strong demand for a robust semantic layer is undeniable, as evidenced by the strategic moves from Snowflake and Databricks. Without this crucial metadata layer, the promise of natural language query for business databases will likely remain unfulfilled.

The pivotal question now is whether the momentum for semantic layers will be sufficient for them to emerge as an independent product category, separate from the BI tools or data platforms they were historically linked to. The recent actions by Snowflake and Databricks, integrating semantic capabilities deeply into their platforms, suggest a preference for ecosystem-bound solutions. However, history offers a different precedent: further up the data stack, the demand for an independent table format led to the widespread adoption of Apache Iceberg. Both Snowflake and Databricks ultimately standardized on Iceberg, a victory for data independence and a setback for vendor lock-in.

The semantic layer has similarly materialized as a vital component for achieving data interoperability and ensuring the repeatability and reliability of AI-powered BI. The ultimate outcome — whether the industry giants will converge on a universal, open standard that benefits all, or if they will seek to make semantic layers a proprietary competitive advantage — remains to be seen.

[[AI’s power relies on understanding data. Now, a fierce battle is erupting for control of the essential ‘semantic layer.’] ]