Databricks and Snowflake Are Fighting Over Who Hosts Your AI Agents, Not Just Your Data
At Data + AI Summit 2026, held in San Francisco June 15 through 18 with more than 30,000 attendees, Databricks reported that over 100,000 agents have now been built on its Mosaic AI platform, processing what it describes as more than a quadrillion tokens a year. The company used the summit to expand Agent Bricks, its agent building product, and introduce a new tool called Genie One.
Snowflake spent the same window expanding Cortex Code, its AI coding agent, adding support for working across Snowflake itself, dbt, Airflow, Databricks, AWS Glue, and Postgres. Snowflake's own framing for the expanded product is explicit: it wants Cortex Code to be the "control plane for the agentic enterprise."
Same infrastructure, different bet
Both companies built their businesses as the place enterprise data lives, Databricks around its lakehouse architecture, Snowflake around its cloud data warehouse. The fight now playing out is not about who stores the data better. It is about who becomes the default place an enterprise's AI agents actually run, query, and act from. Databricks is building that layer inward, agents that live natively inside Mosaic AI and operate on data already sitting in the lakehouse. Snowflake is building it outward, a coding agent explicitly designed to reach across other platforms, including Databricks itself, rather than assuming all the relevant data already sits inside Snowflake.
Why the numbers matter
Databricks' 100,000-plus agent figure and quadrillion token count are the company's own reported usage numbers, but the scale they describe is not disputable in kind, only in exact size: agent workloads on that platform have moved well past pilot territory into something running continuously across a large customer base. Snowflake naming Databricks explicitly as one of the platforms Cortex Code supports is itself a signal. It is easier to compete for a customer's daily agent workflow by meeting that customer wherever their data already lives than by trying to pull all of it into your own warehouse first.
What this means for a buyer
An enterprise choosing between the two is no longer just choosing a data platform. It is choosing which vendor's agent tooling becomes the operational layer developers build against every day, which has real switching cost implications down the line. Databricks' bet rewards teams already committed to the lakehouse and willing to build agents inside that ecosystem. Snowflake's bet rewards teams whose data is scattered across multiple platforms and who want one coding agent that can reach into all of them without a migration first. Neither is obviously the safer choice yet. Both companies are still in the phase of racing to make their platform the habit, not the exception.