Databricks: Data Based Decisions - [Business Breakdowns, EP.134]

0:03This is Business Breakdowns. Business Breakdowns is a series of conversations with investors and operators diving deep into a single business. For each business, we explore its history, its business model, its competitive advantages, and what makes it tick. We believe every business has lessons and secrets that investors and operators can learn from. and we are here to bring them to you. To find more episodes of Breakdowns, check out joincolossus .com. All opinions expressed by hosts and podcast guests are solely their own opinions. Hosts, podcast guests, their employers or affiliates may maintain positions in the securities discussed in this podcast.

0:45This podcast is for informational purposes only and should not be relied upon as a basis for investment decisions. I am Zach Buss, an investor at Avernic Capital. And today, I am joined by Yaniv Suisa to break down the private company Databricks. Grown out of a UC Berkeley research lab in 2013, Databricks has grown rapidly. And after 50 % growth this summer, it was rumored to have last raised at a $43 billion valuation. In the most simple terms, Databricks provides tools for ingesting, transforming, and analyzing large sets of data from multiple sources in multiple formats in order to inform business and engineering decisions.

1:28Databricks is on a crash course with Snowflake to amass market share. In this conversation, we explore the nuances of structured and unstructured data, discuss data lakes, and what it entails to get Hadooped. We hope you enjoy this conversation. Yaniv, thank you for joining us to break down Databricks, a business that's increasingly well -known notoriously for its increasing capital raises. But I thought it'd be interesting to kick this conversation off with a very basic explainer of what the business is and why its success has been so timely in today's environment. Databricks is a data processing platform, but what does that mean?

2:09So there are two kinds of data that we can talk about here. There's what we call structured data and unstructured data. So structured data is data that's organized and formatted in a predictable way. These are the things stored in databases and spreadsheets. So it's really easy to search, process, analyze. You think of your course schedule from college. It has the name, the professor, the time, the location, and it's all formatted in a very clear column by column structured way. So it's easy to analyze. You could think of it as almost creating the Dewey Decimal System for data back when we had libraries.

2:46Then you've got unstructured data on the other side. Unstructured data is where you have a rapidly evolving environment. So things are flying at you all the time. Data is always being produced. You don't just want to look at historical data. You want to look at things that are happening as you're trying to make a decision. So that's the movement -based data or data that's not in a very neat set of columns and spreadsheets. A good example would be social media. You have text, you have images, you have video. Nothing's in a specific standard structure. So it's hard to extract meaning unless you have a unique algorithm or expertise in managing unstructured data.

3:27So Databricks started in the unstructured data world. They made unstructured data accessible to businesses and companies. And Snowflake started in the structured world. So they were helping with the spreadsheets and organized databases, looking for information from past pieces of data that you've collected. And now they both do both. So that's where Databricks plays in this world. So I think to best bring this to home for the audience, if you can offer some real -world examples of how their customers use the product, I think it will really help us to better relate to what the business does. Think about any manufacturing facility.

4:05So Databricks, for example, they work with Honeywell. They work with some of the big healthcare pharmaceutical companies. What are those companies doing? They are producing things in real time on complex manufacturing lines. So you have the inputs and the weight and the amount of the inputs, and you have to refine the temperature at which they're being heated and manipulated. And then you've got to make sure that the quality is good for each of them. and they have the right amount of this or that and that they're packaged properly. And to do that, you need to not only take the data while it's happening from the machines, you also need to take data from sensors that exist in your manufacturing facility.

4:44You wanna use the video surveillance of the lines and the people working on it to make smart decisions. And so in that case, something like Databricks that can process in that real -time unstructured streaming data is really powerful. Another example would be in the financial world. So if you take something like any of the big banks that you bank with, Bank of America, JP Morgan, Chase, et cetera, they're worried about fraud as one of the many things they care about. And it's not useful for me to know that this fraud was committed after it's already happened. What's useful for me to know is these things are being changed and we're seeing these things happening or these qualities coming through on this user right now that suggest threat so that you can in real time make a decision.

5:31And then the last one I would give that I think is probably most relatable are recommendation systems. So in the consumer world, both for the brands themselves and for individuals. So Amazon has this on their website. They have their own recommendation system. But imagine, and this is what Databricks does for customers and agencies, imagine being able to say, all right, well, you're reviewing this right now in real time on this website or in this particular ecosystem or information platform. And I can then make decisions on which ad to serve you or what to recommend in real time while you're doing that when you're most likely to act on something.

6:09Similarly, that's a real generator of revenue for some enterprise companies because they can then say to brands that want to do that advertising, I in real time have this. I can analyze in real time that this is your best opportunity to win this potential buyer. How much are you going to pay me? And they're going to pay a lot higher price for something like that because it should have a higher hit rate. And the way you know that is because of that unstructured streaming real -time data. And then conversely, in a world where Databricks and Snowflake do not exist, how did people handle these problems historically?

6:45So historically, there wasn't this ability to analyze streaming data. So what you would do is you would organize the data that you had, you'd have to organize it first in those files or in those libraries. And then you'd have to go and like your student used to do research way back when, you'd have to go in and read and take out the analytics to understand what insights would be helpful to you. And so everything was historical. But the ability to do things in real time is particularly important. And it's becoming more important and has accelerated because of the environment we're in, in terms of where software is now in the enterprise space.

7:23COVID helped this a lot, but it was already a trajectory that was going somewhere. Companies more and more, all kinds of companies, not just Cisco and Amazon, but also automotive companies and manufacturing companies and logistics companies, energy companies, they're all using data to make decisions in real time. So the ability to do that is becoming more and more important. And that is really the fundamental revolution that Databricks ushered into the market. And you touched on it there, but in terms of the relevance of this business, clearly there was something timely and there's a reason that the business has inflected in the case that it has.

8:03Are there more concrete examples of what happened over the last, call it, 36 months that really brought this product to bear? I think there are two pieces to what made Databricks accelerate so significantly in the most recent timeframe. Apart from inventing streaming analytics, why is it so much more important now? One of those is as you have data in distributed places and coming from more and more locations, coming from your toaster oven and coming from the field of the rails for a train going down the tracks, as you have that data coming from more and more places, you need to be able to process it in a meaningful, quick way to make those decisions.

8:43And we've seen these multiple industries just jump on this bandwagon faster and faster, trying to implement software in an effective way. And I think that's been particularly important during the COVID era and during the supply chain issues, because the focus on increasing efficiency and performance, which inherently requires real -time decision -making in a lot of these industries, became particularly important during that time. The other thing that I think is probably most on people's mind is the expansion of AI. So AI has always existed, but it's now in everybody's day -to -day lingo. And I would argue everybody's using it or most people are using it now day -to -day and that will only increase.

9:26So things like visual identification, like the facial recognition technology that came out, chat GPT, these all are not possible if you can't analyze unstructured data in real time. And that's why Databricks is so exciting. It's their core strength. They were born out of that space. So this streaming data where data is coming in and you're able to look at it instead of having to put it into a spreadsheet or a library book and then go look at it is really powerful. If you would ask ChatGPT a question and they would say, give me a moment, and then I could tell you an hour from now, it wouldn't be that exciting.

10:01But with AI, you could do it in real time and you could only do it in real time through things like Databricks. And then how exactly does the revenue model work here? Just take us through the basic economics of how you convert that revenue into bottom line. The way they make money is a fewfold. So they charge by usage down to increments that are as small as a second. And it's a pay -as -you -go model. And they have different products. So the first product is the analytics, the streaming analytics part of things, which is your basic Apache Spark solution set. I shouldn't say basic. It's very advanced, but that's their core where they started.

10:39They've built the lake house, which allows you to have the best of all worlds. That's where they started actually directly competing with Snowflake. So that was where you could take structured data and unstructured data and use it together. So that's another way they make money. That's one of their products. And then I'd say the third product, and they have some others, but these are the big ones. And I should also say that second product is growing incredibly quickly. In just 18 months, it went from zero to over 200 million in revenue. And then the third product is this new product that came with the acquisition of Mosaic as part of this new product, is the ability to run AI models and use them with your own internal proprietary data at an enterprise to be very effective.

11:20And that's the third product from which they gain revenue. And it's all based on how much you use. And then when people start to talk about the competitive advantages and the growth of Databricks, it generally invites questions about the business and how it compares to Snowflake. So at a basic level, what are the differentiating factors? Do companies generally use one or the other or both? How does Snowflake fit into the equation? So interestingly, they started very different, came to be very similar, and I think now are diverging again. So they started, as we talked about earlier, that distinction between structured and unstructured.

11:58Databricks started as the unstructured streaming, so in those real -time environments, and Snowflake started in a structured environment. They then came together. Even just a year or two ago, their products were pretty similar from a functional perspective. Some were better here on the margins for this or that, but it was hard to really distinguish them completely, except for maybe the open platform. Of course, Databricks was much stronger on the unstructured side, and Snowflake is much stronger on the structured side. Databricks then built its lake house concept with Delta and put both together.

12:31Snowflake has tried to put both together by partnering with folks. What distinguishes them going forward, however, is this AI space that everyone is thinking about, and we've talked a little bit about today. So what that does is really start to distinguish the two companies because AI fundamentally requires unstructured data analytics, which is by far Databricks' strongest muscle, and they have the strongest muscle in the industry for that as well, whereas Snowflake does not come from that world. So I think you'll increasingly see them as being separate entities, whereas Snowflake becomes more of an underlying infrastructure, like AWS or Microsoft has for analyzing those files and structured historical data, where you might be able to do some of the Databricks type of things.

13:18But really, if you want to do things in real time, you're going to invest in purchasing Databricks' product. And then from the perspective of the infrastructure layer, thinking through the value chain here, Google, Amazon Web Services, Azure, Microsoft, how are they symbiotic versus competitive with someone like Databricks? So I think they'd love to be competitive, but they have not taken that approach. And I don't think they could achieve it effectively, to be honest. But they have the marketplace approach. as does Databricks. They're the underlying infrastructure. They're that cloud that most companies are on.

13:55So they're that cloud infrastructure. And then on top of that, you want to be able to do things beyond just buying that infrastructure. And instead of them building everything on their own, they've taken the approach of building a marketplace where lots of people can be on their platform. So you can go just like you go to the supermarket and there's all kinds of brands and all kinds of products you could buy for any different particular use. Same with Amazon and Microsoft. And by the way, they're both partners with both Snowflake and Databricks. So I expect, although people talk about it as them being the competitive models, I actually think Databricks and Snowflake are more of a challenge to Microsoft and Amazon longer term than they are really head -to -head competing or going to be able to compete with one another.

14:37And the reason for that is things like Databricks and Snowflake allow you to use your underlying infrastructure more effectively, faster and more effectively, which presumably Microsoft and Amazon wouldn't like, because that means you're using less of them. But what they're betting on and what we at SignWay bet on by investing in Databricks is that the more you enable people to use data and use these solutions effectively, you're going to compensate for the efficiency by increasing the market size. So there will be maybe the processing will be more efficient, but there will be a lot more of it.

15:14And in order to use Databricks and Snowflake, you're always going to need underlying that Microsoft and Amazon. So I think they're much more copacetic than they are competitive. So to the extent that you can share, how big is this business today, just to get a better appreciation for how it compares to some of the large publicly traded companies in their space? So if we look at Databricks business versus some of the other competitors out there, and I'll give you their numbers in particular. They have tens of thousands of customers, 1 ,200 partners or more now. They have multiple products like we talked about.

15:48So they have many different product lines for the different areas that we also discussed earlier. They allow you to share data, which again is a new element called Delta sharing that they've added that I think will become another revenue stream for them. And their margins are very high. So as software margins should be, their margins are 85%. So think about Snowflake's margins, which are around 56 % as a comparable. The other thing with Snowflake's margins or revenue is some of their revenue is spent and goes back to AWS for the hardware component that underlies Snowflake. With Databricks, they don't have that.

16:25However, they may add that over time. And I think that is actually their intent. So in terms of the comparison, Snowflake is the closest. It's at a 19x multiple. If you look at Databricks' recent round, it about pegs them there. And so I think that those are the closest, but you can see where their growth is. It's just, there's a lot of talk about them going public. And I think the time that that happens is when investors are hungry again for growth, because everything about Databricks' growth stream, number of customers, depth of customers. So Databricks actually has, I think it's over 300 customers now that spend over a million dollars.

17:02So you can see the entrenchment that's really important for a company. When the hunger for that growth, those kind of margins with that kind of growth rate, with that revenue, with that entrenchment comes back to the market, it should have never left. But right now, it's not so much there. When that comes back, Databricks will be incredibly successful in the public market space as well. And then just to give the audience the appreciation for scale, a Fortune 500 company that's a customer of Databricks. It's a usage -based model, and we understand there's not perfect information, but approximately how much are they spending on a product like this a year, and how much could they be spending?

17:39Fortune 500 companies are spending, once they've used Databricks and tried it for a little bit and decided this is what we're going to use, they are definitely using in the seven figures, so in your millions, and many of them, especially in that size company, in the double -digit millions. There are no, I don't think, three -digit million customers, but there are definitely one and two digit million customers. And the key piece for a business model is that that continues to grow within each customer. Ultimate size really depends on the breadth of the company and where they deploy the analytics.

18:10At the end of the day, if AI takes off, these companies could be multiple 10 to double digits of revenue for Databricks each. Particularly if that company is using it in real time often for whatever process they're doing or whatever targeting they're trying to do with customers, et cetera, et cetera, like we've spoken about. This is the key analysis that a customer does when they buy Databricks. Does the increase in efficiency and insight, not just efficiency from using those underlying resources less because I have this pay -as -you -go model, but also the efficiency in making decisions and operating my business.

18:46If Databricks can continue to always increase that efficiency and value, then customers will always pay for it because it lowers costs and increases performance. And at the end of the day, it actually saves them money. And that's the whole business model at the end of the day. So now that we've set the stage for what the business is today, and we'll talk about some of the future implications, I thought it'd be interesting to rewind to a little bit about the founding story, why that's relevant in the data world and the evolution and origins of Spark itself, just to better contextualize everything that came to be over the course of a business that's not that old at this point.

19:20Sure. Databricks was originally founded in 2009. That's when Spark really came to light and they built their own proprietary Spark on top of the existing platform. And in 2013 is when we invested in the company. The Spark technology, the open source platform, was founded out of Berkeley. So Jan Stojka, who is a professor there, and his laboratory, specifically the technologist who's CTO at Databricks, Matei Zaharia, is the one who created Databricks in coordination with Jan and several of the other team members. In fact, it was a group of, I'd say, seven or so. Don't want to not give everyone credit, but it was around seven or eight or so that are considered the founders of Databricks and of Spark.

20:00And all seven of those are still executives at the company. So from that origin, it was this open source community. And what an open source community does is it's free. It allows for free software. It allows for all kinds of developers to contribute to that software, to make it better, to make the applications better. and that was what they really created from the lab at Berkeley. When Databricks became a company, they decided to offer something above and beyond that. So open source has some pros and cons, what Databricks decided to do. And this is something that I can't think of any other company that has done this effectively.

20:37Several have tried, like Red Hat tried and others tried, but very few have been able to create their own version of this open source platform like Databricks did. So they built their own version of Spark. And the reason for that was that they could allow enterprises to use the Spark capabilities or the best of the Spark capabilities in a controlled environment. So they have this nice mix. And that's where Databricks came to existence. A cool, funny thing about it is there's no competitive advantage in open source because it's for everybody. That's the premise of it. But the free piece was actually something that Databricks took advantage of even when it created its own version and brought some of that open source stuff into a proprietary solution set.

21:20Because what they were able to do is make Spark more efficient. So faster performance, lower cost, more effective. All of those increases in performance, whatever synonyms you want to use, they drove the technology so far ahead of where everything else was. And they were able to reduce the cost as a result of the underlying hardware required to run these analytics. So the hardware paying AWS for cloud, paying Microsoft for cloud. They were able to reduce the cost of that so significantly that essentially the software didn't only become free, but it was almost like you were making money because you were spending so much less on these infrastructures.

22:00So out of curiosity, at what point did the success of the business become apparent to you in that you became interested and were able to help support their growth? So at SignWay, we're a very thesis -driven fund, and we had a thesis for a long time. One, how to enable the use of streaming analytics effectively, but also how to put that into a business in a way that really drove turn -on investment for a customer. At the end of the day, a customer cares about whether it can do something that reduces its cost or increases its performance, i .e. its revenue. At the time that we were monitoring Databricks, there were two or three things that I think were most compelling.

22:38Obviously, the open source community was growing like crazy. And this team was the team to bet on since they built the open source community in the first place. If you were going to layer on an enterprise solution on top of the open source technology, their revenue was growing at an exponential rate. VCs talk about looking for companies like this all day long. And this was at 149 % growth rate at the time. They had incredibly strong usage. I think it was 40 % of their customers exceeded their memory quota for an early stage company. So that was really compelling. And I think it goes to the fact that it was saving so much money for people and it was having such an impact on performance for these businesses.

23:17And of course, they had thousands of enterprises. I'll give you a little sneaky piece of data that everyone probably forgets now, or maybe only the internals knew. But there was also an attempt to acquire them back then for a billion dollars. So it was a pretty good bet either way, because we knew where we would house them if anything went wrong. And how important is the partner ecosystem to the continued growth and trajectory of the business? And what are some of the key partnerships that have helped fuel that growth that's so attractive? So this was incredibly crucial to Databricks. So when we came in at Sinewave actually back then, one of the things we helped them with was not only building their public sector vertical, which is hundreds of millions of dollars for them and very sticky, but we also helped them build customer partnership channels.

24:01So as an example, there's a few there, one's IBM and others Deloitte, but I'll use Deloitte as the example. Deloitte has a partnership with Databricks. So it gives them a channel partnership approach. All of Deloitte's customers, by training various folks at Deloitte, all of Deloitte's customers can now deploy Databricks to make smart decisions. So it really increased their sales capacity, even as a small company, by partnering with a lot of these companies early on. The second partnership part of it is they opened up the model. If Databricks had stayed Spark only, I think it would not have had the potential or been as big as it is now.

24:37By allowing other kinds of analytical tools to be used on the same platform, they became the one -stop shop for everything. Why go to anywhere else when you could just come to Databricks and see everything? Why go to any airline's website when you could go to Kayak and see everything? Same situation where they became this democratized platform that everybody could just go to for their analytics. and it was effective with big customers. So they knew how to get in with enterprises, show them results and drive revenue. So that was the other key piece. And you see the third part of that, I think, is when they decided not only to be democratic in terms of analytical tools, but also to broaden the platform across the different pieces of an analytic structure at a company.

25:23So as we talked about the warehouse or the structured data that people have been using forever, what they now call Lakehouse, which integrates that warehouse with the streaming data, which Databricks is famous for having enabled at companies. And now the most important one, I would argue, for their growth going forward is the AI LLMs and all of the different models. So interestingly, open AI is not so open. You're not allowed to use their models for anything commercial. There's actually some very specific restrictions on it. Whereas Databricks allows all different kinds of models to be used on their platform for commercial execution and usage, allow you to integrate it with your proprietary data in a smart way.