Databricks expands billable AI compute

The real win is that Databricks turns outside model usage into inside platform spend. Instead of exporting tables to a separate AI stack, a customer can call OpenAI, Gemini, Anthropic, or Databricks hosted models through Databricks serving and governance layers, with Unity Catalog permissions, logging, and cost controls still attached. That lets Databricks monetize more inference, agent, and coding workflows while keeping the lakehouse as the system where enterprise data stays and AI work happens.1 2 3 4 5

• This changes what counts as billable compute. Databricks is no longer only charging for Spark jobs, SQL queries, and model training. It can meter external model calls, agent orchestration, retrieval, tool use, and governed serving endpoints, which pushes revenue up the stack from data processing into application level AI work.1 2 5
• The product advantage is operational, not just technical. A company can keep customer records, documents, permissions, and lineage in Unity Catalog, then let an agent query that data through Agent Bricks or model serving without copying it into a separate vector database or custom middleware layer first. That removes integration work and preserves security rules.3 4 5 6
• This is also how Databricks competes with Snowflake and point tools. Snowflake is building a similar keep the data in place AI story with Cortex, but Databricks is broader in combining external frontier models, MLflow, MCP based tools, and agent deployment inside one control plane. Point solutions may serve a model, but they usually do not own the governed enterprise data path.6 7 8

Going forward, more of Databricks growth should come from wrapping frontier models with governance and workflow products, not just from raw data infrastructure. As agents touch more enterprise systems and more employee workflows, the platform that keeps data, permissions, tracing, and inference billing in one place will capture a larger share of AI spend.2 3 5 8