Databricks grew out of UC Berkeley's AMPLab around 2009, when cloud computing was still controversial and its founders bet on it from first principles.
Alan Tu of WCM Investment Management says Databricks' core value is data processing, which can consume 80% to 90% of analysts' time and is larger at enterprise scale.
Alan Tu says Databricks' AI-related revenue is about $1bn of more than $4bn in annual recurring revenue, making AI a durable tailwind that raises the priority of data engineering.
How Databricks turned open source Spark into a data platform
Databricks is a long-term bet that enterprises need one platform to turn raw data into decisions. It began as open source Spark research at Berkeley and became a paid business by making its core product better than the free version. Each step, from machine learning to warehousing to AI, widened the market it could serve, and the main risk now is executing across so many products.
Takeaways
Databricks turns messy enterprise data into usable data. Its core is data processing. Analysts can spend 80% to 90% of their time preparing data in a spreadsheet, and Reustle puts preparation at 95% of the work. In fraud detection, Databricks builds the pipelines and models, while a separate company application acts.
A better proprietary Spark solved open source monetisation. The seven founders came from Berkeley's AMPLab around 2009 and bet on cloud, data at scale and open source. Databricks sold a new proprietary Spark implementation with better performance and reliability. Tu says customers pay for core product quality, not add-on features.
Long-term product bets widened the addressable market. Spark led to MLflow, then Delta, then a SQL data warehouse on pace for $1bn in 2025. The Lakehouse label was mocked at first and is now an industry category. Making data usable expanded a market the old data lake category never built.
AI is a durable tailwind with real revenue. Net dollar expansion runs above 140% and annual recurring revenue exceeds $4bn, with about $1bn tied to AI. Tu argues AI raises the priority of data engineering, and Agent Bricks and LakeBase target agentic applications.
Snowflake and hyperscalers are both partners and rivals. Customers often process data in Databricks and store it in Snowflake, while each moves into the other's territory. Databricks avoided storage charges through open formats. Azure Databricks, its first Microsoft partnership, helped jumpstart monetisation. Reustle compares the cloud dynamic to Amazon and its delivery partners.
Usage pricing and private status fund heavy R&D. Databricks charges on compute used, plus governance features built on open source. It is free cash flow positive, which Tu calls a low bar. Staying private let it keep investing through 2022 while public peers could not.
Execution across many products is the main risk. Tu sees R&D execution as the biggest threat as Databricks expands into many products and AI tools. Its culture of long-term, first-principles thinking must hold. Both speakers call the timing of an IPO a trillion-dollar question.
Tap a name to go there · tap a line to see why they are linked