AtScale

A commercial semantic layer that presents warehouse tables to Excel, Power BI, Tableau and AI agents as cubes and metrics, translating their MDX, DAX and SQL into warehouse SQL and building aggregate tables to speed them up.

What it is: Semantic layers
Type: Commercial product
Made by: AtScale, Inc.
License: Proprietary; its Semantic Modeling Language (SML) specification and converters are Apache 2.0
Facts checked: 24 October 2026

What it is

AtScale sits between BI tools and a cloud data platform such as Snowflake, Databricks SQL, BigQuery or Redshift. Modelers define a semantic model, which AtScale calls a cube: datasets, dimensions with hierarchies, metrics, and the relationships between them, built in its Design Center or written as code in SML, its YAML-based Semantic Modeling Language, and kept in Git. Each tool then connects through the protocol it already speaks. Excel and Power BI connect over XMLA as if AtScale were a SQL Server Analysis Services cube, with DAX and Live Connection for Power BI; Tableau and Looker connect with a PostgreSQL driver, because AtScale emulates the PostgreSQL interface. AI assistants connect through AtScale’s MCP server. Whatever the client sends, the engine rewrites it as SQL for the warehouse, so data stays where it is and every tool gets the same definitions of each metric.

To keep interactive queries fast without copying data out, AtScale creates aggregate tables in a schema in your warehouse: system-defined ones that AtScale builds automatically, demand-defined ones generated from the queries it sees, and user-defined ones for cases those miss. It routes later queries to a matching aggregate, rebuilds aggregates fully or incrementally for new fact rows, and drops stale ones. AtScale runs as containers on Kubernetes, or as a Snowflake Native App inside a customer’s Snowflake account. Pricing is not published: it is billed monthly by the number of deployed semantic objects, not by users or queries.

Why teams pick it

Excel pivot tables and Power BI connect to warehouse data as if it were an Analysis Services cube, including hierarchies and semi-additive measures, with no extracts
One set of metric definitions serves Excel, Power BI, Tableau, Looker, Python and AI agents over MCP
Automatic aggregate tables in the warehouse speed up repeated dashboard queries without a separate cache server
Models can be code: SML is an open YAML specification that lives in Git, with converters to and from other semantic formats

Watch out for

Commercial software with no published prices; only the SML specification and tooling are open source
Each BI tool reaches it through an emulated protocol (XMLA, DAX, PostgreSQL), so some tool features behave differently or are unsupported; the docs list functional differences per tool
Aggregates live in your warehouse and cost storage and compute to build and refresh, and need periodic full rebuilds after large dimension changes
Another service to deploy and run on Kubernetes unless you use the Snowflake Native App

Specs

Compare with alternatives →

Data model: Semantic models (cubes): datasets, dimensions with hierarchies, metrics
Query language: MDX, DAX and SQL from clients, rewritten as warehouse SQL
Client protocol: XMLA (as Analysis Services), PostgreSQL wire protocol, MCP
Ways to run it: Containers on Kubernetes, or a Snowflake Native App
Pricing model: Not published; billed monthly by deployed semantic objects

Alternatives

What all semantic layers have in common →Compare side by side →

CubeOpen sourceAn open-source, headless semantic layer: define metrics, joins and access rules once in YAML, query them over SQL, REST or GraphQL, and serve repeated queries from rollups it builds ahead of time.

dbtOpen sourceSQL select statements in git, built into warehouse tables in dependency order, with tests and docs alongside.

LegendOpen sourceGoldman Sachs’s data modelling and governance platform, open-sourced through FINOS: describe business concepts once in the Pure language, map them to databases and services, and query or publish them as APIs while the engine generates the SQL.

How it fits with other technologies

What AtScale uses

An SML model object: relationships, dimensions and metrics in YAML · yaml

unique_name: Internet Sales
object_type: model
label: Internet Sales
visible: true

relationships:
  - unique_name: factinternetsales_Date_Dimension_Order
    from:
      dataset: factinternetsales
      join_columns:
        - orderdatekey
    to:
      dimension: Date Dimension
      level: DayMonth
    role_play: "Order {0}"

dimensions:
  - Color Dimension
  - Size Dimension
  - Style Dimension
  - Weight

metrics:
  - unique_name: orderquantity
    folder: Sales Metrics
  - unique_name: salesamount
    folder: Sales Metrics

See it in a real stack

No architecture uses AtScale yet. Drop it into the playground and wire it up yourself.Open playground →

Where it came from2013–2024

Dave Mariani ran data pipelines and analytics teams at Yahoo! when Hadoop was born, then engineering at Klout, where a Hive warehouse held over a trillion rows. Business users there wanted to explore that data in Excel and Tableau, but Hadoop offered no OLAP server. AtScale, founded by veterans of Yahoo’s data team, launched in April 2015 as a Hadoop-native analysis server: BI tools queried it like a multidimensional cube, and it answered from data in Hadoop without moving it.

As analytics moved to cloud warehouses, AtScale repositioned. In December 2018, raising a $50 million Series D led by Morgan Stanley under CEO Chris Lynch, it called itself a data warehouse virtualization platform for Snowflake, BigQuery and others. It later adopted the term semantic layer, shipped as a Snowflake Native App in June 2024, open-sourced SML under the Apache licence in September 2024, and now pitches its engine as a semantic layer for AI agents, served over MCP.

Why it’s built this way

Speak each BI tool’s native protocol: Excel and Power BI expect an Analysis Services cube and Tableau expects a SQL database. Emulating XMLA and the PostgreSQL interface lets people keep their tools while the data stays in the warehouse.
Build aggregates inside the warehouse: Queries over raw fact tables were too slow for interactive pivots, but copying data to a separate OLAP store creates another silo; aggregate tables in the warehouse speed queries up while keeping one copy of the data.
Open the modeling language: Semantic models were locked into each vendor’s format. Publishing SML, with converters to formats such as Power BI semantic models and Snowflake’s, aims to make models portable.

Timeline

2013
AtScale introduces its semantic layer, according to the company
2015
Launches publicly in April as an OLAP server for Hadoop
2018
$50 million Series D led by Morgan Stanley; repositioned for cloud warehouses
2024
Snowflake Native App in June; SML open-sourced in September

Sources7 · facts checked 24 October 2026

Improve this page on GitHub