Databricks wins AI‑data premium

The fight here is not really “warehouse versus warehouse” anymore. It is about who owns the layer where companies build AI systems on top of their data. Databricks spent the last two years pushing hard into that layer — training, model ops, vector search, agents, operational databases — and the market has started treating it like AI infrastructure, not just analytics software. Snowflake is still growing and still important, but the gap in narrative has gotten real. ### What changed? The biggest signal is simple — Databricks stopped being valued like a data-tools company. In December 2024 it announced a Series J round at a $62 billion valuation. By August 2025 it said a new round would value it above $100 billion. Then in February 2026 it said it had crossed a $5.4 billion revenue run-rate with growth above 65% year over year. That is not normal “mature data platform” pricing. That is an AI premium. ### Why does Databricks get that premium? Because Databricks has been selling the idea that your data stack and your AI stack should be the same stack. That pitch got much stronger after the MosaicML acquisition, which gave it a native story for model training and fine-tuning, not just storing data for someone else’s models. Since then it has layered on Mosaic AI, MLflow 3, vector search, serverless GPU features, Agent Bricks, and Lakebase for operational AI apps. (databricks.com) Basically, it keeps moving one step closer to where the actual AI application gets built. ### Where does Snowflake sit in that picture? Snowflake has not stood still. Its pitch is now “AI Data Cloud,” and it has been adding Cortex features, coding agents, and model access through partners like Anthropic, Google Cloud, and OpenAI. Financially, it looks healthy — fourth-quarter fiscal 2026 revenue was $1.28 billion, product revenue was $1.23 billion, and remaining performance obligations reached $9.77 billion. So this is not a story about Snowflake failing. (databricks.com) It is a story about Snowflake still being seen first as the governed data layer, while Databricks is seen as the place where the AI pipeline actually runs. ### Why does that distinction matter so much? Because AI budgets are not being assigned like old BI budgets. Companies are spending on systems that connect raw data, model training, retrieval, evaluation, deployment, and monitoring in one workflow. If a platform sits in the middle of that loop, it captures more spend and gets framed as strategic infrastructure. If it mainly stores and serves data cleanly — valuable, but narrower — investors tend to give it a lower multiple. (snowflake.com) Think of it like the difference between owning the factory floor and owning the warehouse beside it. ### Is this just investor hype? Not entirely. The revenue numbers support part of the story. Databricks said in September 2025 that it had exceeded a $1 billion AI revenue run-rate inside a broader $4 billion company run-rate, then accelerated again by February 2026. That suggests customers are buying the AI layer, not just the legacy lakehouse pitch. Snowflake’s numbers are also strong, but its disclosures still read more like enterprise software expansion than AI platform breakout. ### Could Snowflake close the gap? Yes — but the catch is product perception changes slowly. Snowflake has the enterprise relationships, the governance reputation, and a lot of customer trust. If Cortex and its app-building tools become central to production AI workloads, the market can rerate it. But right now Databricks has the cleaner story: data in, models built, agents deployed, all in one place. That story is easier to reward. ### What is the real takeaway? (databricks.com) The market is telling you where it thinks AI value will concentrate. Not in the database alone, and not in the model alone, but in the platform that binds data, compute, and model operations into one system. Right now, Databricks looks closer to that center of gravity.