Why every major data platform is investing in semantics
Over the last couple of years, the analytics industry has quietly converged on a single idea. Nearly every major data platform has introduced some form of semantic capability. Snowflake released Semantic Views. Databricks invested in Metric Views and Genie. Microsoft Fabric continues expanding its Semantic Models. dbt introduced its Semantic Layer, and Dataiku added semantic modeling capabilities of its own.
These companies compete aggressively. They have different architectures, different customer bases, and different product strategies. Yet they’ve all arrived at essentially the same conclusion: business meaning needs to be captured explicitly if organizations want AI to produce consistent, trustworthy results.
If this were only one vendor’s strategy, it might be easy to dismiss as product positioning. When multiple platforms independently converge on the same idea, however, it’s usually worth asking why. That kind of convergence often signals that the industry is responding to a fundamental shift rather than a temporary trend.
In this article, we’ll explore why semantic capabilities have become such an important investment across the analytics ecosystem, and why we believe they’re pointing toward a much broader shift in the way organizations will prepare their data for AI.
The bottleneck has moved
A few years ago, most enterprise AI conversations centered around the models themselves. We compared context windows, benchmark scores, hallucination rates, and reasoning capabilities. Those discussions were important because the technology was evolving rapidly, and organizations were trying to understand what these new models could actually accomplish.
Today, those conversations tend to sound very different. Most modern language models are already capable of generating SQL, summarizing datasets, writing code, and answering questions posed in natural language. As we’ve seen over the past several years, improvements in reasoning have been remarkable. The question is no longer whether AI can access our data. Increasingly, it’s whether AI understands what that data actually means. The bottleneck has moved from generating answers to understanding business context, and that’s a very different problem to solve.
Business meaning has always existed
Consider a question that sounds deceptively simple:
What was revenue last quarter?
At first glance, it appears there should be a single answer. In practice, there often isn’t. Depending on the audience and the business context, “revenue” might refer to gross revenue, net revenue, booked revenue, recognized revenue, adjusted revenue, or revenue converted into a common currency. Every one of those answers can be technically correct while producing materially different results.
It’s worth noting that the difficult part isn’t generating the SQL. Modern AI systems are remarkably good at writing queries once they understand which definition they’re supposed to use. The challenge is understanding what revenue means within your organization and applying that definition consistently.
Experienced analysts make these decisions almost instinctively because they’ve accumulated years of business context. AI doesn’t have that advantage unless we provide the context explicitly.
Why semantics matter
This is exactly the problem semantic models were designed to address. Rather than leaving business definitions scattered across dashboards, SQL queries, documentation, and institutional knowledge, semantic models provide a consistent description of the concepts that matter to the business. Metrics, dimensions, calculations, business rules, and relationships all become part of a shared vocabulary that both people and AI systems can use.
The result is greater consistency, improved governance, and a stronger foundation for natural language analytics. It’s easy to understand why every major platform has decided to invest here, because semantic capabilities address one of the largest obstacles organizations encounter when moving AI beyond experimentation and into production.
This is rapidly becoming table stakes – a baseline expectation, not a differentiator – for enterprise analytics.
But definitions are only the beginning
What’s particularly interesting, however, is that semantic models solve only part of the problem.
Knowing what revenue means doesn’t explain how revenue relates to margin. Understanding what an active customer is doesn’t explain what influences customer retention. Business definitions provide a vocabulary, but organizations also operate through relationships, dependencies, and business processes that extend well beyond individual metrics.
As AI evolves from answering questions to assisting with investigations and business decisions, those relationships become increasingly important. Understanding why something happened often requires traversing multiple business concepts rather than simply retrieving a definition.
That’s where the next phase of enterprise AI begins.
Looking beyond the semantic layer
Semantic models represent an important step forward, and the industry’s investment reflects that. At the same time, we believe they’re part of a larger evolution. We’re beginning to see organizations recognize that business meaning itself – the definitions, relationships, business rules, and institutional knowledge that describe how the company operates – has become increasingly important as AI becomes part of everyday business operations.
Exactly what that evolution looks like – and what it means for enterprise architecture – is a much larger conversation.
