dbt Summit 2026 Demo Shows Cross-Platform dbt Mesh Built on Apache Iceberg

At a breakout session during dbt Summit 2026, Jack Lowery, lead sales engineer specialist for data lakes and transformations at Fivetran, and Anna Lee, software engineer at dbt Labs, walked attendees through what Open Data Infrastructure looked like in practice, built around Apache Iceberg and a live, cross-platform dbt Mesh demo.

Open Data Infrastructure, or ODI, is the idea that data should live in an open format any compute engine can read, rather than being locked inside one warehouse's proprietary storage. Lee explained the technical backbone: a new catalogs.yml v2 spec that decoupled dbt's Iceberg configuration from the older, Snowflake-specific concept of a catalog integration and write integration. That change meant a dbt project could define where its Iceberg tables lived independently of which warehouse did the writing.

The session's centerpiece was a live demo of a cross-platform dbt Mesh spanning three teams. A platform team on Snowflake wrote to a catalog Lowery and Lee referred to as Horizon, a finance team on Databricks wrote to a catalog called Unity, and a marketing team on Snowflake read from both, using catalog federation and catalog-linked databases to make tables written by one engine visible to another without copying the underlying data. That was the practical promise of ODI: three teams, two warehouses, one shared set of Iceberg tables.

Lowery also introduced two services built around this model. Managed Data Lake writes Iceberg or Delta Lake metadata directly into a customer's own cloud storage, using the open source Polaris catalog rather than a proprietary one. And Lake Compute, then in private beta and built on DuckDB, lets specific dbt models run on a different, often cost-optimized compute engine than the rest of a project, configured through a simple adapter setting rather than a platform migration. Lowery demoed Lake Compute live, routing a single model to run on DuckDB while the rest of the project kept running on Snowflake.

The through-line of the talk was that open table formats and shared catalogs are what make dbt Mesh genuinely cross-platform, rather than cross-platform in name only. Teams no longer need to standardize on one warehouse to share governed data with each other. They need a shared catalog and a table format every engine involved agrees to speak.

Source: dbt Labs Blog