Snowflake Summit 2026: Context, custom model training, Iceberg V3

Snowflake is doubling down on data interoperability via Iceberg, creating a context layer for its Horizon Catalog, adding data streaming to Snowflake CoCo, formerly Cortex Code, and enabling custom model training via Snowflake CoWork, formerly Snowflake Intelligence, to reach business users and teams.I

The announcements landed at Snowflake Summit 2026, a deal with AWS, the acquisition of Natoma and a strong quarter. Snowflake's announcements include a bit of rebranding, your typical dose of private previews and a strategic bet that the interoperability will be a major selling point for customers delivering data and context to AI agents.

• Snowflake expands AWS partnership, acquires Natoma, delivers strong Q1
• Snowflake rolls out Snowflake Intelligence, Cortex Code updates
• Snowflake launches Project SnowWork, eyes AI workflows
• Snowflake: Cortex Code, Semantic View Autopilot GA
• Snowflake Marketplace: The Future of Data Monetization and Enterprise AI

First, the rebrands. Snowflake Intelligence is now Snowflake CoWork and Snowflake Cortex Code is CoCo, which is what the company has called the platform internally. Speaking on Snowflake's first quarter earnings call where the company raised its fiscal 2027 growth target to 31% from 27%, CEO Sridhar Ramaswamy said:

"AI is accelerating consumption in our core platform as customers migrate workloads to Snowflake faster in order to access the data, context and governance needed to power AI securely and at scale. Snowflake Intelligence (CoWork) and CoCo are seeing the fastest adoption of any new products in our history, opening new opportunities for growth as the first major product surfaces of the agentic control plane.

Adoption of these AI products is increasing core platform consumption as customers move from questions to answers from prompts to pipelines and from ideas to production workflows."

Here's a look at what Snowflake announced in order of what's closer to generally available. The goal for Snowflake is to show how out-of-the-box agents on Snowflake CoCo and CoWork perform better than other AI agents.

The interoperability story. Snowflake said it connecting Apache Iceberg v3 to Snowflake Storage for Apache Iceberg Tables to work with data inside and outside of the company's platform. The interoperability push is powered by Snowflake Horizon Catalog and designed to position the company's governance and management tools as one control plane for data.

• Constellation ShortList™ Automated Cloud Data Warehouse Services

Databricks recently announced that Apache Iceberg V3 support is generally available.

Key items include:

• Interoperability between Apache Iceberg v3 and Snowflake Storage for Apache Iceberg Tables. Support for Apache Iceberg v3 is generally available.
• Horizon Catalog, powered by Apache Polaris, will enable bi-directional read and write access.
• Support for Iceberg REST Scan Plan API.
• The strategy for Snowflake is to combine open connectivity and intelligent querying to create one data foundation to access multiple data engines.
• The company announced zero-copy integrations with Aveva and IBM to go along with partnerships with SAP, Salesforce and Workday.

Snowflake Horizon Catalog gets Horizon Context enhancements. Snowflake said it is adding a context layer for AI and business intelligence to deliver one version of the truth. Horizon Context brings together context across the enterprise data estate, maintains context via connected views and enables Snowflake CoCo to get a trusted answer.

The company also added Semantic Studio to enable teams to define shared business logic without SQL expertise. Semantic View Autopilot automatically refines semantic views for the context layer. In addition, Snowflake said Horizon Catalog will get a feature called Adaptive Compute, which will automatically create the optimal mix of compute and software resources for AI workloads.

Snowflake also said it is adding new security features including:

• Agent Identify, which provides every agent with a verified identity.
• New observability features in Snowflake Trust Center.
• Machine learning detection of prompt injection flaws and other vulnerabilities.

The Snowflake Horizon Catalog and Horizon Context additions are either in public preview, coming soon to general availability or generally available.

Snowflake CoCo gets data streaming. The company launched Snowflake Datastream, a managed streaming service for Apache Kafka that combines real-time data and AI in one platform. Snowflake Datastream is in private preview.

In addition, Snowflake CoCo will integrate with desktop, mobile and Slack interfaces to provide unified workflows.

Key items include:

• Snowflake said Snowflake CoCo is now available as a native desktop app that can extend to environments where work is completed.
• Snowflake said there are extensions and plug-ins to connect to multiple environments.
• Snowflake CoCo can execute tasks autonomously without prompting.
• Snowflake CoCo includes skills and a skill catalog for common data engineering and AI workflows.
• Integrations between Snowflake CoCo and tools that deploy apps.

Snowflake CoWork adds customization tools. Snowflake said Snowflake CoWork enhancements will give knowledge workers more tools and a unified experience via an AI agent. The headliner of Snowflake CoWork is Cortex Training, which gives enterprises the ability to train models at scale.

Cortex Training enables custom training for foundational models on fully managed GPUs. Cortex Training is designed to enable enterprises to customize open-weight LLMs.

Other Snowflake CoWork additions include:

• Cortex Sense, a new feature that brings together data, business definitions and operational knowledge for AI agents. Cortex Sense works in CoWork and CoCo. Cortex Sense also features a set of prebuilt plugins and MCP connectors.
• Artifacts include publishable dashboards with governed view of trusted data that can be explored via conversations.

Constellation Research's view

Constellation Research analyst Mike Ni said:

"Snowflake is preparing for a world where it no longer needs to own the data format to own the customer relationship.

Iceberg V3 is largely table stakes. The market has already moved there, and while V3 improves performance, the bigger shift is read/write interoperability that opens the door to more operational and AI-driven workloads.

You see the Summit 2026 announcements are not data formats, but around meaning, trust, permissions, context, and governance. The announcements play to the Natoma acquisition, as well as their previous acquisitions like Select Star."