Databricks Shifts From Warehouse To Platform
Databricks at $4.8B ARR
The key split is that Databricks is no longer just selling a faster warehouse, it is turning its data stack into a broader software surface where customers can build, train, and run more of their workflow in one place. That shows up in mix. Databricks already gets roughly 40% of revenue from newer warehousing and AI lines, while Snowflake is still overwhelmingly monetized by core warehouse consumption.
-
Databricks started from Spark, notebooks, and ML pipelines, so its core buyer was often the data engineer or ML team running many kinds of compute jobs. Snowflake started as a cleaner cloud warehouse for SQL analytics, so most of its dollars still come from storage and query spend tied to BI and analytics workloads.
-
That difference matters because adjacent products attach differently. Databricks SQL and Mosaic AI extend an existing engineering workflow, so they can scale into $1B lines. Snowflake has added Cortex, Snowpark, and Streamlit, but official reporting still describes product revenue mainly as one integrated compute, storage, and transfer offering, with warehouse usage doing most of the monetization work.
-
The next step is moving above infrastructure into apps and transactions. Lakebase adds Postgres for low latency application reads and writes, Databricks Apps gives a way to ship front ends on top of lakehouse data, and Agent Bricks packages agent building into the platform. That creates a path to earn revenue from the software sitting on the data, not just from processing the data itself.
Going forward, the competitive line gets clearer. Snowflake remains strongest where the job is centralized analytics on warehouse data. Databricks is pushing toward being the system where raw data lands, models are trained, agents run, apps are served, and operational databases connect back into the same stack. If that bundle keeps working, more enterprise data spend shifts upward into the application layer.