Apache Iceberg Lakehouse: Snowflake & Google Cloud

The announcement introduces a shared open table format, Apache Iceberg, enabling multiple analytics engines such as Snowflake, BigQuery, and Spark to read and write directly to the same data without proprietary adapters. This approach aligns with the Data Locality principle, which prioritizes moving compute to data rather than transferring large datasets across networks, improving speed, cost efficiency, and security. Iceberg serves as a common language for these engines, reducing interoperability challenges while requiring a governance layer—the catalog—to manage metadata, access policies, and concurrent writes. The solution addresses a longstanding industry need for a unified data architecture that avoids silos and redundant data movement.

Snowflake and Google Cloud now support catalog federation, allowing engines from one platform to access Iceberg tables managed by the other without copying data. This is achieved through standardized protocols like the Iceberg REST Catalog (IRC) and vended credentials, which grant temporary, scoped access tokens to requesting engines. For example, Snowflake users can query Google Cloud’s Lakehouse Iceberg tables using standard SQL, while Google Cloud services can access Snowflake-managed tables via Horizon’s IRC endpoint. The result is a fully interoperable lakehouse where governance, security, and access controls remain consistent across platforms.

Customers can choose between self-managed or managed Iceberg REST Catalogs, with options like Apache Polaris for DIY deployments or Snowflake Horizon and Google Cloud’s Lakehouse runtime catalog for managed solutions. Managed catalogs reduce operational overhead by handling infrastructure, scaling, and patching, while self-managed options offer full control at the cost of added responsibility. The announcement highlights that most organizations will prefer managed catalogs to focus on data rather than catalog operations, though both approaches support bidirectional federation.

The interoperability framework extends to AI applications, with programmatic access provided through standards like the Model Context Protocol (MCP) and REST APIs. These enable AI agents to query Iceberg tables directly, ensuring governed and authenticated access similar to human analysts. Additionally, semantic models and contextual intelligence layers—such as Snowflake Horizon Context and Google Cloud’s Universal Semantic Layer—ground AI responses in accurate, business-defined logic and metadata. This reduces hallucinations and ensures AI systems interpret data consistently, bridging the gap between accessible data and reliable AI-driven insights.