Snowflake's Lakehouse Aims for Data Agency

AI's relentless march is forcing a reckoning with existing data architectures. When teams can't access data directly, they resort to copying it, leading to sprawl, fragmented governance, and stale insights. Snowflake's new Interoperable Lakehouse, now generally available, seeks to change that.

StartupHub data

Companies working on this

Profiles of the companies named in this story, with founding year, headquarters, and a short description from our database.

Google Cloud

A suite of cloud computing services offered by Google, providing infrastructure, data analytics, and machine learning.

Founded

2008

Location

Mountain View, United States

SAP

Global leader in enterprise software, specializing in ERP, cloud solutions, and business intelligence.

Founded

2026

Location

Germany

Snowflake

A cloud-based data platform enabling data warehousing, data lakes, data engineering, and data sharing.

Founded

2012

Location

Bozeman, Montana, United States

Valuation

$12.4B

Built on Apache Iceberg, Apache Polaris, and Open Semantic Interchange (OSI), the platform offers a blueprint for managing a single, governed copy of data regardless of its location. The goal is to grant 'agency over data' back to organizations, cutting costs and providing a reliable basis for AI.

Act on Data In Place

Central to this is the ability to act on data where it resides, eliminating the need for costly data copying and movement. Snowflake's support for Apache Iceberg v3 is now production-ready, offering enhanced capabilities for semi-structured data, row-level deletes, and high-frequency time series. This marks a significant step forward for interoperability, as detailed in Iceberg v3 Ushers In New Data Era.

To simplify management, Snowflake Storage for Apache Iceberg tables provides a fully managed experience for AWS and Azure, with Google Cloud support coming soon. This feature, part of Snowflake Simplifies Iceberg Storage, allows data to be governed through Horizon Catalog and accessed by any compatible engine.

Parquet Direct, in private preview, enables querying existing Parquet files with Iceberg-class performance. Zero-copy integrations bring critical business data from systems like SAP and Salesforce into Snowflake without ETL pipelines, preserving semantic context.

Unified Meaning and Governance

Connecting systems is only half the battle; ensuring data has consistent meaning across the enterprise is crucial. Horizon Context acts as a central layer for business definitions, linking scattered definitions across databases, data lakes, and BI tools. This ensures all teams and AI agents operate from a single source of truth.

Features like Semantic Studio, an AI-assisted IDE, allow teams to define shared business logic without deep SQL expertise. Semantic View Autopilot automatically generates and refines semantic views based on query patterns.

Universal governance is another cornerstone. Horizon Catalog, based on Apache Polaris, now extends governance to all Iceberg tables, not just those managed by Snowflake. This unified approach means policies are set once and honored across engines, eliminating the complexity of multi-catalog environments.

External engines like Spark, Trino, and PyIceberg can now read and write to the same governed data copy as Snowflake users. Fine-grained access controls, including row-level policies and dynamic data masking, follow data regardless of where it's queried.

Enterprise-Ready Operations

Snowflake is also addressing the operational burden of managing lakehouse architectures. Comprehensive auditing in Access History logs all external engine operations within Snowflake, providing a single, auditable record.

Iceberg Health Insights in Snowsight offers a connected operational view of externally managed Iceberg tables, surfacing freshness and refresh issues proactively. Managed Iceberg replication will provide resilience against outages.

These advancements aim to reduce the integration projects typically required to make lakehouse architectures production-ready.