Snowflake and Google Cloud unite data platforms for AI with open standards

Snowflake and Google Cloud have announced an integrated data platform architecture built on Apache Iceberg and catalog federation, designed to let multiple data engines coexist without data duplication while maintaining enterprise governance.

The foundation is Apache Iceberg, an open table format originally developed at Netflix for petabyte-scale table management and later donated to the Apache Software Foundation. Iceberg is now supported by Spark, Trino, Flink, BigQuery, Snowflake, and dozens of other engines. Instead of each engine maintaining its own copy of data, all engines agree on how data is physically laid out so each can read and write directly to the same table in the customer's own storage bucket. This follows the Data Locality principle: it is faster, cheaper, and more secure to move executable code to where data lives than to move massive data volumes across a network.

The catalog layer provides governance. Snowflake's Horizon Catalog and Google Cloud's Lakehouse Runtime Catalog each serve as the single authority managing table metadata, access policies, concurrent writers, and consistency guarantees. Both expose their tables via the Iceberg REST Catalog (IRC)—a standardized REST API that defines how clients discover namespaces, load table metadata, and commit updates. Both catalogs use vended credentials: when an engine requests table data, the catalog returns short-lived, narrowly scoped storage tokens rather than persistent bucket credentials, so engines gain no persistent access to the underlying storage layer.

Catalog federation enables bidirectional access across platforms. When Google Cloud manages a catalog via Lakehouse Runtime Catalog, Snowflake connects through a Catalog-Linked Database (CLD) that automatically discovers and syncs tables from its IRC endpoints; Snowflake users interact with those tables using standard SQL or via Snowflake's CoCo or CoWork as if they were native tables, with Snowflake governance (RBAC, masking, lineage) layered on top. Similarly, when Snowflake manages a catalog via Horizon, Google Cloud services—BigQuery, Managed Service for Apache Spark, and other Iceberg-compatible engines—reach those tables through Lakehouse catalog federation, connecting to Horizon's IRC endpoint to read and write directly.

For AI readiness, the architecture addresses three gaps. First, programmatic access: Snowflake exposes Model Context Protocol (MCP) servers allowing MCP-compatible clients—Gemini Enterprise, code assistants, and custom agent frameworks—to discover, query, and reason over lakehouse data through a universal interface. BigQuery also offers an MCP server accessible to first-party and third-party AI agents. Both also support REST APIs for applications needing direct programmatic control. Second, semantic grounding: Semantic View Autopilot, part of Horizon Context, automatically discovers and maintains semantic models by learning from query patterns, user behavior, table relationships, and BI tool integrations. Snowflake's Semantic Studio (in private preview) allows teams to author and refine models in natural language with a visual and YAML editor and Git integration, so semantic definitions can be version-controlled. Google Cloud's Universal Semantic Layer leverages Looker's support for in-database analytics models with new integration with BigQuery Graph and Snowflake Semantic Views. Third, contextual intelligence: Snowflake Horizon Context applies table and column descriptions in business terms, data quality signals (freshness, completeness, caveats), usage patterns, and domain relationships. Google Cloud Lakehouse provides trusted context through Knowledge Catalog, delivering Gemini-powered insights for lakehouse tables. This context is governed the same way as the data itself, so agents and users only receive information they are authorized to see.

With this architecture in place, Snowflake CoWork is positioned as a personal work agent for knowledge workers, and Cortex Agents and Gemini Enterprise agents can operate across structured and unstructured data with high accuracy and governance. The result is a fully open, interoperable data platform where all engines read and write directly from the customer's Cloud Storage Bucket with no ETL between systems, each team picks the best tool, and the catalog layer enforces unified governance across all of them.

The announcement reflects a fundamental shift in how enterprises approach multi-engine analytics. For years, organizations faced a binary choice: replicate data across systems and accept redundancy and drift, or lock teams into a single engine and lose the ability to pick specialized tools. The Apache Iceberg standard, combined with standardized catalog protocols, eliminates that trade-off by keeping data in one place and bringing compute to it—a principle known as Data Locality.

The catalog layer is the governance linchpin. Without a shared authority managing table metadata, access policies, and concurrent writes, open formats alone create ungoverned data. Snowflake's Horizon Catalog and Google Cloud's Lakehouse Runtime Catalog each provide managed, serverless catalog services that expose their tables via the Iceberg REST Catalog (IRC) standard, allowing engines built by competitors to access each other's tables with fine-grained, time-limited credentials. This two-way federation is the technical enabler of what the industry calls an "open lakehouse."

For AI specifically, the architecture bridges three critical gaps: programmatic access (via Model Context Protocol and REST APIs so agents can query data like human analysts), semantic grounding (embedding business logic in semantic models so AI inherits correct definitions), and contextual intelligence (layering table descriptions, data quality signals, and usage patterns so agents interpret data correctly). Snowflake's Semantic View Autopilot and Google Cloud's Knowledge Catalog automate much of this work, reducing the manual overhead of maintaining semantic models at scale. The result is not just interoperability for its own sake, but a platform designed from the ground up to let AI agents operate with high accuracy, governance, and trust.