Databricks hits $4.8B revenue run rate with 55% growth as Ali Ghodsi argues LLMs are already commoditized

to hold up the process until you can get clarity on how they're going to treat that that IP, right? That's really

fogghorn leg horn is very very important. Um anyway, uh our next guest is already in the reream waiting room. We have Alli from Datab Bricks. Welcome to the show. How are you?

So sharp. So sharp.

You look very sharp.

It feels like you're wearing a suit, but you're not. It's your own thing.

But good to see you again. How are you doing?

Likewise. Good. Good. Good. I didn't get the memo. I'm not dressed up the right way. We should have. No, you you look fantastic. Don't worry about that.

Come back on tomorrow. We'll we'll ship you a Santa outfit. Can raise the next round. We still have a few letters left.

But uh congratulations. Give us the news. What what what happened?

Yeah. What happened is that uh you know we're just seeing an acceleration in the business especially with respect to AI and actually we slice and dice. We're data companies. We slice and dice that data in many many different ways. Is it the particular customer concentration? Is it the particular cloud? Is the particular product? No. It's like, you know, broad-based acceleration throughout the business. U you know, we passed $4.8 billion run rate revenue.

Awesome. Oh, wow.

I was asking, will I get a gong? Thank you.

Yes, of course. Of course.

Of course. Maybe multiple.

Maybe multiple.

Perfect. Yeah. So, over 55% growth. Um and uh but you know, more interestingly, we uh we're now passing $1 billion on our data warehousing product, which we announced four years ago.

Yeah. And we also passed $1 billion run rate in AI revenue which you know makes it you know over 25% of our revenue. Um you know not that many companies you know there's a lot of talk about AI what's going on with AI is there bubble or not many companies have over 25% of their business in AI and you know over a billion dollars. So um we're excited about that.

Yeah. How are you thinking about classifying that? because I imagine that in some ways um a lot of the data warehouse revenue is from AI companies that want to warehouse data because there's more and more need for data because everyone's doing more things with it. We're in this era where uh everything is becoming AI and yet it can feel a little disingenuous when companies slap AI on the back of their name.

It's hard to classify too because we were looking at some data earlier this week from RAMP that basically said like 50% of businesses aren't spending any money on AI directly yet they're clearly buying things from companies that are using AI. So

yes, where in the supply chain is it? But how do you think about uh what is the main driver of that data warehouse growth? Yeah, I mean that's kind of interesting. So, you know, um we we started the company 2013. Yeah. And one of the reasons we started was because we wanted to do AI and the problem is since 2013 up until November 2022, no one wanted to hear AI.

Okay. No one was interested in it.

And then come November 2022, Chat gets released and now everyone wants to just talk about AI,

you know. So now it's like, you know, well, you know, are you an AI company? Are you doing AI? What's AI? But yeah, know when we we split our revenue up, right? You can think of it like you know roughly two billion of data munching and securing that data

one billion of data warehousing and then one billion of AI. So you can like separate these out. So it's not like over overlapping triple quadruple countering or anything like that. Uh so what are they doing with AI? I mean there's just lots of use cases. So let me give you an example. So we have this product called agent bricks. Um you know one of the favorite use cases I have is that Merc actually created a model. These are called transformer models. these AI models they created something called Teddy which stands for transformer enabled drug discovery

and this is an AI model instead of predicting the next word in English sentence it can predict cells and cell expressions so it's called gen gene regulatory network this is super important for drug discovery because then the drug companies can target those genes that are responsible for diseases

so that's like a game changer um and like lots of use cases like this 7-Eleven

was able to automate most of their marketing stack. So before you would have to do your own segmentation like who should I show this material to who this will be interesting to and then you would have to create the material for that segment manually. Now the agents can actually prepare a lot of that material themselves. They can generate images. They can rewrite the text to target a specific group. You can do that much faster now. Um you know I could keep going. RBC Royal Bank of Canada.

Basically when a earnings call comes out you guys will like this. Earning coms comes out. It fetches the earnings call. It looks at all the previous earnings calls, looks at earnings calls of competitors, looks at what's going on in the market, looks on, you know, the sentiment, and then writes up the equity research report using the tone of the equity research analyst. And they can get this out in 15 minutes,

which is, you know, usually in the industry, it's at least two hours or several hours.

So, every every minute, every minute matters, every second.

Exactly. Exactly. So, those are some of the use cases you're seeing out there. So, it it is taking off. You know, we're kind of focused on this. you know, I call it sort of, you know, uh, get stuff done or GSD AI rather than, you know, super intelligence AI. How much of those examples that you I mean there's a great case studies very very tangible in a world where there's you know not many tangible AI narratives or or um how many of those how many of those um case studies came from companies that were already using lakehouse or a data warehousing product and then you came to them and they were comfortable working with you or they felt like they would be sped up on working with you um on the AI side because they were already comfortable with you housing the data for probably years.

Yeah, almost all of them. Let me explain why though. Yeah, there's a reason why. It's not just, hey, we know each other and I'm comfortable with you. Yeah.

Uh it's the fact that AI today, like large language models

are actually a commodity. I mean, they're the coolest thing that have ever happened, but they're already a commodity. What I mean by commodity is just like oil. You know, you could get your gas from this gas station, you can get it from that gas station. Slightly different in price, but doesn't really matter for your car, right? Same thing with LLMs. You can get this LLM or that LLM. You don't even know the difference. This week, this one is better. That week, that one is better at, you know, coding. This week, this one can do reasoning. This one can do images. So, it's a commodity. So, actually, it's now a commodity to have a generative AI model that can answer any question about the world and the history and it's super smart.

What is super elusive and you can't get your hands on is an AI that can actually reason on proprietary private enterprise data.

Mhm.

Okay. that's that actually doesn't exist because the techniques used to generate these AI models work on huge amounts of data like when I mean huge I mean like the whole internet like take all the data and crunch it doesn't work on a smaller data that enterprises have so you how do you then make the AI work on that data that you already have inside the enterprise right this could be electronic medical records that are sensitive it could be financial models that are super secretive it's your recipe for your alpha generation um so what you need to do is you need to get a good data foundation there and that's the lakehouse that you're talking about and then the governance and the security on top of that with unity catalog those are the kind of things you have to get in place first before you can unleash the AI on it so that's why it's not that they just know us it's that they've been at it for a long time and actually if you haven't gotten that foundation right if your data is a mess it doesn't help if your data is sitting in on premise in some old database you know some Oracle database and it's locked away what help is this commodity AI that you have on the top Yeah. Um, what advice do you have for founders that are building um, their business around in and around open-source technologies? We we we talked to a lot of young founders that um, you know, are are are open sourcing something really cool. They have a bunch of GitHub stars. There's energy. They don't really have a business model yet, but you can see that they're delivering value and at some point they're going to capture it. What are the dos and don'ts of growing up and being a good steward of an open source project or being a good partner to the open source ecosystem? What would you counsel a founder in that situation?

Yeah, that's a billion dollar question. It's very hard to get that right. Um, you know what I would say is number one, there's actually kind of two requirements and there's like an analogy I have that you kind of have to get right. First of all, realize it's a very difficult sport. the sport that you're about to get into open source and to monetize and create. It's like not an easy sport. So, you know, preferably don't get into it because you might, you know, get hurt. Yeah.

But if you decided if you're hellbent on doing this,

then, you know, you can think of it as, you know, your strategy basically is to hit two home runs after each other consecutively. The first home run and you're banking on hitting that home run is for your open source project to take over the world. Everybody wants to download it. It's like the greatest thing ever. If you don't hit that first home run, you just shot yourself in the foot because you just gave away all of your intellectual property and you got nothing back for it. And now you have nothing. So you better have a first home run and the open source project takes over the world. So that's like step number one.

Now you've achieved that. Congrats to you. Everybody's paying attention. They're downloading your open source project. The GitHub stars are just like, you know, going vertical. It's awesome. You don't have any monetizable business. You cannot make any money. So then comes home run number two. Home run number two is now you have a 10x just as good as the first one innovation on top of the open source and that's the proprietary value that you're going to provide to these companies. That's why they come to you otherwise you know they can get open source anywhere u and that's how you monetize and now you have the sort of secret sauce to really successful company. If you don't hit the second one, the problem is now the big guys, you know, the hyperscalers and others just pick up your open source model and they offer it up for almost free. How do you compete with that? How do you compete with that free? That's where you need your second home run. So, it's very difficult.

Yeah. So, I want to double click on that particularly. uh how do you compete that we have because we have other founders that have an open source project and it's growing and they're doing well and then uh we see the press release from one of the hyperscalers that their product is now a feature. Their child has been abducted into the cloud uh and it's always a you can just tell no matter how they respond it's a scary moment. It's an existential moment. We know that there's a board meeting happening happening. If uh if you're in, you know, a board meeting, a fly on the wall, an adviser or something, and there's a founder who says, "Oh, one of the hyperscalers just launched a clone of our product. They're really going to come for us this quarter, maybe next quarter." Um, what do you advise them to stay in the game while the capital cannon of the Mag 7 is pointed directly at them?

Well, well formulated. So, I actually get that call from the founder day before the board meeting and they're like, "Oh my god, you know, what do I do? My life is over."

And I tell them, "Congratulations,

you just proved that you hit your first home run. They're paying attention to you. So now they're coming after you. So now you need to do another home run. Okay. But your first home run, you just actually succeeded. Everybody's attention. It's a home run business.

They would not Yeah. They would not direct all those cannons towards you if you didn't have anything. So you have something now. They're direct. Now they're coming after you. So now you need to innovate and you need to actually monetize that and that portion you cannot open source if you do they'll just keep you know milking that with their tenants right uh so that's the way to go about it and it's normal part of life and by we all went through it I went through it I remember like you know waking up you know cold sweats in the middle of the night and you know freaking out about like oh my god this thing will kill us uh but you know the good news if you continue innovating you can stay ahead because you know the people that want to copy you they're always going to be a few years behind so if you continue executing you can stay a ahead and if you did a home run once you can probably do it again if you set your mind to it. So just the trick is to stay behind stay ahead right because the way it works is that startups the ma the main thing going on is that they don't have time on their side right like if you could freeze all startups for 12 months all the max 7 would catch up immediately and there would be no startups right so your whole thing is outrun the big guys very very fast so that's another thing I tell them don't slow down move very very fast continue innovating

yeah I love that

how do you uh how do you categorize everything that's happening in AI like what what is your personal framework because like what you're doing is very different than let's say a foundation model lab which is different than uh you know maybe maybe an end uh rapper company

is picks and shovels a porative is is is that a pjorative

no I mean look Levi still around I don't know how many of those gold diggers are around

that's true [laughter] that's a great point I never thought about that

yeah still here kicking you know here in San Francisco but I would say there there's kind of like three paradigms going on at the same time paradigm number one is the quest for super intelligence and the quest for super intelligence uh you know requires bigger and bigger data centers more and more gigawatts and it's very resource intensive and the idea is to come up with such a smart model that can improve itself to the point that we don't like you know it's just curing cancer and coming up with new AI models and of course we should be really afraid of the you know safety aspects of that so that's like you know what the frontier labs are doing uh I would say you know category two is the researchers that created these models in the first place like Yan Lun or Rich Sutton who created reinforcement learning that technique, they'll actually tell you like, hey, that's not going to actually work. You're not going to get to super intelligence that way because that's not how humans learn or animals learn and so on. So, they're like, hey, we're, you know, leave us alone and we'll get get to something in 20 years,

which to me sounds like a little bit like, hey, we don't really know. Leave us alone. Uh, then there's a third camp. The third camp is, hey, we already have all the AGI we already need. We might not have super intelligence, but we have artificial general intelligence. I'm in that camp. I believe we already have artificial general intelligence. It's generally intelligent. Just use it and you'll be convinced within like five minutes. It's pretty smart. Okay. So now start applying it and get stuff done inside of an enterprise or inside of an organization. Just GSD deliver AI value on basic tasks and if you're unsure start with the simplest mundane tasks that you know are, you know, really low cost right now. See if you can start augmenting and automating them.

How did you how did you process the talent wars this year? Uh I think the question on my mind is will we see you know a1 billion dollar pay package in two in two years from now or even in in next year right uh but I'm curious how you processed it as somebody

was a one time thing or is it normal

that was last year guys uh you know I think I think no one at character got like a billion right I mean they paid billions for character.ai AI and that was literally a billion dollars for one guy at least 1 billion. So that's already done that. So I mean 10 billion is the next I guess package to go for [laughter] and maybe that was Alex at Meta but uh no but uh you know joking aside um I think that it's also in everybody's interest to exaggerate these numbers.

Yeah.

Like the joke among CEOs is let's exaggerate that number right oh you know companies are making like hundred billion dollar offers for my employees

because when you say that what happens is that first of all you must have a great team. Yeah. First of all, it's like a compliment to your team. Second, if someone comes after someone in your team now, they're going to expect at least that amount. If they don't, they feel insulted. You know, you're not even you're not even offering me a billion dollars. What's going on here?

So, that's a that's value my skill set clearly.

Exactly. Exactly. Right. So, it dissuades everyone. So, I think there's also a little bit like self-reported numbers here going on. uh the market is not as crazy as you know people say but it has been pretty nutty and especially on the research side uh in AI but actually I think it's coming uh like it used to be the case that if you're a big shot professor in AI you could you know get funding huge amounts of funding for crazy valuations and if you you could just say I want to just do research and get to super intelligence those days are over actually already it's no longer the case that you could that pitch doesn't work anymore uh and there's no capital like you're not going to get a billion or $2 billion to go just do research on AI. You need to have something more than that. So I do think actually the market is coming to its senses.

The timeline the last 24 hours has been fixated on the role of storytelling within companies. You're a fantastic storyteller like very just fun to to listen to and and animated. I'm curious were you were you always goated or did you did you have humble beginnings?

There's no way to answer that question, right? [laughter] It's like a trick question.

I'm saying no it's not. No, I No, I No, I I I genuinely believe there are founders that come out and they're seed stage and they're hyper engaging and they're incredible to listen to and they get you excited and motivated and then there are other founders that like have to really learn and but I but I think it's like in some ways it's like you either have it or you don't.

Yeah. Look, I think that there are many things that go into like running companies. Like one of them ingredient number one is to be engaging and excite people and get people excited about your vision, right? But number two, you also need a good strategy. If not, if you don't have a good strategy, eventually it's going to catch up to you. Number three, you need to put together amazing, stellar, world-class team, right? That's all that matters because if you're amazing, but your team is not, then you're not going to go anywhere. Even if you have all of these, it's not enough. You need to also execute. Like, you need to actually get the team to execute again and again and again and get to the results. And that kind of matters more than anything else. I mean, look at AWS. Like, they're phenomenal at executing, right? And then, uh, finally, five, you need to do it with good culture. If the culture is not good, it will eat you up from inside. the key people will leave. Um, you know, we've seen, you know, companies that went under because the culture was bad or the CEO got fired and, you know, they went through tough times and, you know, the, you know, authorities were after them for fraud and these kind of things. So, you kind of have to nail all these five. Um, you know, it's the unfortunate truth.

Uh, well, uh, is it series L? Now, you're quickly running out of letters of the alphabet. Uh, we have to ask, are you going to be hanging out with us at the New York Stock Exchange anytime soon or are there are there no plans to go public because we'd love to hang out?

We can hang out there. We can hang out there anytime at the New York Stock Exchange or NASDAQ. Uh, you know, but uh but whether we're going to go public, look, we're I mean, I wouldn't be I wouldn't rule it out that we would be public by end of next year. Uh, but you know, it's also not like it's guaranteed because the thing I want to avoid is what happened in 2022. Yeah,

in 2021 I had a lot of peers that were so excited. They took their companies public. They're ex great storytellers, great execution machines. They were awesome.

But then interest rates went up, inflation went up. And in 2022, they said, "Hey man, you got to like get IBIDA. You got to produce, you know, margins." So what did they do? They all did layoffs. Yeah. And they all produced IBIDA. And what did they cut down on? All the futuristic stuff, all the innovation, all the new products. And also employee morale kind of tanked. Yeah. Because you know employees are like I don't trust you anymore. You said this company is gonna do well but now you know you're firing my friend here. So um you know and you know as you can see we're now growing twice the growth rate of many of those companies and I think staying private in 2022 had something to do with it as well. Uh so you know um you know we're not trying to exactly time the rock market. We're trying to win it.

Yeah. Do you think that the the narrative around open-source LLMs is like is over? Do you think it's understood? Because uh you know the deepseek moment was a moment when everyone was saying oh maybe open source will win or it'll have a serious uh shaping of the market structure in AI. It feels like uh folks have a much more uh concrete thesis around the shape of the AI market with you know like a you know an aggregator winning in consumer and then maybe more of an igopoly and in enterprise codegen and API businesses um and and certainly the the open source models fit into that but how are you thinking about open-source AI models uh developing maybe in next next year the year after? Yeah, [clears throat] I would just say that open source has always been a commoditizer in business.

Sure.

You know, like when Linux showed up,

like is the greatest thing that's happening right now in the world, Linux? No. But what did it do? It kind of commoditized Windows,

right? And significantly so. And we've seen that again and again and again. So what it does, it really it's it sets the price of the IP of the open source project down to zero.

So that price of that intellectual property shrinks to zero overnight. Yeah.

Uh, and that's great because that basically puts pressure on the LLM foundation model layer to have lower and lower prices. And the lower those prices go, the complimentary market, the, you know, market on top, which is us with agent bricks and, you know, Blakebase, our, you know, our database for agents, uh, helps that market cuz, you know, it makes our COGS, our cost of goods sold, which is we would pay for those LLMs, lower. So, I think that's going to continue happening. Uh and in particular one interesting trend is that China is putting the pressure on just open sourcing everything. They're not just open sourcing the large language models and the weights. They're open sourcing the whole stack. Everything they did. This is the way we open source it. This is the data we use. Here's a research paper. And they're having really fast diffusion of innovation happening in China. And that also puts pressure on the west over here. We're like, okay, we can't fall behind. So we're doing the same thing over here. So I think that's just going to continue. So I mean people I I I completely understand that narrative and uh it does seem like it's feels like China's commoditizing American Foundation Labs. Uh and it and you easily like fall into this like geopolitical narrative. But is there a world where in China it's it's really just like uh the the classic open- source technology with a for-profit corporation wrapped around it and they're trying to build like the red hat of LLMs over there and like that's the real uh motivation more than just some like crazy wrench in the system.

Well, I think that I I don't exactly know how the communist party thinks and what they are doing, but I think that over here we will have companies that are already doing that. Like if you talk to startups already,

many many of the startups in the United States are actually using the open source models and the Chinese models. By the way, what's already happened which people don't talk about openly, it's really hush hush is that they're taking these Chinese models

and kind of distilling them and you get an American model,

right, with an American name and you don't know that it actually tracks back to Quen from Alibaba or something like that. That's already happening in this market. But like look at the vast majority of startups. the huge amount of open- source Chinese models that are being used by American open source models. And I think you'll have actually Red Hat-like companies here, even here in the United States, that will say, "Hey, you know, we'll help you get up to speed, use the right ones, make sure it has the right safety and guard rails and you're not taking any risk and we do it at low cost." That's already happening in some sense. And then there are people who do that in the cloud. We actually help you do that. We at data bricks we are multi-lm

so we have OEM so openai is available natively to all of our customers and tropic is natively available to all of our customers and Gemini is natively available so it's like if you're using dataix those three are actually under the hood and then we have all the open source models to let you so since we're the next layer on top right that's the complimentary market that I was talking about uh so I think this will continue and I think actually that's good for innovation that that's happening there's so much pressure on that middle layer because it is a very lucrative layer right I mean you you know as judging from the valuations of entropic and open AI that are very welld deserved um you know that's an important market so uh I think this will just continue we'll see more and more open source that's not going to stop and it's going to be from China my prediction is that you're going to also see that now from Europe and United States

cuz you know do we want uh Chinese models to take over the whole planet in open source I think that some nations will say hey why don't we just get together and fund a supercomput and then build one here by the way we used to do that back in in the day was called supercomputers. Yeah. National labs that had supercomputers. The largest computers on the planet used to be supercomputers state funded. That's going to start happening you know in either this country or that country. And once that starts you know everyone wants to get in like you know United States is going to have the biggest one right. So that's my prediction. [screaming]

What does your sales process look like at this stage? Like where where are you investing time specifically uh with with uh customers that are that are in the pipeline? like what's the what's the highest leverage and what is even what does even the workflow look look like?

I mean we have many different segments at data bricks. There's a self-s serve segment.

Yeah,

you can actually sign up. There's a free edition of data bricks that's really important funnel for us. The free edition anyone can just sign up with their Gmail and use data bricks forever but you don't get you know lots of compute. You don't get lots of AI. You get a little sliver

so you get a taste of data bricks forever.

Then you know we have

good this is your sales process. you're selling selling right now. [laughter]

You should try it out. You guys have an email address. If you have a Gmail address, you know, go to database.com/try.

And so once you've done that, so then comes Yeah. Then comes a trial. The tri, okay, but that was just a sliver of compute. You know, I want to I want to process a lot of LLMs. I want to use the bigger models. I want to do more interesting things with agent bricks. I want to have my database with Lakebase. Then uh I would say get the um free trial. Mhm.

So free trial, we give you I think $200, something like that and you get to spend $200 for free with us.

And so that's sort of the funnel and then from there people convert. This is really how most SMBs, mid-market companies, that's how they get started on data bricks.

I'm talking about I'm going to I'm talking about the big dogs.

Yeah. How different is the organization? Do these people like the the the person who maybe is responsible at some level for the biggest deal you've ever done and the person who's responsible for the smallest deal? Are these like have these people ever met in the organization? Are they completely separate? Is there some crossf functional opportunity? Because they're very different functions. I imagine one is steak dinner, the other's like UIUX. I would imagine, but I don't know.

Well, believe it believe it or not, the steak dinner days are kind of over. Okay.

You know, I mean, like it's not like, hey, we'll just take you out and just we have like the best golf players and we have the best drinkers that you know.

To be clear, if you take us out to a steak dinner, I mean, we'll go to databases.com. I would

Yeah. [laughter] Yeah. And and you pay me $0. Exactly. Right. Uh but

I was hoping I would sneak that by you and you know exactly futures are selling off massively right now.

Exactly. [laughter] Are just going down. Yeah, exactly. I'm sorry guys. But what I would say is that uh you know um we actually have a funnel. So uh you know you start maybe as a sales development uh leader. Actually we have one at data bricks. you know, guy called Matt started just, you know, hitting the phone, just getting these SMBs and then grew with the business, then got into the sort of more mid-market accounts and then now like on the enterprise side. So, you know, we have people who grew up with databooks and they kind of, you know, you learn to do the whole all the way to the, you know, it's a lot of it is just getting comfortable, right? Because it's like it's one thing if I'm negotiating with you a $30,000 deal.

It's another thing if I'm negotiating with you, you know, $250 million deal. And, you know, a lot of folks will just get nervous and they'll just start shaking, right? So, uh, but how do you do that? How do you get that, you know, 200 million, $200 million deal? Um, you know, it actually goes back to those free additions and those things.

In that conversation with an exec high up, we will say, did you know that you're already leveraging open source spark throughout the organization already?

Did you know that you're already having people using the free edition with their Gmail accounts? Uh, you know, you have all these trials. By the way, the things I'm telling you about today, if you don't trust me, you can go yourself try it out on the free edition. So actually that kind of motion bottom up motion helps a lot even in the top down big enterprise sales.

Yeah.

Um and that but there the goal is what are the transformational projects that you want us to help you with and how do we really drive business value for your organization like what's a big strategic board level initiative?

Let us help you make you successful with that.

We're trying to get to we're trying to get to 24 hours a day. And so we're really trying to use AI 24 hours a day live. Use AI to optimize when John