The Data Brief · Issue #5 · Semantic Layer Series
dbt, Cube, AtScale, Looker, and the platform-native options — assessed without vendor influence
Every vendor says they are the semantic layer. That sentence has never been more true — or more confusing — than it is in 2026. The market has fragmented into standalone platforms, warehouse-native solutions, BI-embedded layers, and emerging AI-first tools.
84% of data teams still encounter conflicting versions of the same metric, with more than a third experiencing this regularly. This issue cuts through the noise. Five tools, assessed honestly. What each one does well, where it falls short, and the watch-out each vendor is unlikely to put in their own marketing materials.
The wrong question is "which tool is best?" The right question is: which tool fits my architecture, team skills, and use case today?
Ideal for: Teams with mature dbt practices who want metrics version-controlled alongside transformation code
dbt Labs acquired Transform — the company behind MetricFlow — and integrated it into dbt Cloud. Metrics are defined in YAML in the same Git repo as transformation models. BI integrations include Tableau, Google Sheets, Excel, Hex, Mode, Klipfolio, Omni, and Power BI (Preview — available via Power BI Desktop; Power BI Service requires IT-configured On-premises Data Gateway).
Ideal for: Teams building embedded analytics, custom data products, or multi-tool architectures
Cube is the most developer-friendly semantic layer — built open-source with nearly 20,000 GitHub stars. The 2025 release added roll-up anytime materializations and a WASM-powered query engine achieving sub-second P95 latency on Snowflake. Cube exposes metrics through REST, GraphQL, SQL, MDX, and DAX APIs — the broadest API surface of any standalone tool.
Ideal for: Large enterprises with complex governance requirements and Excel-heavy user bases
AtScale sits across Snowflake, Databricks, BigQuery, and on-premise systems simultaneously, exposes metrics through XMLA/MDX interfaces that Excel understands natively, and provides the deepest governance capabilities. In January 2026, AtScale joined the Open Semantic Interchange initiative and introduced an open-source Semantic Modeling Language.
Ideal for: Google-first organisations and shops where analytics flows through a single BI tool
LookML is the original code-based semantic layer and remains one of the most mature modeling languages available. The 2025–2026 updates integrated Gemini NLQ, with meaningful AI accuracy improvements when grounded in LookML's governed model. Google has been developing decoupled semantic capabilities — though these are still maturing.
Ideal for: Teams fully committed to a single data platform who want the lowest-overhead path
Snowflake Semantic Views reached full GA with standard SQL querying in March 2026 and introduced automated metric discovery capabilities. Databricks Metric Views extend Unity Catalog with metric governance that inherits existing access controls.
| Criterion | dbt SL | Cube | AtScale | Looker | Native |
|---|---|---|---|---|---|
| Multi-BI portability | ●●● | ●●● | ●●● | ●○○ | ●●○ |
| Setup speed | ●●○ | ●●○ | ●○○ | ●●○ | ●●● |
| Governance depth | ●●○ | ●●○ | ●●● | ●●● | ●●○ |
| AI / LLM readiness | ●●● | ●●● | ●●● | ●●○ | ●●● |
| Embedded analytics | ●○○ | ●●● | ●●○ | ●●○ | ●○○ |
| Cost at small scale | ●●○ | ●●● | ●○○ | ●●○ | ●●● |
| dbt-native teams | ●●● | ●●○ | ●●○ | ●○○ | ●●○ |
| Vendor stability | ●●○ | ●●● | ●●● | ●●● | ●●● |
● limited ●● solid ●●● best-in-class for this criterion · Ratings reflect architectural fit, not overall product quality.
"35% of teams still define metrics ad-hoc in BI tools or custom code — not because they don't know better, but because they haven't found a tool that fits their workflow without adding friction."
Next in the series
Build it: step-by-step implementation walkthrough with dbt and Cube