Ali GhodsiCo-Founder & CEO

The board was interviewing other people and I was hearing about it through the grapevine. I assumed I was not going to be picked. The company had great open source success with Apache Spark but only about 1.5 million in revenue. I was actually applying for a faculty job at Berkeley at the same time. I was facing the choice between becoming CEO, which I never wanted to do, or going back to my dream of being a professor. What made me choose CEO was a pattern I noticed in my own life. Every time I faced a choice between the comfortable known path and the genuinely challenging unknown, I had picked the challenge and it had opened my horizons each time. So I took the CEO role as the hardest thing I could possibly do.

Find the one major bottleneck and put an orders-of-magnitude disproportionate amount of attention on it. Not just a little more attention. A lot more. Almost to the point of going overboard. Because even with that level of focus, you probably will not fully unclog it. There is so much other stuff landing in your lap every day that you will be pulled in every direction. If you do not deliberately overallocate attention to that one thing, you will make no real progress on it. For me in 2015 and 2016, the bottleneck was simple. Open source success, check. Commercial success, zero. So everything went toward building the commercial engine.

I ran a mental experiment. I looked at the companies where enterprise sales was working best and asked myself, are the best salespeople the most technical people? The ones with PhDs? And the answer was clearly no. None of the top performers were technical. So I concluded not only is technical knowledge not an advantage in sales, it is probably a disadvantage. That told me we needed to completely change our hiring profile. I was looking for people who were professionally aggressive, who had exceptional emotional intelligence in high-pressure situations, and who could read and navigate the power base inside a large enterprise organization. Those are very different skills from what I had spent my career developing.

The first AE Ron hired, a guy named Dave, got us a meeting with a customer we could not get a meeting with. In that meeting the executive told us to get Dave out of the room and never bring him back. He had emailed the boss, the boss's boss, everyone in the building. After the meeting I confronted Dave and he said, Ron told me to get into the building. Nobody was responding. So I went through every door I could find. I got you the meeting. What are you complaining about? That was my education. The squeaky wheel gets the grease, and in enterprise sales, if you are not willing to be professionally aggressive, you do not get the meeting.

Ron was what they call in the industry a PTC or BMC-style seller. Classic meat-eating enterprise sales DNA. But he also had an engineering undergraduate degree and an engineering master's from Stanford GSB. That combination meant he could talk to our engineering founders as equals. We were not oil and water. He had also built a sales organization from zero to 50 million ARR at a startup called Cyclone, then seen the journey up to hundreds of millions at Axway. He knew how to build the car, not just drive someone else's car. And he had been at the same company for 11 years. That told me he had loyalty and staying power. If it got rough, he was not going to quit. That turned out to be crucial because great enterprise salespeople are very hard to manage.

We studied their weaknesses carefully. Three stood out. First, their stack was fully proprietary. The data sat in their format and customers were worried about lock-in. Second, their AI and machine learning story was weak. Third, they were expensive. We went after all three simultaneously. We positioned Databricks around the open lakehouse, meaning you own your data in open formats, you can do AI on it natively, and the total cost of ownership is about a third of the alternative. We also ran a coexistence strategy rather than a rip-and-replace message. We would go into accounts and identify the specific workloads suited to machine learning, pull those out, and move them to open formats. It was a very precise playbook executed account by account. That strategy took about four years to play out. You cannot do something like that overnight.

Internally, the seasoned people said do not do it. The word sounds ridiculous in an enterprise context. Strategy consultants ran surveys and told us nobody wanted it. They came back with category names like Unified Data Science Platform. Our sales reps said that made no sense either. Externally, when we launched it, people made fun of us online. Data rivers. Data rapids. Ha. But I held firm and made it a maniacal obsession across the whole company. If a press article about Databricks did not mention Lakehouse, that was a failure. If a customer win did not reference Lakehouse, that was a loss. Even if our conversion data showed that mentioning Lakehouse hurt ad performance, I made us use Lakehouse in the ads anyway. It took multiple years of that kind of religious commitment before the category took hold.

As a CEO, everyone is pushing you for something constantly. If you are conflict averse, you are going to say yes to things you should say no to, or worse, you will avoid the conversation entirely and leave people confused. The company then starts going in all directions at once instead of focusing all its energy on that one bottleneck you are trying to unclog. And I think conflict aversion and truth-seeking are fundamentally incompatible. If you want to see clearly what is not working in your company or your strategy, you have to face those gaps head on. You have to be aggressive about it. I tell people it is like going to the gym. It is hard for everyone. Nobody loves it. But if you want to be a CEO, you cannot say I am the kind of CEO who does not do conflict. That is like saying I am the kind of athlete who does not train.

I ask this question at every audience I speak to. How many of you think AI is smarter than most of the people around you most of the time? Ninety percent of hands go up. At Moscone Center with 32,000 people, almost everyone raised their hand. That is what I define as AGI. It may not be superintelligence but it is performing at or above the level of most human workers in most contexts. The disagreement about AGI being here is a definitional problem, not an empirical one. The real problem is that enterprises are not using these capabilities. They are still using AI as a chatbot or a coding assistant. The gap between what the models can do and how enterprises are actually deploying them is enormous. I believe it will take at least a decade for that gap to close.

The models are smart but they do not have the context of your business. They do not know what decisions are being made in your meetings, what your priorities are, who the key people are, what your data means. We call capturing all of that the ontology. Genie is our product that builds and maintains that ontology for you and feeds it continuously to the AI. At Databricks internally, I can ask Genie any question about any part of the business at any point in a meeting and get a real answer. We have built up our ontology over years. When I look at our customers using the same product, the ones who have not built up the ontology yet get a fraction of the value. The intelligence is not the bottleneck. The context is.

A lot of what managers do is fundamentally human work. Someone is unhappy. Career goals are unclear. There is a conflict between two team members. These things require presence and attention. If you have 25 direct reports and you are also expected to vibe code 80 percent of the day, you are going to have a lot of unhappy employees who do not know what is happening, whose manager they barely see. I support the philosophical direction that AI changes how information flows through an organization. I am not convinced yet that it should mean one manager for 25 people. Those two things can be true at the same time.

We will go when the waters are more calm. Right now the public markets are going through wild swings about what AI means for software valuations. One week SaaS is dead, the next week it is a buy. That kind of volatility is not good for a company in the middle of a major transformation. We also do not need the capital. We are free cash flow break even. Unlike companies like Anthropic or OpenAI that need to raise constantly, we do not. The reason to go public eventually is the complexity of running private liquidity programs across 100 countries for 15,000 current and former employees. At some point it is simpler to just be a public company. But that point is probably another year or so away.

I see it as three buckets. The first is the main bottleneck I am personally obsessed with. That is the thing only I will work on because everyone else is busy running their trains on time. I want to carve out as much time as possible for that. The second bucket is all the reactive CEO work that fills your calendar. I try to compress that as much as I can. The third bucket is being a product CEO. I spend significant time with our engineering and product teams staying deeply familiar with what we are building. If I have a day that is completely packed from eight in the morning to six at night, I tell my team that was a wasted day. I did nothing to move the needle. I just worked for my calendar.