Snowflake Semantic View Defined
A Snowflake Semantic View is a schema-level database object that stores business metrics, dimensions, and entity relationships directly inside Snowflake. Core DDL reached general availability in mid-2025, and standard SQL querying against Semantic Views reached GA on March 2, 2026, with continued enhancements through Snowflake Summit 2026.
A Semantic View sits between the raw tables in Snowflake and the business language data teams and AI agents use to ask questions. It encodes the authoritative definition of a metric once, inside the warehouse, so every agent and analyst draws from the same governed source.
A native Snowflake database object that defines metrics, dimensions, relationships, synonyms, and verified query examples on top of raw tables.
A BI tool, a separate semantic layer product, or a metric store that runs outside Snowflake. A Semantic View lives inside Snowflake and only sees Snowflake data.
Data engineering and platform teams to author the definitions, Snowflake's AI products (Cortex Analyst, CoCo, CoWork, Cortex Sense) to consume them, and a universal semantic layer like AtScale to extend them to every BI tool and external agent the enterprise runs on.
How a Snowflake Semantic View Works
A Semantic View is defined as a schema-level object with five main components:
- 1
Tables and join paths — Predefined relationships so the engine doesn't have to guess at joins.
- 2
Dimensions — Name and describe the attributes the business cares about (region, plan tier, fiscal quarter), with synonyms so natural-language queries can match them.
- 3
Metrics — Define the calculations the business has agreed on (Q3 revenue, active customers, gross margin), with comments and verified examples.
- 4
Time intelligence — Anchors fiscal calendars, snapshot logic, and period-over-period comparisons.
- 5
Governance — Flows through Snowflake's role-based access control and row-level masking, so a Semantic View restricted in Snowflake stays restricted at query time.
Querying a Semantic View can happen two ways: through the original SEMANTIC_VIEW clause, or through standard SQL clauses, which reached GA on March 2, 2026.
What Snowflake Semantic Views Solve
Snowflake Semantic Views solve the meaning problem inside Snowflake, with several concrete benefits:
One definition for Snowflake's AI surfaces. Cortex Analyst, CoCo, CoWork, and Cortex Sense all consume Semantic Views as their governed foundation, so the same metric returns the same number whichever Snowflake AI product asks for it.
Versioned business logic inside the warehouse. Semantic Views are first-class database objects, which means they can be versioned, tested, and managed with the same tools the data team already uses for tables and views.
Natural-language accuracy. Cortex Analyst reads the descriptions, synonyms, verified examples, and join paths in a Semantic View to generate SQL against physical tables, with Snowflake reporting roughly 85–90% accuracy on natural-language queries against well-defined Semantic Views.
Foundation for Horizon Context. Semantic Views are the construct Horizon Context uses to store governed business meaning, which makes them the unit of work for any Snowflake governance program.
Where Snowflake Semantic Views Stop
Snowflake Semantic Views are warehouse-bounded, which is the most important limit to understand. They see Snowflake data and nothing else. They don't reach dbt models that live outside Snowflake, Looker metrics, Salesforce objects, Jira data, or any of the dozens of other systems most enterprises run on.
They also don't natively render in the tools the business uses every day. Power BI, Excel, Tableau, Looker, and ThoughtSpot don't consume Snowflake Semantic Views without a layer in between. The AtScale Snowflake Semantic Views guide describes the gap directly: "Snowflake Semantic Views centralize governed metric definitions inside Snowflake, but Microsoft tools like Power BI and Excel do not natively consume those definitions."
Other constraints worth noting:
1 MB YAML spec limit per Semantic View, with a recommended ceiling around 50–100 columns for Cortex Analyst.
No cross-account replication for Semantic Views as of mid-2026.
No automatic time grains in the native construct.
Business Glossary is on the H2 2026 roadmap, so the natural-language layer above Semantic Views is still maturing.
How to Extend Snowflake Semantic Views to Every Tool the Business Uses
A universal semantic layer like AtScale for Snowflake reads the Semantic View definitions already governed inside Snowflake and serves the same definitions to Power BI, Excel, Tableau, Looker, Google Sheets, and external AI agents through standard interfaces (XMLA, MDX, DAX, JDBC, MCP).
The AtScale for Snowflake Semantic Views XMLA Endpoint ships as a Snowflake Native App with a single DDL install, Snowflake RBAC passthrough, and an adaptive aggregate cache that compounds Snowflake performance instead of bypassing it. The data team continues to govern Semantic Views inside Snowflake, and the business continues to consume those definitions in whatever tool is on their desk.
The data team governs inside Snowflake. The business consumes everywhere else — in Power BI, Excel, Tableau, and AI agents — from the same definitions.
What the Anthropic Benchmark Says About Semantic Views and AI Accuracy
The accuracy case for Semantic Views stopped being theoretical when Anthropic published its own numbers. In How Anthropic's AI Accuracy Went from 21% to 95%, AtScale CTO Dave Mariani walks through what happened when Anthropic's data science team gave Claude raw access to thousands of SQL files, dashboards, transforms, and notebooks. Claude answered correctly about 21% of the time, and accuracy moved to roughly 95% once a semantic layer was in place with a rule telling Claude to consult it first.
AtScale's production benchmark with the commercial banking arm of a Tier 1 global bank cut compute by up to 21,000x and lifted accuracy from around 70% to 100% on a set of five common questions, with projected savings of about $9 million a year on that one question set. The Snowflake Semantic View is one piece of that picture, and the universal semantic layer is what carries the same accuracy and cost profile out to every consumer.
After adding a semantic layer to Claude's analytics workflow.
Accuracy from ~70% to 100% on five common questions, $9M/yr projected savings.
Frequently Asked Questions
What's the difference between a Snowflake Semantic View and Horizon Context?
A Snowflake Semantic View is a single schema-level object that defines metrics, dimensions, and relationships. Horizon Context is the broader governed semantic and context layer built on top of Horizon Catalog, which includes Semantic Views along with governance, lineage, and the active context that CoCo, CoWork, and Cortex Sense consume.
Does Power BI read Snowflake Semantic Views natively?
No. Power BI doesn't natively consume Snowflake Semantic Views. A universal semantic layer like AtScale is the bridge that exposes those governed definitions to Power BI through XMLA, MDX, and DAX.
Can a Semantic View pull from data outside Snowflake?
No. Semantic Views are warehouse-bounded. For cross-system context (dbt models, Salesforce, Oracle, SAP, BI tool metrics), the governed definitions either need to land in Snowflake first or be extended through a universal semantic layer that spans the broader stack.
What's Semantic View Autopilot?
Semantic View Autopilot, which reached GA on February 3, 2026, analyzes existing Snowflake tables and auto-generates Semantic View DDL, mapping entities, inferring relationships, and suggesting metrics. It's intended to reduce the time to build a first Semantic View from days to minutes.
Do CoCo, CoWork, and Cortex Sense all depend on Semantic Views?
Yes. Snowflake's AI products consume Semantic Views as their governed foundation. The quality of the Semantic Views directly bounds the accuracy of the AI surfaces above them.
What's Semantic Studio?
Semantic Studio is the AI-assisted IDE for building Semantic Views without writing SQL, announced at Summit 2026 in private preview. It pairs with CoCo for AI-assisted authoring.
Key Takeaways
A Snowflake Semantic View is a schema-level database object that stores governed metrics, dimensions, and relationships inside Snowflake — the unit of work that Horizon Context, CoCo, CoWork, and Cortex Sense all depend on.
Semantic Views are warehouse-bounded, which means they see Snowflake data and nothing else, and they don't natively render in Power BI, Excel, Tableau, or external AI agents.
A universal semantic layer like AtScale extends Snowflake Semantic View definitions to every BI tool and AI agent the enterprise uses, without data movement and without rebuilt logic.