The Rundown - How Databricks Became One of the World's Most Valuable AI Companies

Episode Date: June 29, 2026

Databricks is one of the most valuable private AI companies in the world, but what exactly does it do and why are investors paying so much attention? In this episode, co founder Arsalan Tavakoli explains why enterprise AI is becoming the next major battleground, how companies are trying to rein in soaring AI costs, and why the winners may not be the companies building the biggest models. We also discuss the competitive race between OpenAI, Anthropic, Google, and open source AI, plus why Databricks believes its biggest opportunity is still ahead. Finally, Arsalan shares his thoughts on the company's eventual IPO and what needs to happen before it goes public.

Welcome back to the rundown interview edition. Today, I am talking to Ursula Tavacoli, one of the co-founders of DataBricks. Now, this is a little bit different from our usual guest because DataBriggs is a private company, meaning you can't buy their stock. But it's one of the most valuable private companies in the world, last valued at $134 billion. And Databricks is quietly powering the AI behind something like 70% of the Fortune 500 companies out there. In today's conversation with Arcelon, we get into what DataBriggs actually does, their new AI agent called Jeannie One, the absolutely insane AI bills that some companies are racking up and how Databricks is helping to solve that. And his thoughts on what roles the model companies like OpenAI, Anthropic, and Google will play in the future of AI. And of course, we also talk about IPO plans.

This was a really fun one. I think you guys are going to love it. So let's get into it. All right, guys, today we are talking to Arcelon, Tavacoli, the co-founder and senior vice president of field engineering at Databricks. Arcelon, welcome to the rundown. Thanks for having me. Appreciate it. I'm super excited for today's conversation.

You know, we mostly cover publicly traded companies on this podcast. So I imagine a lot of our audience might not be familiar with Databricks. So for the people that aren't familiar, can you give us a brief recap on what data brink? Bricks does. And imagine that I'm like a 62-year-old uncle at Thanksgiving and I'm in a slight food coma. So if you can explain to me what Data Bricks does, I'd really appreciate it. In a slight food coma, everybody's been there, right? Before, look, at a high level, you know, Databricks is a data and AI company. So what does that actually mean? A lot of the things that you are really excited about in the world that's happening of applications. For example, you know, Adidas is trying to think about how do they put new products out that

really understand what customers want, what they care about, how do they get it out? Or you're talking about 7-Eleven wants to know what are the main products that you actually want in there? How do I stock the latest things in my stores? Or EasyJet is trying to figure out what are the right routes to fly to get you to where do you want to go into Europe? What that takes is a bunch of frankly, unsexy, boring things to happen. There's like, how do I go grab all of that data? How do I pull it in? How do I figure out some intelligence and insights from it, and then how do I build an application that you as an end user? That's really hard. People stick in the Cragmire. Databricks is the leading platform that allows you to do all of that end-to-end to deliver those applications that consumers and

businesses really need. So it's doing a lot of the boring, unsexy stuff in the background that no one thinks about, but that help enterprises kind of run more efficiently and gather insight, things like that. Yeah, exactly, right? So for example, if you wanted to, you've got a podcast. you want to say, hey, about any company, I want to understand what are the best, you know, private companies that are doing X, Y, and D, who's been on my podcast that has done, you know, this? Or if I'm sitting at a pharmaceutical company, I want to be able to ask a question and say, what are the products that are savings people's lives? Like, how are our sales reps doing that? How do I, how do you give that capability? How do you make that happen, which is what people

want, especially in the AI era? You know, Databricks does everything from soup to nuts about bringing you the data, allowing you to build those AI applications. Of course, making it cost efficient and secure to do so. I think that's the holy grail that everybody's figuring out, how do we move faster? And that in Databricks is allowing tons and tons of customers over 20,000 globally to do that. That's incredible.

I mean, and you guys have been around for a long time with 2013, founded in 2013, and I was doing some background on you. Before that, you were a McKinsey consultant. So you're actually the perfect person to ask this question. Are consultants cooked? Because with like more and more enterprises, just integrating. AI into their systems and having access to all their data.

Like, what's going to happen to consultants? I mean, as a former consultant, what do you think about this? I always get worried when somebody gives me a preamble. I was like, nothing easy is going to come out of this afterway. Look, I think the reality is the same for consulting as it is for kind of any other job. There is a bunch of what folks are doing today that AI can help automated. And if you assume that, hey, what I want to do as an enterprise is the same thing I've been doing for the last 10 years and not change, not change. AI is going to give other people a leg up, right? But I think consulting as well,

now looks at it and says, well, look, tons of amazing data is coming out. And if we have something like a data, it can give me access to that data. It can give me access to analytics. Now, what are the newer capabilities that customers are asking for? How do we help advise them how to get there? So I think it's more about for organizations, how do you leverage the benefits of what AI has to open up an entirely new wing of things that you can do. Gotcha. I mean, like, a lot of consultants listen to this podcast. I have some consultant family members, and that's the number one thing they're always asking me. He's like, what's going to happen to this space? And I think about it myself, too, especially some of the publicly traded consulting companies have been not doing so well. So I want to talk more about Databricks again and some of the specific AI products that you guys have released. I saw that you guys are getting into a Gentic AI now released a tool called Jeannie One. It's kind of like an AI co-work. And that's like the hot thing these days, right? Like an AI co-work. So can you talk more about that? Like how?

How is it going to help someone that uses this tool on a Tuesday afternoon? Yeah, on a Tuesday, but not on a Wednesday. Look, here's the thing. What everybody got introduced through chat, GPT of great, we have these AI tools that we can ask questions. And in the consumer space, I think the models do a fantastic job. You're like, you know, show me like, how do I plan my next trip to Greece? Or you want to say, can you help me understand? who has the highest likelihood of winning the World Cup.

They're really good at that, right? But when you're sitting inside of an enterprise, you want to ask those similar types of questions, but there's a whole bunch of more complexity in there. You want to be able to ask you questions about enterprise data. Like, I'm trying to make a decision at a company of, should I basically expand into this country in Asia and why, right? And that requires it to understand, well, what locations do you have in Asia right now?

What are the reasons for going into that? What does your cause structure look like? What are the different metrics? How do you do hire? Those are things that are not available on the public internet. And so what I think a lot of people got frustrated is that they want to be data driven. They want to ask those questions. But without that context of the enterprise, you get like generic ones.

Like you will be in finance. You're a bank. You ask that question. They're like, well, let me tell you what the worldwide web search says about this. That's not helpful to you. What GD1 does, it says, what if I bring you those capabilities that you want? You want really, really high quality answers. I want to ask a question and get an answer like I would if I had a really, really smart data

analysts who understood everything about the business, everything about our metrics, everything about the relationships, and went and did that analysis for me and gave me that response, can I self-serve that, right? And so what I care about is security. I can't access data that I shouldn't know about, right? I'm not in all the meetings. Can it be really high quality? If you give me something that's only accurate 50% of the time, who cares, right?

And can you do it at a cost-effective one? because you have a bunch of finance folks. You're hearing all these comments about token maxing and tokenomics that basically the spend is going on there. So can you do all of those things for me without me having to worry about the detail? And that's what Genie 1 delivers, right? And again, has access to the data. It has access to your context.

It applies security. It automatically is choosing the best model underneath and training it to basically give you the most cost-efficient answer and does it with really high accuracy. And that's why we've seen incredible adoption of something like that. You answered a ton of my questions that I had in my head about like what models you guys are using. So I'm assuming you're using Anthropic, Open AI, Gemini, all across the board, maybe some open source models as well? Or do you have some specific ones?

Yes, to all of those. Right. So I think that there's a couple of things I would say. One, from a Databricks perspective, one of the nice parts is when people use it is that you get access to all the models. So everybody is deathly afraid, right, of, oh, I'm going to pick one foundation. model provider and then what happens if tomorrow the next one is or the government took access to this one away and so that notion of hey inside of data bricks i can use any of them that i want so you mentioned anthropic you mentioned open ai you mentioned google and gemini and actually grok as well um you know

from x-a-a-a-a-a-c so all of those are in there and at the same time it's access to all of the open-source models because the other thing that you're very quickly realizing is models are expensive and so what i will give you is that, like, imagine you have a cardiothoracic surgeon, like a one for if you have a really life-threatening, you know, kind of surgery, you want that person to do it. But if you're basically trying to teach your five-year-old anatomy, you do not need that. And it's the same thing, right? Like, you have these great models that are really, really powerful, but they're also expensive. And sometimes you're like, you do not need that for crafting a 10-word email response to your basically like coworker. So inside of Databricks, there's two.

things. It basically has all the models and using something called Unity AI Gateway and Omnigent, it'll automatically route to what is the most cost efficient but effective model that you can have. Second thing you asked before is we took some of those basic questions that you asked of like, hey, help me. I am, you know, I am Adidas and I want to ask a question and understand about what is like all of my customers doing. If I just went and got the best models and asked a question, my accuracy was 50%. So a bunch of it. of the work that we do around quality and context outside of the model is what allows Genie to basically give you north of 85% accuracy. So models are great, but there's a whole

lot of work that you can do for making them cost efficient and much higher quality that we also do. Yeah, I mean, something that I'm guilty of is like using the most powerful model for like the basic, basic questions and then I'm hitting my usage limits pretty quickly. So I think, I think that's one of the things that people I've been talking about recently is like the token maxing stuff, causing the shock invoices and now companies are starting to cut back. So, so you have some sort of router within your system to allow that to pick the most efficient model for that specific task, which is, I think, going to be very useful for companies as they, you know, start using more and more AI. But that does beg the question, like, how does that impact your business model? Because, I mean, I'm assuming it's like a usage-based business model, right? So like if you're, um, If you're capping the overspend with some of the tools that you have now, doesn't that also cap how much money Databricks makes? Yeah, but look, I think that it's a great question, right? But it was one of the reasons that from 13 years ago,

one of the core pillars that we had out there is we said like Databricks is going to be customer obsessed, right? And let me give you why. And so we do a lot, both from manually our technical teams, our field engineering team to do it, but more recently, automatically with AI, we do a bunch of things that will automatically optimize people's like workloads and reduce their spend. And you're like, why would you do that? It's going to hurt your revenue, right? But let me kind of explain. If you don't, here's what happens.

A lot of people get going and they're really, really eager to, left to their own devices. They're really eager to basically show something works well. So they basically spin it up. They don't worry about optimization. And then, hey, it worked. Let's scale it up. And now you have an unoptimized thing that you're scaling up. inevitably, that line item is going to get big enough that a CFO is going to say,

hey, what are we spending on this thing? Do we really need that much? And then somebody's going to go back and do the work. And they're going to be like, hey, look, we did all of this stuff and we optimized it. We took out 50% of spend. The second you do that, the CFO is now like, well, clearly these guys is highway robbery. They've been jacking us. They've been doing all of those kind of pieces, right?

And so then you're in trouble. So instead, we found that, look, if we build an incredible product and we continue trying to basically make the customer successful, so they're only paying what they get value out of and you deliver that, there is tons of workloads, you know, basically that they will then want to put on the platform. And I'd like to say that strategy has served us well. I think we recently said what we are, we're north of 6.9 billion in ARR growing north of 80% year over year. So clearly like, you know, know, the value proposition is resonating with customers. I did see that. I saw the CEO said that you guys are growing 80%. But I think I also saw that margins are falling. So is that just the price of just doing the land grab right now? Is that just like the cost of doing business? Yeah, look, I think that in that world, I mean, there's no secret about it in in pure software areas. And then when you look at it right now from more of an AI piece as people use you, the margin profile is different between the two. I think that there is an element of everybody's driving adoption,

but we saw this in the earlier days of Databricks also when we got going. As you drive more adoption and as you deploy some of these strategies we said where it's like, how are you routing it? How are you building kind of intelligent small models that are lower cost and can do it? I think that you can basically bring margins increasingly in line as well. Gotcha. I'm really curious to get your take on this, on what role Open AI and Anthropic are going to play in all of this, right? Because you talk about how they're, you know, you're using their models right now, but they're also trying to build out their own, you know, enterprise side of the business. They're trying to get access and get closer to enterprise data as well. How do you see that playing

out? Do you just, do you, do you, first of all, do you care if, do you care if one or the other wins, Open AI versus Anthropic? Or do you just, do you just, do you just, do you, do you, do you, do you, do you, do you, do you, do you, do you, are you just happy to plug in whoever's tools work best? Are you nervous that you're using their frontier models and they could potentially gatekeep it and just use it for their own uses? Like, how do you think about all that? Lots of, lots of questions in there. So, right?

So first off, we consider all of them partners, right? You know, Open AI, Anthropic, you know, Google and XAI with GROC. I think we think of all of them as partners. And I think one thing that they do really well that the world needs is that they build incredible like foundation models and they're going to continue to do so, right? Which I think is awesome. I think at the same time for us, we look at what is it that customers care about? And as I said, one, I don't think this is a winner take all market just to be.

It's just like the hyperscalers as well, right? Like you have multiple hyperscalers. There's always going to be a benefit of having multiple ones out there. Customers want choice. Increasingly as the spend on many of these foundation models becomes one of their large line items, they want economic leverage to be able to do it. And so what I hear from customers is saying, look, those are great. You know, I mean, I know it's cliche, once a consultant, always a consultant, you use frameworks, right? So I've got some customers are like, look, what I care about is choice.

I care about control. I care about context. So making them high quality and I care about costs. That's what I really care about, right? And I think part of that means that they want to be able to leverage all of those, but also the open models at any given point that they want to take. And then secondly, the model is a piece of it. I think the one thing that Databricks is done over time, which is, you know, very different, is that this whole notion of how do we have governance, how do we bring all the different data, how do we build context? I think that's something, as I mentioned, we've got over 20,000 customers who've trusted us on that. And the model is a key ingredient to bring in it. I think it's very difficult to say, hey, somebody else will kind of try to run the full gamut.

So for right now, I think that they're most focused on building incredible models and then partnering with the Databricks to basically say, they need all of those other pieces to make the models work in the application. Do you see the cost of these models just, do you see as a race to the bottom? Because that's kind of been my theory is that at some point, like the cost of these models, whether it's from OpenEI or Anthropic or just some of these open source models from China, the cost are just going to go lower and lower. It's a race to the bottom. And the companies that are going to end up winning are the ones potentially like data breaks

that how the relationship with the customers that have access to, you know, enterprise data. I mean, how do you see that playing out in the next two, four, six, 10 years? Yeah. So I'm clearly a little bit biased, you know? I know, you're biased, but I just want to get your take on it. Look, I think it's an interesting, uh, it's an interesting dynamic because on the one hand, everybody will say, uh, the price of tokens today is heavily subsidized. Everybody will say, hey, look, they've got to go up because frankly, when you look at some of the

valuations also, you're going to need not. just revenue, but you're going to need basically gross profit and margins to be able to justify that. So I think that you're caught into two interesting models. One, you're building new models that are incredible. And I think the models that they build will have some capabilities that for a specific set of use cases, you absolutely need those models to be able to solve those, right? And you're able to charge a premium for it. And this is true of any market that you have. If you have something that you basically consult that in its use case is great. But I think what many people are going to find is, one, that is expensive.

And two, for you and I talked about, for a vast majority of your use cases that you have today, you don't necessarily need the greatest and best, like, you don't need the best model on extra high reasoning. It's just going to be expensive to take you longer and it's going to overthink. So I think you're going to have that, you know, that balance. And I think that that's a good thing. You're going to have some amazing foundation models, but you're going to have some really, really good smaller models and the open models. We have that thriving ecosystem.

And I think that there is a lot of value will accrue to kind of Databricks of strategy, which says, hey, customer, whatever use case that you have, we will automatically find the kind of most cost-effective model that can give you the quality you need for this use case. You can easily switch between them. You can bring all of your data to make sure that they're high quality. And then what people care about. people don't care about a model. This is like in the old days when everybody said,

which Pentium, um, right, I remember that. Number 133, who cares? Like, what you care about is, can you build this application, right? And those applications and customers were unlocking, having that relationship with the customers, I think it's really, really important, you know, and sets it, sets it up well. Yeah, I mean, I think it's going to be interesting to see how it plays out. I think that my personal theory is that it just all becomes a commodity and then ultimately the people that win are the ones that and help route to the best models. And I mean, we've seen the prices of these models come down.

If you look at the price of a model that came out a year ago, you know, 4.5 or some of these older models, I mean, there's so much cheaper than they were when they first came out and the prices are just going to keep going down. So it'll be interesting to see how this plays out in the role that a company like Data Bricks plays. I have to ask you this question before we wrap up here. When you see some of these massive IPOs, I'm sure you have your, you know, you keep an eye on it. You know, the SpaceX was obviously huge. Cerebris was huge. there's always there's always there's always there's always there's always there's always

chatter about data breaks potentially IPOing soon I mean do you do does that does that get you a little excited do you do you is your inbox blown up with some of these bankers emailing you nonstop like how do you how do you how do you take this all in uh look it I you know it's funny I think uh the only question I get asked more right now than when you can IPO is what do you think of the SaaSpocalypse I think those two like equal of like how often I get asked about them Look, there's no, like, Databricks, it's not a if, it's a, it's a, it's a, it's a win. We're going to be a public company, but, you know, candidly at the same time,

you know, we've always said when the time is right, because, you know, just being fully transparent, when you look at why organizations IPO, it's for one of a couple of things. It's one, um, you need visibility from a name perspective, uh, that historically has not been Databricks's issue. Um, you say a lot of people will be like, hey, we can't, you know, if you're not public, we don't trust you, you're a small company, we can't buy from you. Right now, like, what was last round, like 134 billion? That is in our third is you're like, you need access to capital. Many of the reasons why these foundation model providers are going is because the size of capital that they need is just, it's not

available in the private market. That also hasn't been our issue. And then the last, most important one is that you want to be able to give liquidity to these employees that help built the company. And, you know, we've now set it up that we do kind of regular tenders where basically employees can kind of get liquidity for their share. So with all of those, there's not this inherent pressure. No, I don't look at X and say, oh my God, they went. Well, how do we go basically the week afterwards? We'll be a public company when the time is right, but also being private has it enabled us to move faster and nimbleer and focus on kind of long-term strategy in a way that I think has been incredibly beneficial to us. Yeah, but think about how cool it's going to be ringing the bell at the NASDAQ or the New York

Stock Exchange, getting your poster on up there. So, I mean, like you're saying, it's a matter of when, not if. We look forward to that date. Arslan, I appreciate all the time today and looking forward to seeing all the cool stuff that Databricks does now and into the future. Sounds great. Thanks again for having me. Appreciate it.

Appreciate it. Cheers. Have it going. Bye. Well, all right, guys, hope you enjoyed that conversation with Arcelon Tevacoli. It'll be really interesting to see what role DataBriggs will play in the enterprise AI space moving forward. Personally, I am pretty bullish on companies like Databricks

and already have a relationship with enterprise customers and access to their data. And right now, DataBriggs is doing pretty well. Their revenues are growing by 80% and I'm really curious to see when they're actually going to IPO. Let me know what you guys thought about today's conversation. Drop your comments on Spotify and YouTube. And while you're at it, consider giving us a five-star rating as well.

All that engagement really does help us out and it helps other people find the show. Thank you guys again for listening, watching, and commenting. Shout out to Mike for all the work behind the scene. And we'll see you guys back here tomorrow.