Databricks’ Ali Ghodsi Never Wanted to Be CEO. Now He’s Among the Best

Summary · 5 min read instead of 1 h 15 m

Ali Ghodsi is the founder and CEO of Databricks. He became CEO in 2015, when the company had a successful open source project in Apache Spark but about $1.5 million in revenue. Before that he was a postdoc at Berkeley and had been a professor in Sweden.

Ghodsi says he became CEO as a trial, with the board interviewing other candidates, and that he runs the company by finding its single biggest bottleneck and focusing on it for two to three years. He describes how he moved Databricks from product-led growth to enterprise sales, hired executives ahead of need, and beat Snowflake by targeting its weaknesses instead of copying it. He argues that the Lakehouse category took years of conviction against internal and outside resistance. He says conflict aversion is the worst trait in a CEO. He is skeptical that AI justifies much flatter org charts, and says Databricks will go public once markets settle.

Minute by minute

1. 1:14

   The board tried CEO candidates first. Ghodsi says the board interviewed other CEOs in 2015, when Databricks had $1.5 million in revenue despite a successful Spark project. He had applied for a faculty job at Berkeley. He thinks Ben Horowitz favored founders and gave him a trial: no CEO salary at first, then a bigger equity package about a year in.

2. 4:25

   Always pick the harder option. Ghodsi traces a pattern in his choices: going to university instead of game programming, taking a PhD over an easy professorship, and accepting a Berkeley postdoc. He took the CEO job on the same logic, since he knew nothing about running a business. He calls challenging himself his philosophy.

3. 8:38

   Focus everything on one bottleneck. Ghodsi says first-time CEOs should build their own playbook from first principles, as Bruce Lee did with martial arts. Companies usually fail on one giant bottleneck, and he says to give it orders of magnitude more attention. His was turning Spark into commercial success. Such efforts run one to three years, and Lakehouse took multiple years.

4. 12:40

   PLG failed, so Databricks hired enterprise sellers. Ghodsi first replaced the executive team and handed engineering to a cofounder. The company's zero-touch experiment flatlined revenue for two quarters. He found that top sellers were not technical and that a PhD was a disadvantage. Great sellers were professionally aggressive, had strong emotional intelligence and could map a customer's power base.

5. 19:11

   What made Ron Gabisco the right sales leader. Ghodsi says sales leader Ron Gabisco was a classic enterprise seller with engineering degrees who could speak the engineers' language. He had helped build a company from zero to $50 million ARR and had stayed 11 years at his previous two companies. Ghodsi credits him with the sales engine that grew from $1 million to $7 billion ARR.

6. 23:21

   Hire executives before you need them. Ghodsi says that by the time you know you need an executive, it is too late. He hired ahead of need to run searches of six months to a year. He was very picky, said no to people who later succeeded, and counts about two clear hiring mistakes. He recommends many back-door references with past managers; he says only 10 to 20 percent of front-door references are truthful.

7. 27:07

   Beat Snowflake by studying, not copying. Ghodsi says Databricks went after three Snowflake weaknesses: a proprietary format, weak AI support and high cost. Sellers ran an account-by-account playbook of coexisting, pulling out machine learning workloads and moving data to open formats. He cites a professor's advice to watch the competition but never follow it.

8. 33:22

   Lakehouse took years of conviction. Ghodsi says seasoned employees, strategy firms and some sellers opposed the Lakehouse name, and people ridiculed it online. He mobilized the whole company behind it and used compensation plans to direct sellers. He says Databricks made big bets around 2018-19, when ARR was in the hundreds of millions, to avoid becoming a one-product company. Public investors ignored the vision until results showed.

9. 41:03

   Conflict aversion disqualifies a CEO. Asked about his reputation as a killer, Ghodsi attributes his aggression partly to growing up as an immigrant in Sweden. He says conflict aversion is the worst trait for a CEO. Everyone pushes a CEO for approvals, and a conflict-averse CEO says yes or dodges. Clear, black-and-white messages keep a 12,000-person company from canceling its own efforts.

10. 44:05

    Org charts and AI need separating. Ghodsi says structure became necessary around 200 to 250 employees, when the CEO could no longer know everything. He agrees with Jack Dorsey on AI context but separates the org chart from how information flows. He says enterprises mostly use AI as chatbots and for coding, and need an ontology of company context; Databricks' Genie product uses one. He calls 25-person manager spans plus coding a mistake.

11. 52:50

    Two days for a connector exposed the gap. Ghodsi says he built a connector in two days that teams took three quarters to build. The team found the gap was in requirements, system setup and end-of-cycle testing. After reengineering, the team went from seven and a half months per connector to seven connectors in one quarter. He concludes that absorbing AI will take humanity at least a decade.

12. 57:17

    His time goes to one bottleneck. Ghodsi splits his time among the main bottleneck, routine CEO work and product, and tries to keep calendar blocks free. His staff meets Monday for an hour to an hour and a half on a few annual goals, and for 30 minutes on Wednesday and Friday. He says teams need to meet in person so department loyalties don't override company priorities.

13. 1:04:47

    Databricks will go public, but not yet. Ghodsi says Databricks would be worth more as a public company today, citing $20 billion of demand in a $5 billion raise. Running employee tenders across about 100 countries is already like operating a private Nasdaq. He wants to wait until markets settle, because he says public markets handle the AI transition badly. Databricks is free cash flow breakeven.

14. 1:08:11

    Get over conflict aversion. Ghodsi says developing a bullshit detector starts with not lying to yourself about what is wrong. He compares conflict to going to the gym: it is unpleasant for everyone, yet people push themselves. He says anyone who wants to be CEO must get over conflict aversion and face disputes head on.

Published Sep 17, 2026.