Summary
On the All-In interview series, Databricks CEO Ali Ghodsi discusses how the company grew from a UC Berkeley AMPLab project into a business with a reported $7 billion run rate, growing about 80 percent year over year. He recounts starting Databricks in 2013, when there was little industry interest in what was then called machine learning, and why legacy infrastructure makes enterprise AI adoption so hard.
Ghodsi argues that today’s models already provide enough intelligence to transform the economy, and that most companies are still using AI in rudimentary ways, such as a glorified search engine. The hosts press him on running large numbers of agents, rising token costs versus revenue, whether GDP will move quickly or take decades, and the recursive self-improvement debate. His view is that even if AI writes 98 percent of code, automating the last 2 percent does not by itself produce runaway self-improvement.
The conversation includes a concrete demo of an internal AI analysis of employee survey data. Using an ontology of company metrics, the system separates correlation from causation and breaks results down by department, work that would otherwise take a data team a week. The interview also touches on Databricks’ new product launches and the question of when the company will go public.
📺 Source: All-In Podcast · Published November 17, 2026
🏷️ Format: Interview







