Semantic Layer Platforms

This is the most widely adopted category in analytics that Forrester has never ranked. Eighty four percent of enterprises report using semantic layer platforms, the firm's newest research says. And there is still no Wave. The category is planned, the report that precedes it is published, and the gap between the two is where this market's real story lives.

The preceding report, Make Data AI Ready Via Semantic Layer Platforms, appeared in June 2026 under VP and principal analyst Boris Evelson. It is a definition, a market map, and a set of statistics, everything a Wave needs except tiers. The flag on this category is correct. What is missing is not a scorecard to wait for. It is a market that has been adopted faster than it has been ranked, and the reasons for the delay are the article.

The report that precedes the scorecard

Evelson's June 2026 report carries the numbers that explain the category's position.

Eighty four percent of enterprises already use semantic layer platforms. Yet the average enterprise runs three BI and analytics platforms and four data platforms, and the result is inconsistent metrics, duplicated business logic, and disconnected access policies. Adoption without governance is how metrics multiply. The category's founding problem is not that organizations lack a semantic layer. It is that they have several, none of them shared.

The report also carries the infrastructure context. Eighty seven percent of surveyed cloud decision-makers run multiple cloud platforms, and thirty eight percent of enterprise data still lives on-premises. The semantic layer's job is to sit across that hybrid mess and hand every consuming tool the same meaning. And the report's sharpest number concerns AI: natural language to SQL accuracy runs at seventy to eighty percent in controlled benchmarks, and it degrades on real enterprise queries. The benchmark number is the case for the whole category.

The definition that changed the category

Forrester's definition of the next-generation semantic layer is worth reading in full, because it demotes the category's entire history.

A semantic layer is no longer a BI modeling convenience. It is "a control plane for data and analytics applications which centralizes and governs metric definitions, dimensional models, calculations, and access policies, and delivers them consistently to BI tools, data products, and AI agents, seamlessly integrated with the enterprise data fabric."

A control plane is where decisions about data stop being local. A metric definition written once, governed once, delivered everywhere: to a dashboard, to a data product, to an agent. The old semantic layer served one BI tool. The new one serves every consumer a company has. That is why the word platform replaced the word tool, and why the category now attaches itself to the data fabric instead of to a vendor's stack.

The semantic layer was the industry's most boring middleware. Agentic AI made it load-bearing.

Why AI made the boring layer load-bearing

The mechanism is the accuracy number.

An LLM that writes SQL from natural language gets the answer right seventy to eighty percent of the time in laboratory conditions, and worse against real schemas with real naming. For a dashboard that error rate is embarrassing. For an agent acting on the answer, buying inventory or moving a budget, it is dangerous. The fix Forrester is advancing is grounding: point the model at a governed semantic layer where metrics, dimensions, and calculations are already defined, and the model no longer has to guess what revenue means. Semantic grounding reduces hallucinations and improves explainability.

That is the quiet revolution in this category's fortune. For twenty years the semantic layer was infrastructure for the twenty percent of users who are power analysts. AI agents are the first consumer that cannot work without it. Every agent deployment inherits the metric chaos of the average enterprise, three BI platforms and four data platforms deep, unless someone builds the shared layer. The planned evaluation will be scored in this shadow: not which vendor models data prettiest, but which vendor can hand governed meaning to software.

The six factions that will share one scorecard

The June report maps the market across six architectures, and the planned Wave will have to rank them in one column.

The BI-native faction grew the semantic layer inside a BI platform, the model Strategy (formerly MicroStrategy) carries. The full-stack faction sells the layer as the product itself: AtScale, Cube, Kyligence. The data virtualization faction arrives from the query-federation side: Denodo, Dremio. The lakehouse faction embeds the layer in the platform: Snowflake, Databricks. The specialist faction is code-first and developer-owned: dbt Labs, GoodData. The data-graph faction argues the future is knowledge graphs: Stardog.

Six approaches, one definition, no tiers yet. The tension the planned Wave must resolve is structural. A BI-native vendor and a code-first specialist are not competing products. They are competing philosophies of who owns metric definitions, and a single axis will flatten them. Watch whether the eventual criteria favor governance depth, developer workflow, or agent delivery. That choice will decide which faction the category is written around.

The layer that sits on the fabric

Forrester's definition says the semantic layer integrates seamlessly with the enterprise data fabric, and the two categories must be read together.

The fabric carries the data. The semantic layer carries the meaning. This project's coverage of Forrester's Data Fabric Platforms Wave, the third name in a seven-year research stream, sits next to this article for a reason. A buyer evaluating fabrics is buying movement: integration, catalog, connectivity. A buyer evaluating semantic layers is buying agreement: one definition of revenue that every system obeys. The June report's market map, BI-native through data-graph, is the agreement market. The planned Wave will formalize it, and the first thing to check when it lands is how it treats the fabric vendors that already ship a semantic layer inside their platforms. Adjacent categories ranking overlapping products is how buyers get told the same vendor is in two different markets.

What the planned Wave will have to decide

The honest limitation of this category is that its scorecard does not exist, and the report that does exist refuses to fake it.

The June research is deliberately pre-ranking: a definition, a map, a set of survey statistics. It names no tiers because the market has not yet separated. That is defensible and useful, and it leaves a buyer with nothing to shortlist against except the map. If you need a ranked list today, this category cannot give you one. The coordination problem the report documents, three BI platforms and four data platforms, also describes the analyst's dilemma: the semantic layer market is genuinely fragmented across six architectures, and ranking them fairly requires criteria that do not exist yet, most obviously a way to score agent delivery, which every faction now claims and few can demonstrate at production scale.

Second, the planned flag itself is unconfirmed in public. No announced quarter exists for the Wave. The June report and the September 2026 webinars built on it are the only public anchors. Plan around those, not around a date.

Three questions the June report already raises

One: which faction already owns your stack? Your BI estate, your lakehouse, and your transformation tooling pin you to one or two of the six approaches before any vendor demo. The planned Wave will rank factions together, but your migration cost will not appear in the scores. Map your own faction first.

Two: can the layer serve an agent today, not just a dashboard? The report's central argument is that agents need governed metrics to be safe. Ask every shortlisted vendor for production agent integrations, not roadmap slides. The seventy to eighty percent accuracy number is the question every vendor should be able to answer against.

Three: what happens to your existing platforms? Forrester's research frames the semantic layer as bridging hybrid, multicloud, multi-tool complexity without rip-and-replace modernization. Make the vendor show the bridge to your three BI platforms and four data platforms, in your environment, before you believe the control-plane story.

The governed-access argument driving this category resurfaces in Data Fabric Platforms, where the two confirmed Leaders represent opposite answers to where an organization's data is allowed to live, a decision that determines which of this layer's six factions actually fits.

Forrester does not endorse any vendor named here, and tier placement should not be read as a recommendation to buy.

This market sits next to Metadata Management Solutions, covered separately on this site. Alation, Informatica, and Atlan lead Gartner's 2025 quadrant on the strength of one shift: metadata stopped being the librarian's back-office record and became the thing AI models read before every answer.