dbt Labs and Fivetran used their dbt Summit 2026 session to argue that data lakes built on open table formats are the next step for the data stack, and that teams can get there in small moves rather than one big migration. Russell Christopher, senior director of product strategy at dbt Labs, and Fivetran's Casey Karst delivered the talk dressed as ducks.

Christopher described the problem with the modern data stack: it appears to separate storage from compute, but most teams write data into a proprietary format tied to one compute engine, which raises costs, limits which engine can do which job, and makes later migrations painful. He called his own past migrations three-to-nine-month "death marches."

The proposed alternative puts Apache Iceberg at the center, exposed through an Iceberg REST catalog (dbt Labs and Fivetran support Apache Polaris), so engines such as Snowflake, Databricks and BigQuery can read and write the same data without copying it. Karst demoed Fivetran's managed lake setup, which asks only for storage provider, bucket, path and how many Iceberg snapshots to retain, with defaults for catalog integration with AWS Glue or Databricks Unity Catalog. He also showed a "Move" capability that migrates an existing Snowflake-native table to Iceberg in place, keeping the same table namespace so dependent dbt models and BI dashboards keep working; the live demo finished in a couple of minutes. Because Fivetran had already been writing and re-loading that data through Snowflake, the move cut roughly 15 to 20 percent from the Snowflake compute bill for that workload at no extra cost, since pricing stayed on monthly active rows.

Christopher then extended the idea to transformation with a concept called Lake Compute. Since a dbt project's models don't all need the same compute, some being simple filters, casts or renames and others heavy joins or warehouse-native AI functions, Lake Compute lets teams move specific low-complexity models onto cheaper lake-native compute, model by model, without restructuring the DAG or project. He shared directional figures suggesting ingestion and transformation together account for roughly 75 percent of a typical data stack's compute spend, and said the approach targets that spend while the gold layer serving analysts and BI tools stays put.

The session closed by separating the reference architecture, an open lake on open table formats and open catalog interfaces that preserves engine choice, from the build-versus-buy decision on top: teams comfortable with open source can assemble their own lake stack, while others can use Fivetran's managed lake for ingestion and Lake Compute for targeted transformation savings without giving up open standards. The speakers pointed attendees to a follow-up roundtable on Open Data Infrastructure later that day.

Source: dbt Labs Blog