Snowflake pivots to execution

Snowflake spent years selling faster queries and cleaner dashboards, and now its chief executive, Sridhar Ramaswamy, is telling people the next fight is about software that can actually do work. In a TechCrunch interview published on April 9, 2026, he said Snowflake is moving from a data warehouse into an artificial intelligence and applications platform built for agents. (techcrunch.com) That shift sounds abstract until you look at what Snowflake shipped on March 4, 2026. The company put support for Apache Iceberg version 3 into public preview so customers can read and write newer open-format tables inside Snowflake instead of treating outside data like a second-class citizen. (snowflake.com) Apache Iceberg is a table format, which is basically the filing system that tells software where data lives and what changed. Version 3 adds row lineage, so a company can track which exact record changed in a change-data-capture pipeline instead of reprocessing an entire table like a store recounting every item after one sale. (snowflake.com) Snowflake says that Iceberg version 3 support works for Snowflake-managed tables and externally managed Iceberg tables. Its March 4 release notes say the feature reaches across ingestion, transformation, analytics, machine learning, artificial intelligence, disaster recovery, and external catalog integrations. (docs.snowflake.com) The second part of the move is governance portability, which is less flashy and more important. In an April 8, 2026 report, Snowflake said it is working on policy exchange standards, governance federation, and read-restriction application programming interfaces so security rules can travel with data across systems instead of stopping at Snowflake’s border. (siliconangle.com) That solves a problem every big company now has: the data sits in one place, the model runs in another place, and the action happens in a third place. If an autonomous agent is going to approve a refund, reroute inventory, or trigger a customer email, the permission rules have to survive that trip. (siliconangle.com) Snowflake is making this push while investors still watch its consumption model closely. In its fiscal fourth-quarter 2026 results, reported on February 25, 2026, product revenue grew 30% to $1.23 billion, net revenue retention was 125%, and remaining performance obligations reached $9.77 billion, which means the company is growing but still has to prove customers will keep expanding usage. (investors.snowflake.com) That pressure helps explain why “store data here” is no longer enough. When customers can keep data in open formats like Iceberg and connect multiple engines to it, the platform that wins is the one that can govern the data, run the models, and let the software take action without breaking compliance. (snowflake.com) (siliconangle.com) The rivalry underneath this is also changing shape. Databricks has spent years pushing lakehouse architecture and open formats, and even its own documentation now shows how to connect Unity Catalog and Iceberg environments with Snowflake, which tells you the market has moved from closed stacks to negotiated interoperability. (community.databricks.com) So the pitch Snowflake wants customers to hear in April 2026 is not “put your tables in our warehouse.” It is “keep your data open, keep your rules attached, and let our platform help agents execute tasks on top of it,” which is a much bigger promise and a much riskier one if the governance fails. (techcrunch.com) (siliconangle.com)