Data platforms push openness; Snowflake slides

Databricks just made a direct pitch at one of the software industry’s biggest fears: getting trapped in one vendor’s data format. On April 9, 2026, it put Apache Iceberg version 3 into public preview and said customers can use one governed copy of data across different engines instead of rewriting it every time they switch tools. (databricks.com) That landed one day before Snowflake shares were quoted around $132.24 after an 11.83% one-day drop on April 9, 2026. Market coverage tied the selloff to two things hitting at once: fear that artificial intelligence reshapes software economics, and lawsuits tied to older claims about Snowflake’s growth outlook. (google.com) (fxleaders.com) The fight here is about the plumbing under corporate data. A table format is the filing system that tells software where rows live, what changed, and which engine can read the data without breaking it. (databricks.com) Apache Iceberg is one of those filing systems, and it is designed as an open standard instead of a single company’s private format. Databricks says Iceberg version 3 adds deletion vectors, row-level lineage, and a Variant type so teams can update records faster, trace where rows came from, and handle messy semi-structured data without custom hacks. (databricks.com 1) (databricks.com 2) Deletion vectors are the part investors care about even if they never hear the name. They let software mark changed rows without rewriting whole Parquet files, which is like editing one line in a spreadsheet instead of printing the whole workbook again. (databricks.com) Databricks paired that with Unity Catalog, its control layer for permissions and governance. Its documentation says managed Iceberg tables, foreign Iceberg tables, and Iceberg client access through the Iceberg Representational State Transfer catalog application programming interface are all in public preview, which means outside tools can read and write while Unity Catalog still handles the rules. (docs.databricks.com 1) (docs.databricks.com 2) That is the pressure point for Snowflake. If customers can keep one copy of data in an open format and still share, govern, and query it across platforms, the cost of leaving any one platform drops. (databricks.com 1) (databricks.com 2) Snowflake is not standing still. Its investor page says it now markets itself as an “AI Data Cloud” and reported $9.77 billion in remaining performance obligations as of January 31, 2026, which shows customers are still committing large future spending even as the stock gets repriced. (investors.snowflake.com) But the legal overhang gives traders another reason to hit the sell button. Multiple shareholder notices say a federal securities class action targets statements made between June 27, 2023 and February 28, 2024, and the lead-plaintiff deadline is April 27, 2026. (fxleaders.com) (markets.financialcontent.com) So the market is pricing two separate questions at once. One is whether artificial intelligence shifts value away from storing data and toward whatever software can reason over it fastest, and the other is whether open formats like Apache Iceberg make the storage layer easier to swap out. (fxleaders.com) (databricks.com) Databricks is betting that “open” is now a product feature, not a concession. Snowflake is betting that customers will still pay for speed, tools, and ecosystem even if the raw data becomes more portable. (databricks.com) (investors.snowflake.com)