The fastest revenue engine in SaaS history: $5.4B run rate in 10 Years - Ron Gabrisko [Databricks]

Databricks built a $5.4B engine by giving software away

Databricks built a business by giving away the core of its technology and charging for consumption of the managed platform on top. That works because open source keeps handing the company new products to sell, and the buyers are engineers who want help rather than a relationship. AI workloads are the next thing pulling them through the same door.

Takeaways

• Databricks scaled from $1M to $5.4B ARR. Ron Gabrisko joined in 2016, when Databricks had seven PhD founders, a couple of salespeople and under $1M in annual recurring revenue. Ten years later it runs at a $5.4B run rate, growing 65% year on year at a $134B valuation.
• Technical sales beats relationship selling. Gabrisko says Databricks built a very technical sales team, and that reps gained trust by helping users of the free community edition with technical problems instead of pitching them. Relationship selling does not work with technical products.
• An open source core keeps retention above 130%. Databricks kept moving features into open source and layering new projects into one platform, Spark, Delta, MLflow, Iceberg through the Tabular acquisition, and Mosaic AI. Customers start with one use case, expand, and retire Hadoop and older warehouses, driving net retention past 130%.
• Usage pricing beats value pricing at renewal. Databricks prices on consumption and publishes its rates rather than charging for the value customers get, the way Palantir does. Gabrisko says value-based pricing falls apart at renewal, when customers ask what they consumed.
• Databricks concentrates on 7,000 accounts. Databricks segmented the market using purchased datasets on cloud and product spend and found that 7,000 companies make up about 90% of its total addressable market. It spends 30% or more of its resources on new business acquisition.
• AI is already $1.4B of booked revenue. Databricks partners with Anthropic, OpenAI, Gemini and GCP for model optionality, runs as a first party on Azure and on AWS and GCP, and has booked $1.4B in AI revenue. Gabrisko says customers keep their data in open storage.
• Agents will multiply application traffic a hundredfold. Gabrisko expects millions of agents to hit applications the way humans do today, generating 10 to 100 times the traffic. He calls Genie, which lets anyone query data in English, his most exciting product, and Databricks bought Neon to build Lakebase for agent workloads.

How Databricks built its first sales motion

1. Ben Horowitz told Gabrisko that Databricks, whose founders had invented Spark and given it away, had more upside than anything in his portfolio.