For years, the semantic layer quietly sat behind the scenes of enterprise data systems, translating complex data models into the business logic that made them meaningful. It was a critical component for BI teams but was rarely viewed as a strategic asset.
That has changed. In the age of AI, the semantic layer has become the connective tissue between enterprise data and the large language models (LLMs) that increasingly drive analytics, decision-making, and automation.
The 2025 GigaOm Radar for Semantic Layers & Metric Stores highlights this shift. For the first time, GigaOm classified the semantic layer category as mature, moving beyond its earlier status as an emerging space. The reason is clear: semantic layers have evolved from helpful abstractions into essential infrastructure. Without them, AI cannot think or reason accurately.
From Business Intelligence to Artificial Intelligence
When AtScale was founded more than a decade ago, the mission was to create a universal semantic layer: a single, governed source of business truth that every analytics tool, data application, and AI system could share.
At the time, that idea felt ambitious. Today, it is a necessity.
A universal semantic layer separates business logic from BI tools and data platforms, ensuring that metrics, dimensions, and relationships remain consistent everywhere they are used. That consistency is no longer just valuable; it is mandatory for AI systems that depend on accurate, context-rich data to operate correctly.
As GigaOm’s Category Lead for Data & Analytics, Andrew Brust, put it in our recent discussion,
“The word ‘semantic’ used to be aspirational. Now it’s literal. If AI doesn’t understand what your data means, it’s going to have to guess or hallucinate.”
When data is grounded in a semantic layer, every query, whether it comes from a BI dashboard or an AI agent, returns consistent and governed results. Trust becomes part of the architecture, not an afterthought.
Why AI Needs a Semantic Foundation
AI models are not inherently fluent in business logic. When prompted with a question such as “What were Q3 sales in Canada?”, an LLM does not know which table to query, how to define “sales,” or which joins to apply. It has to infer meaning, and that is where mistakes happen.
The semantic layer provides the missing context. It defines relationships between tables, standardizes metrics, and embeds business rules directly into the data model. It effectively translates enterprise data into a format that AI can interpret without guesswork.
A universal semantic layer does more than expose metadata; it operates as a semantic engine. It simplifies complexity for the AI model, ensuring that queries are consistent, accurate, and reproducible. By removing ambiguity, it minimizes the chance of LLMs misinterpreting data or fabricating results.
When combined with the Model Context Protocol (MCP), an open standard for connecting AI models to external data systems, the semantic layer allows LLMs to reason over data with precision. The result is AI that is grounded in governed truth instead of statistical probability.
Seeing Semantics in Action
During the GigaOm webinar, I shared a live demonstration that showed how these concepts come to life in practice.
Using AtScale’s Semantic Modeling Language (SML), I asked the platform to automatically generate a new semantic model from raw data in Snowflake. Within seconds, the model was created, versioned in Git, and deployed, no manual coding or data movement required.
Then, using Claude, an AI assistant connected through MCP, I began querying the model in natural language. Claude automatically discovered the model, understood its schema, and generated SQL queries that respected the business logic encoded in the semantic layer.
When I asked, “Show me sales by product for Canada,” the AI did not have to guess which tables to join or how to calculate revenue. It simply used the semantic model as its map. The query executed against Snowflake returned accurate, context-aware insights every time.
This wasn’t just a convenience demonstration. It was a proof point for a larger shift in how AI interacts with data. By pairing an LLM with a semantic layer, we transform AI from a pattern matcher into a reasoning system.
Because the semantic layer abstracts the complexity of joins, hierarchies, and time intelligence, the AI can focus on generating insights rather than constructing queries. This makes analysis both faster and more reliable, while ensuring that results are deterministic, governed, and repeatable.
As Andrew noted during the discussion,
“AI is eager—it wants to please. Without guardrails, it can return arbitrary results that look authoritative but aren’t. The semantic layer makes sure it can’t hurt itself.”
That’s the power of pairing semantics with AI: the machine gets smarter, and the business gets safer.
Generative AI Meets Generative Models
The relationship between AI and semantics now goes both ways. Just as AI benefits from the context provided by the semantic layer, it can also contribute to building it.
AtScale’s open-source SML allows AI systems to automatically generate semantic models, commit them to Git for version control, and deploy them instantly across connected tools.
This development closes the loop between AI and data governance. AI no longer just consumes context; it helps create it. This evolution represents the next stage of enterprise data maturity: semantic intelligence, where meaning itself becomes a programmable asset.
Universal Over Embedded
In recent years, major data and analytics vendors have started introducing their own embedded semantic solutions, such as Snowflake’s Semantic Views, Databricks’ Metric Views, and BI tools that bundle semantics directly into their products.
This progress validates the importance of semantics, but it also risks repeating past mistakes. When semantic logic is embedded in individual tools, it fragments. Each platform defines its own version of the truth, leading to drift and duplication.
The only way to ensure consistency across the enterprise is to keep semantics independent and universal. A universal semantic layer ensures that every tool, agent, and model uses the same data language, without bias toward any single platform.
In a world where interoperability drives innovation, openness is not an option. It is the requirement for trust.
The ROI of Context
Organizations that adopt a universal semantic layer consistently report measurable improvements in analytics performance and data efficiency. In real-world AtScale customer deployments, teams have seen dashboards refresh several times faster, query workloads consume significantly less compute, and AI models deliver more consistent results.
Across industries, these benefits often translate to:
- Faster insights through centralized logic and optimized query performance
- Lower compute costs enabled by intelligent caching and workload management
- Higher AI accuracy when large language models are grounded in governed semantics
These outcomes are drawn from aggregated results observed across AtScale’s customer base and internal benchmark testing. By establishing a shared semantic foundation, enterprises achieve both speed and confidence in their analytics and AI workflows.
Data teams spend less time reconciling differences and more time driving insight. AI systems become reliable collaborators rather than unpredictable black boxes.
The Semantic Decade
The enterprise AI revolution is accelerating, but it is built on a fragile foundation. Without a semantic layer, AI lacks context, and context is what turns data into knowledge.
We are entering the Golden Age of Semantics, a decade in which the semantic layer transitions from supporting analytics to powering intelligence itself.
Enterprises that invest now are not just modernizing their BI; they are future-proofing their AI. The companies that lead in the next decade will not be the ones with the largest models but the ones whose models understand what their data means.
Download the 2025 GigaOm Radar Report for Semantic Layers & Metric Stores
Discover why AtScale was recognized as a Leader and Fast Mover for innovation, universality, and AI readiness.
See the full conversation between GigaOm’s Andrew Brust and AtScale CTO Dave Mariani, including the live demo of how AI and semantic layers work together to deliver governed, intelligent insights.
SHARE
Case Study: Vodafone Portugal Modernizes Data Analytics