theCUBE + NYSE Wired: Mixture of Experts Series | Aaron Katz, ClickHouse

Clips
More from theCUBE + NYSE Wired: Mixture of Experts Series

Aaron Katz

CEO & Co-Founder

Clickhouse

play_circle_outline Introduction of Aaron Katz, co-founder and CEO of ClickHouse.

play_circle_outline Exploring ClickHouse: A Key Open-Source Asset for AI Video Cloud Innovations and Its Origins

play_circle_outline Growing trends in business intelligence and AI integration.

play_circle_outline Update on ClickHouse's funding and investor diversification strategy.

Info

Aaron Katz, ClickHouse

Aaron Katz

CEO & Co-Founder Clickhouse

search

Dave Vellante

>> Hi, everybody. Welcome back to the New York Stock Exchange. We're here at the prestigious Buttonwood Podium, overlooking the options exchange, NYSE Wired and theCUBE's ongoing series of Mixture of Experts. Aaron Katz is here. He's the co-founder and CEO of ClickHouse. Aaron, thanks for coming in. It's good to see you face-to face.

Aaron Katz

>> Yeah, likewise. Thanks for the invitation.

Dave Vellante

>> So, you bet. So, we are a customer of ClickHouse, our AI video cloud, which we built many years ago. We now use ClickHouse to ingest all of our video, all of our data. I mean, it is our core data asset, and it's super fast and we love it.

Aaron Katz

>> We like to hear that.

Dave Vellante

>> Thanks for building a great product.

Aaron Katz

>> Yeah, you bet. You're not the only one.

Dave Vellante

>> I'm glad to hear that. So, why did you start the company?

Aaron Katz

>> Yeah. So, I started the company four years ago with my two co-founders, Alexey, the creator of ClickHouse and Yury, who is a technology executive who I met when he was running engineering at Netflix. And I had seen ClickHouse emerge. It was open-sourced in 2016 and was quickly adopted by thousands of companies around the world, from Deutsche Bank to Microsoft, to Uber and eBay, for a diverse set of use cases. And I thought if we could unleash the potential of the technology by forming a company around it, it would be a fun project to work on. And so, I partnered up with some venture capitalists to initially fund the company and we got things off the ground. So, we're a Silicon Valley headquartered, venture-backed startup. And fast-forward four years later, we now have close to 400 employees, several hundred million dollars on the balance sheet, several thousand customers on our managed service, ClickHouse Cloud. But it feels like we're just getting started.

Dave Vellante

>> Well, it's interesting, because you take the open-source model. And we've seen it ebb and flow. Obviously, Red Hat was very successful, but then you saw in the big data days, Cloudera, Hortonworks really struggled with it. And it wasn't until Databricks came along with the managed service concept. But there was a lot of pressure on the development teams, because they had to commit resources to open source, but they really didn't have a business model on the back end. You clearly do. I presume you're committing a lot to the project, but also you have a managed service, so you make it easier for customers to adopt. Can you explain that in some depth?

Aaron Katz

>> Sure. I mean, I'll start by saying open source is not for the faint of heart. And really, there's only a handful of independent successful open source companies which remain today. You've got several in the public markets and you mentioned Confluent. You also have Mongo and Elastic. You've got several later stage in the private markets, like Databricks, and Grafana, and now ClickHouse. But you saw other companies like HashiCorp, for example, get acquired or Red Hat get acquired by IBM. And so, it's a wonderful innovation model, but it can be a very tricky business model. You've got to really strike the right balance between what are you release into the open source and how do you engage with the community, versus what do you build that's proprietary that you can monetize.

Dave Vellante

>> Where does ClickHouse fit in the whole database? Database, 15 years ago, was such a boring market and then it just exploded with very special purpose databases and you've seen some converged databases coming in lately. Help us understand the picture of where ClickHouse fits in the competitive landscape.

Aaron Katz

>> Well, I think why the database market is so exciting and why the category is so big is because the use cases are so diverse. And it goes from fraud detection to sentiment analysis, to risk modeling. And people use ClickHouse for all of those, in addition to observability, which is a very common use case around analyzing logs, metrics and traces. People use ClickHouse for that, like OpenAI, and Anthropic, and Tesla.

Dave Vellante

>> Security is a use case as well? I think it would be great for that.

Aaron Katz

>> Yeah. We're the back end to a number of leading security and cyber companies.

Dave Vellante

>> I bet.

Aaron Katz

>> Sierra, and Exabeam, Huntress, Simplicity are all using ClickHouse on the back end.

Dave Vellante

>> Right, right. Okay. So, help us understand the differentiation. Why does the world need ClickHouse?

Aaron Katz

>> Well, it was originally designed to be the back end for web analytics. And you think about, what are the characteristics for a database where you've got thousands of concurrent users running hundreds of thousands, if not millions of queries over petabytes of data? So, it needs to be able to store that data very efficiently and in a cost-effective way. And then it needs to be able to provide lightning-fast query experiences, so sub-second latency over both petabytes of historical data and data that's streaming in. And there are very few technologies in the world that can satisfy those requirements. And we maintain a public open-source benchmarking site called ClickBench, where we talk about and the community maintains these performance characteristics.

Dave Vellante

>> Databases clearly aren't going away. And you might not like this, but we've noted in our research that the point of control is shifting to the governance catalog layer. Oracle. by the way, says you're out of your mind. Database is still where it's at. You may in fact agree with that. But how do you see the stack changing, is really my question. And AI is really shining a light on BI. It seems like there's disruption going on there. There's new layers emerging. We call it system of intelligence that are serving up agents. There's an agent control framework, so the whole software stack is reforming. How do you see ClickHouse's future in that software stack?

Aaron Katz

>> Well, a lot of these agentic workflows are just generating extraordinary amounts of data. If you think about a simple semantic search, whether it's through ChatGPT or any other interface, it could generate up to 50 SQL queries. And you think about the performance characteristics that are required there in terms of the interactivity, and the latency, and then the amount of data that that interaction is generating. And so, we're seeing a lot of tailwinds from AI, companies like LangChain, Weights & Biases, Vercel, Anthropic, OpenAI. Yesterday in New York, we had Modal Labs talk about how they're using ClickHouse, but they're not just AI native companies. We had Capital One present yesterday on their use case. And there are a lot of AI themes even in financial services today. And so, we're seeing that change, especially you mentioned business intelligence, where right now if I need to ask a question about how the company's performing, I don't go to a data analyst and say, "I need you to build a stack bar chart that shows me regional distribution of revenue." I go to a prompt. And we use our own MCP server, integrating with Anthropic's model that queries a ClickHouse cloud service. And I can ask this service, build me a bar chart. Or I can say, "Who are the top 10 customers on AWS or Google Cloud, or Microsoft Azure over the last two weeks in Europe? And how does that compare to the performance in our top markets in Asia?" And within seconds, I get not only just the answers to the questions that I ask, but a bunch of other insights into the business. So, I think the BI category especially is being disrupted at a very fast pace by these agentic workflows.

Dave Vellante

>> So, you're democratizing that as that example that you just gave. Are you able to do more sophisticated things, like what if pricing... changing some of the assumptions and predicting what's going to happen, because those are some of the things that the BI analysts would do. Is that function getting democratized as well?

Aaron Katz

>> Absolutely. I'll give you an example. We've got over 2,000 customers on our managed service, ClickHouse Cloud. And our business model is based off usage or consumption, typically a combination of storage and compute. And so, we can get very granular in terms of looking at how our customers are consuming our service to forecast out how much revenue should we be able to predict over a certain period of time. So, I can ask the same model, build me three scenarios, a pessimistic scenario, a realistic scenario, and an optimistic scenario, in terms of where we're going to finish the quarter. And it will go and look at all of the historical growth. It'll look at all of our customers, in terms of how they're consuming our service. And it'll build me three different outcomes that I can then take to my investors and my company and say, "Hey, we're predicting that on the low end we're going to land here and on the high end we're going to land here." And then I can go even one level further and say, what are some of the factors that can lead to a more favorable outcome for the business? And it can go and interpret how our customers are using our technology and let us know what we could be doing differently to optimize their consumption.

Dave Vellante

>> So, because you have a consumption model, which is I think the forward-thinking way to price this stuff, you went through the cost optimization days of the cloud. Did you see Jevons Paradox kick in? Because that was the premise that, well, look, we want to pass on savings to customers, whether it's using more efficient processing power or whatever it is. And that might've lowered revenue for a period of time, but ultimately the bet was is it going to pick back up. Have you seen that?

Aaron Katz

>> Well, absolutely. Even internally ourselves, we were using a best of breed observability solution internally that we recently migrated off of, because it was extremely expensive. And obviously, the concept of dogfooding is very important for a software company. And we said, we've got to use our own technology. We had made an acquisition earlier this year called HyperDX, which is a visualization layer that sits on top of ClickHouse. And so, we made that migration, we wrote a blog about it, and it saved us millions of dollars and gave us better performance. And it was rumored recently that OpenAI also migrated off of an observability solution to an open source stack, actually on ClickHouse. They presented recently at our user conference in San Francisco. And it was rumored that they saved nine figures based off that migration. So, we're seeing a lot of companies that... because these services can get very expensive, as you know. And with just the emergence of a lot of different alternatives, especially around open source, you can see tremendous cost savings while getting better performance.

Dave Vellante

>> Aaron, I like that you called it dogfooding, because it's not always pretty, so it's an honest answer.

Aaron Katz

>> Yeah, it's not. I mean, especially some of these services that have been in the market for a long time, they're highly curated. They really are tailored to the developer experience. And making that migration can be difficult. And there's a lot of change management that goes along with it, but such is the nature of these emerging technologies. And we're seeing the switching costs actually go down over time and the migration effort go down as well.

Dave Vellante

>> There's obviously a lot of pressure from the market on open table formats. Iceberg seems to have at least a lot of traction. How does that fit in with your strategy? How do you accommodate that? What are you seeing as patterns and trends as well?

Aaron Katz

>> Yeah. I mean Databricks is really on the forefront here in terms of supporting Iceberg and the Delta Lake. And so, we're investing heavily with integrating and supporting those open table formats, for example, and integrating with various catalogs, like the Unity Catalog or the Polaris Catalog. And so, we're seeing that ecosystem really open in terms of these table formats, and develop a system of standards that companies like ours can integrate with.

Dave Vellante

>> So, it's an interesting trend. You mentioned Unity and Polaris. And of course, there are proprietary governance catalogs as well. How do you see customers dealing with that as they move toward open table formats? Question I always ask customers, "Well, how are you going to govern it?" And they're like, "Well, we're still figuring that out." Is that the case that you see?

Aaron Katz

>> I mean, our approach is that we really want to have an integration for all of the most popular table formats, and catalogs, and ingestion frameworks, and visualization layers. And so, that's why we're investing so heavily in what we call ClickPipes, which is, how do you get data into a ClickHouse service? How do you transform that data and load that data, extract insights out of that data? If you want to use Tableau or Looker, or Grafana as your visualization layer, making sure that we support those standards.

Dave Vellante

>> How do you handle transactions? I mean, we saw Snowflake announce Unistore, which never saw the light of day. They had to buy Crunchy. We saw Neon get acquired for transaction. What's your play on transactions?

Aaron Katz

>> I talked about this when we started the company. We started to see this convergence of what are traditionally described as transactional databases, like Postgres or MySQL, or a document store, like MongoDB with an analytical database like ClickHouse or others, or Snowflake, for example. That's also a columnar database. And at the time, the Venn diagram there was very minimal overlap. And that has really increased over the last, let's call it 18 months. And you saw, as you mentioned, Snowflake's acquisition of Crunchy data and Databricks' acquisition of Neon. We acquired a company last year called PureDB, which enables what's called CDC or change data capture to replicate data or migrate data from a transactional database, like Postgres into ClickHouse. We're investing heavily in this, and I think the industry is more broadly. And you also saw the emergence of these specialized databases over the last five years, like a vector database, for example, like Pinecone or Weaviate, and they grew in popularity very quickly. And now what I'm hearing when I talk to customers and OpenAI is a great example, or users of ClickHouse, is that if they could satisfy the same requirements that historically they would need three or four specialized databases, like a transactional database, and an analytical database, and a document store, and a vector database. And now, one database like ClickHouse can satisfy all of those requirements and use cases, that really simplifies their architecture and lowers their costs, which is what they're looking for.

Dave Vellante

>> And you've certainly seen that virtually all the database companies have built a vector capability inside. You as well.

Aaron Katz

>> Yeah. Just yesterday, as I mentioned, we heard not only from Capital One and Modal Labs, but from Ramp here in New York about how they're using ClickHouse to store embeddings and support vector search.

Dave Vellante

>> So, give us the update on the company. Where are you guys at in your funding rounds and your aspirations? What can you share with us?

Aaron Katz

>> It's a pretty unique company in that sense, because when we started the company, we did it with $50 million, which is not your typical seed or series A round. And then we quickly followed it up with another $250 million. This was back in 2021, so kind of peak Zerp period.

Dave Vellante

>> Well done.

Aaron Katz

>> And we didn't have really a commercial product at the time. It took us a year and a half to design ClickHouse Cloud. That's now been in the market for two and a half years and is growing very quickly. And it supports workloads on AWS, Google Cloud, and Azure. And we've got a partnership in China with Alibaba to make sure that we support those customers in Asia that want to deploy on other cloud providers. And so, we launched the product and we've got now over 2,000 customers. We did a series C financing, which was led by Khosla Ventures in May of this year. That was a $350 million round. And we had participation from other investors, like IVP, and Bond, and Bessemer, and Battery, and others, and a lot of insider participation. And we just recently announced that we extended that round and brought in a great group of new investors, like Citibank, for example, and a couple prominent individuals, like Nico Rosberg, the Formula One champion, and Brock Purdy, and Christian McCaffrey, Kyle Juszczyk. So, I'm trying to build this really diverse investor base. We're an enterprise software, so it's not the most glamorous industry in the world. And so, if you can make it a little bit more interesting by getting a broader group of people helping you solve these problems, I find that to be effective.

Dave Vellante

>> I always like to ask entrepreneurs, and founders, and CEOs, when you decided to scale your go-to market, and how did you determine that you had product market fit? What were the signals? And then how did you scale your go-to market?

Aaron Katz

>> Well, this is the beauty of open source, is that we had thousands of companies already using the software to help us design a managed service. In terms of if you were to go from self-managing open source to a cloud service, what are the characteristics of that cloud service? And it was really around price and performance. And so, it took us a year and a half to design the system to enable what's called the separation of compute and storage, which you and I were talking about earlier today. And what does that mean? It means your data's stored in object storage, so something like S3 or GCS or Azure Blob storage, and then your compute nodes are stateless. And that enables auto-scaling and idling, which means you only pay for what you use. So, the performance is significantly faster, but at a lower price. And those are really the two characteristics of a database that people are looking for. And so, that was the original design behind the service when we launched it two and a half years ago. And we've been able to have a lot of the AI companies that I mentioned deploy it. You can run it in our managed service. We have what's called bring your own cloud, where the data plane is decoupled from the control plane. That can run in your own VPC. We've got a private binary that you can run yourselves, like Tesla does. Tesla now ingests a billion events per second into ClickHouse. All observability data is being ingested into ClickHouse. They've got a table with a quadrillion rows. And so, there's very few databases that can support that type of load.

Dave Vellante

>> Okay. And you do this on object. You don't need block to get the performance. And I'm sure that's why our developers in part anyway chose it, because it's much more efficient from a cost standpoint.

Aaron Katz

>> It's absolutely more efficient. But what you also want is you want the performance of SSDs. And that's the beauty of ClickHouse, where you can get the cost benefits of object storage, but the performance as if it's running on local disks.

Dave Vellante

>> And then so, aspirations for the company. Where do you see it going? What's the vision?

Aaron Katz

>> I was very fortunate to join Salesforce when it was a small startup in '02. It went public in '04. I spent 10 more years there. And then I joined Elasticsearch when it was a small startup. And we took it public here at the New York Stock Exchange six years ago, seven years ago in October of 2018. And so, I think it's healthy for companies to enter the public markets. I'll start there. And so, I know it's in vogue for companies to stay private longer, and I understand the reasons as to why they do that. They have this asymmetric advantage, where they can have access to a lot of private capital, grow their company very aggressively, and not deal with the constraints of being a public company. I'm a little more traditional in the sense that I think a technology company, when it reaches a certain scale, and has a lot of predictability in the business, and a lot of all of the unit economics are strong, and you can predict where you're going to land the plane for the next few years, that you should go public. And it's part of the deal that we sign up for, is my belief. But I'm biased, obviously, with these two great experiences of taking these companies public. In the case of Elastic, where I was an officer, in the case of Salesforce, where I was a mid-level manager at the time. And so, that's the objective here. This market opportunity is enormous. Our revenue's growing extremely quickly. And you asked a question about when we were thinking about the company, what was kind of the distribution model that we were thinking about? And I looked at two very successful companies at the time, Datadog and Snowflake, and in infrastructure as a service. And where I sat, they approached go to market a little bit differently. Datadog was very developer-led, self-service, PLG, whatever you want to call it these days. And Snowflake I think was a little bit more traditional, in terms of investing heavily in sales, and marketing, and going after the enterprise. And they were both successful with those strategies, but I thought it was going to be a lot easier if we follow the Datadog approach, where somebody can come, experience the service, load their data, run their queries, add their credit card, and they don't ever have to talk to anybody in sales, for example. And that's proven to be the case. We add over a 100 customers a month to our managed service. Last month, we added over 200 customers to our managed service. The vast majority of those never talk to anybody in sales. It's this frictionless experience. And then when they're ready and they say, "Hey, we want to make a longer term commitment, this is more strategic, and we want to have a multi-year agreement, we need some discounts, we need some SLAs that you offer to your committed spend customers," we engage with them in that sales dialogue.

Dave Vellante

>> Well, and it means that you're more efficient with your capital, and you can put more into R&D, and invest there, versus having to charge more to pay for your sales and marketing.

Aaron Katz

>> I think that's the right way to look at it. We're a very efficient company. Our burn multiple is... Some people would actually say I'm probably too efficient and I need to invest more heavily in sales and marketing, which we're doing. We're going to double the sales capacity over the next six months from where we were. But I really didn't want just a bunch of salespeople to be a crutch for product quality. I've seen that in my career. And I really wanted the pressure to be on engineering and product to deliver a service that delivers real value without a salesperson needing to foster that evaluation.

Dave Vellante

>> And on your IPO comments, we're the same. We're biased because the retail investor can participate in IPOs. It's harder for them to participate in private markets. They get edged out or they have to pay massive premiums. So, we love public markets. So, hopefully, we can cover you ringing the bell here. I really appreciate your time.

Aaron Katz

>> All right. Well, I hope we can get to that stage. Thanks for having me.

Dave Vellante

>> Thank you so much for your time. And thank you for watching our Mixture of Experts Series. This is Dave Vellante for NYSE Wired in theCUBE. We'll be right back, right after the short break.

Aaron Katz, ClickHouse

search

Dave Vellante

>> Hi, everybody. Welcome back to the New York Stock Exchange. We're here at the prestigious Buttonwood Podium, overlooking the options exchange, NYSE Wired and theCUBE's ongoing series of Mixture of Experts. Aaron Katz is here. He's the co-founder and CEO of ClickHouse. Aaron, thanks for coming in. It's good to see you face-to face.

Aaron Katz

>> Yeah, likewise. Thanks for the invitation.

Dave Vellante

>> So, you bet. So, we are a customer of ClickHouse, our AI video cloud, which we built many years ago. We now use ClickHouse to ingest all of our video, all of our data. I mean, it is our core data asset, and it's super fast and we love it.

Aaron Katz

>> We like to hear that.

Dave Vellante

>> Thanks for building a great product.

Aaron Katz

>> Yeah, you bet. You're not the only one.

Dave Vellante

>> I'm glad to hear that. So, why did you start the company?

Aaron Katz

>> Yeah. So, I started the company four years ago with my two co-founders, Alexey, the creator of ClickHouse and Yury, who is a technology executive who I met when he was running engineering at Netflix. And I had seen ClickHouse emerge. It was open-sourced in 2016 and was quickly adopted by thousands of companies around the world, from Deutsche Bank to Microsoft, to Uber and eBay, for a diverse set of use cases. And I thought if we could unleash the potential of the technology by forming a company around it, it would be a fun project to work on. And so, I partnered up with some venture capitalists to initially fund the company and we got things off the ground. So, we're a Silicon Valley headquartered, venture-backed startup. And fast-forward four years later, we now have close to 400 employees, several hundred million dollars on the balance sheet, several thousand customers on our managed service, ClickHouse Cloud. But it feels like we're just getting started.

Dave Vellante

>> Well, it's interesting, because you take the open-source model. And we've seen it ebb and flow. Obviously, Red Hat was very successful, but then you saw in the big data days, Cloudera, Hortonworks really struggled with it. And it wasn't until Databricks came along with the managed service concept. But there was a lot of pressure on the development teams, because they had to commit resources to open source, but they really didn't have a business model on the back end. You clearly do. I presume you're committing a lot to the project, but also you have a managed service, so you make it easier for customers to adopt. Can you explain that in some depth?

Aaron Katz

>> Sure. I mean, I'll start by saying open source is not for the faint of heart. And really, there's only a handful of independent successful open source companies which remain today. You've got several in the public markets and you mentioned Confluent. You also have Mongo and Elastic. You've got several later stage in the private markets, like Databricks, and Grafana, and now ClickHouse. But you saw other companies like HashiCorp, for example, get acquired or Red Hat get acquired by IBM. And so, it's a wonderful innovation model, but it can be a very tricky business model. You've got to really strike the right balance between what are you release into the open source and how do you engage with the community, versus what do you build that's proprietary that you can monetize.

Dave Vellante

>> Where does ClickHouse fit in the whole database? Database, 15 years ago, was such a boring market and then it just exploded with very special purpose databases and you've seen some converged databases coming in lately. Help us understand the picture of where ClickHouse fits in the competitive landscape.

Aaron Katz

>> Well, I think why the database market is so exciting and why the category is so big is because the use cases are so diverse. And it goes from fraud detection to sentiment analysis, to risk modeling. And people use ClickHouse for all of those, in addition to observability, which is a very common use case around analyzing logs, metrics and traces. People use ClickHouse for that, like OpenAI, and Anthropic, and Tesla.

Dave Vellante

>> Security is a use case as well? I think it would be great for that.

Aaron Katz

>> Yeah. We're the back end to a number of leading security and cyber companies.

Dave Vellante

>> I bet.

Aaron Katz

>> Sierra, and Exabeam, Huntress, Simplicity are all using ClickHouse on the back end.

Dave Vellante

>> Right, right. Okay. So, help us understand the differentiation. Why does the world need ClickHouse?

Aaron Katz

>> Well, it was originally designed to be the back end for web analytics. And you think about, what are the characteristics for a database where you've got thousands of concurrent users running hundreds of thousands, if not millions of queries over petabytes of data? So, it needs to be able to store that data very efficiently and in a cost-effective way. And then it needs to be able to provide lightning-fast query experiences, so sub-second latency over both petabytes of historical data and data that's streaming in. And there are very few technologies in the world that can satisfy those requirements. And we maintain a public open-source benchmarking site called ClickBench, where we talk about and the community maintains these performance characteristics.

Dave Vellante

>> Databases clearly aren't going away. And you might not like this, but we've noted in our research that the point of control is shifting to the governance catalog layer. Oracle. by the way, says you're out of your mind. Database is still where it's at. You may in fact agree with that. But how do you see the stack changing, is really my question. And AI is really shining a light on BI. It seems like there's disruption going on there. There's new layers emerging. We call it system of intelligence that are serving up agents. There's an agent control framework, so the whole software stack is reforming. How do you see ClickHouse's future in that software stack?

Aaron Katz

>> Well, a lot of these agentic workflows are just generating extraordinary amounts of data. If you think about a simple semantic search, whether it's through ChatGPT or any other interface, it could generate up to 50 SQL queries. And you think about the performance characteristics that are required there in terms of the interactivity, and the latency, and then the amount of data that that interaction is generating. And so, we're seeing a lot of tailwinds from AI, companies like LangChain, Weights & Biases, Vercel, Anthropic, OpenAI. Yesterday in New York, we had Modal Labs talk about how they're using ClickHouse, but they're not just AI native companies. We had Capital One present yesterday on their use case. And there are a lot of AI themes even in financial services today. And so, we're seeing that change, especially you mentioned business intelligence, where right now if I need to ask a question about how the company's performing, I don't go to a data analyst and say, "I need you to build a stack bar chart that shows me regional distribution of revenue." I go to a prompt. And we use our own MCP server, integrating with Anthropic's model that queries a ClickHouse cloud service. And I can ask this service, build me a bar chart. Or I can say, "Who are the top 10 customers on AWS or Google Cloud, or Microsoft Azure over the last two weeks in Europe? And how does that compare to the performance in our top markets in Asia?" And within seconds, I get not only just the answers to the questions that I ask, but a bunch of other insights into the business. So, I think the BI category especially is being disrupted at a very fast pace by these agentic workflows.

Dave Vellante

>> So, you're democratizing that as that example that you just gave. Are you able to do more sophisticated things, like what if pricing... changing some of the assumptions and predicting what's going to happen, because those are some of the things that the BI analysts would do. Is that function getting democratized as well?

Aaron Katz

>> Absolutely. I'll give you an example. We've got over 2,000 customers on our managed service, ClickHouse Cloud. And our business model is based off usage or consumption, typically a combination of storage and compute. And so, we can get very granular in terms of looking at how our customers are consuming our service to forecast out how much revenue should we be able to predict over a certain period of time. So, I can ask the same model, build me three scenarios, a pessimistic scenario, a realistic scenario, and an optimistic scenario, in terms of where we're going to finish the quarter. And it will go and look at all of the historical growth. It'll look at all of our customers, in terms of how they're consuming our service. And it'll build me three different outcomes that I can then take to my investors and my company and say, "Hey, we're predicting that on the low end we're going to land here and on the high end we're going to land here." And then I can go even one level further and say, what are some of the factors that can lead to a more favorable outcome for the business? And it can go and interpret how our customers are using our technology and let us know what we could be doing differently to optimize their consumption.

Dave Vellante

>> So, because you have a consumption model, which is I think the forward-thinking way to price this stuff, you went through the cost optimization days of the cloud. Did you see Jevons Paradox kick in? Because that was the premise that, well, look, we want to pass on savings to customers, whether it's using more efficient processing power or whatever it is. And that might've lowered revenue for a period of time, but ultimately the bet was is it going to pick back up. Have you seen that?

Aaron Katz

>> Well, absolutely. Even internally ourselves, we were using a best of breed observability solution internally that we recently migrated off of, because it was extremely expensive. And obviously, the concept of dogfooding is very important for a software company. And we said, we've got to use our own technology. We had made an acquisition earlier this year called HyperDX, which is a visualization layer that sits on top of ClickHouse. And so, we made that migration, we wrote a blog about it, and it saved us millions of dollars and gave us better performance. And it was rumored recently that OpenAI also migrated off of an observability solution to an open source stack, actually on ClickHouse. They presented recently at our user conference in San Francisco. And it was rumored that they saved nine figures based off that migration. So, we're seeing a lot of companies that... because these services can get very expensive, as you know. And with just the emergence of a lot of different alternatives, especially around open source, you can see tremendous cost savings while getting better performance.

Dave Vellante

>> Aaron, I like that you called it dogfooding, because it's not always pretty, so it's an honest answer.

Aaron Katz

>> Yeah, it's not. I mean, especially some of these services that have been in the market for a long time, they're highly curated. They really are tailored to the developer experience. And making that migration can be difficult. And there's a lot of change management that goes along with it, but such is the nature of these emerging technologies. And we're seeing the switching costs actually go down over time and the migration effort go down as well.

Dave Vellante

>> There's obviously a lot of pressure from the market on open table formats. Iceberg seems to have at least a lot of traction. How does that fit in with your strategy? How do you accommodate that? What are you seeing as patterns and trends as well?

Aaron Katz

>> Yeah. I mean Databricks is really on the forefront here in terms of supporting Iceberg and the Delta Lake. And so, we're investing heavily with integrating and supporting those open table formats, for example, and integrating with various catalogs, like the Unity Catalog or the Polaris Catalog. And so, we're seeing that ecosystem really open in terms of these table formats, and develop a system of standards that companies like ours can integrate with.

Dave Vellante

>> So, it's an interesting trend. You mentioned Unity and Polaris. And of course, there are proprietary governance catalogs as well. How do you see customers dealing with that as they move toward open table formats? Question I always ask customers, "Well, how are you going to govern it?" And they're like, "Well, we're still figuring that out." Is that the case that you see?

Aaron Katz

>> I mean, our approach is that we really want to have an integration for all of the most popular table formats, and catalogs, and ingestion frameworks, and visualization layers. And so, that's why we're investing so heavily in what we call ClickPipes, which is, how do you get data into a ClickHouse service? How do you transform that data and load that data, extract insights out of that data? If you want to use Tableau or Looker, or Grafana as your visualization layer, making sure that we support those standards.

Dave Vellante

>> How do you handle transactions? I mean, we saw Snowflake announce Unistore, which never saw the light of day. They had to buy Crunchy. We saw Neon get acquired for transaction. What's your play on transactions?

Aaron Katz

>> I talked about this when we started the company. We started to see this convergence of what are traditionally described as transactional databases, like Postgres or MySQL, or a document store, like MongoDB with an analytical database like ClickHouse or others, or Snowflake, for example. That's also a columnar database. And at the time, the Venn diagram there was very minimal overlap. And that has really increased over the last, let's call it 18 months. And you saw, as you mentioned, Snowflake's acquisition of Crunchy data and Databricks' acquisition of Neon. We acquired a company last year called PureDB, which enables what's called CDC or change data capture to replicate data or migrate data from a transactional database, like Postgres into ClickHouse. We're investing heavily in this, and I think the industry is more broadly. And you also saw the emergence of these specialized databases over the last five years, like a vector database, for example, like Pinecone or Weaviate, and they grew in popularity very quickly. And now what I'm hearing when I talk to customers and OpenAI is a great example, or users of ClickHouse, is that if they could satisfy the same requirements that historically they would need three or four specialized databases, like a transactional database, and an analytical database, and a document store, and a vector database. And now, one database like ClickHouse can satisfy all of those requirements and use cases, that really simplifies their architecture and lowers their costs, which is what they're looking for.

Dave Vellante

>> And you've certainly seen that virtually all the database companies have built a vector capability inside. You as well.

Aaron Katz

>> Yeah. Just yesterday, as I mentioned, we heard not only from Capital One and Modal Labs, but from Ramp here in New York about how they're using ClickHouse to store embeddings and support vector search.

Dave Vellante

>> So, give us the update on the company. Where are you guys at in your funding rounds and your aspirations? What can you share with us?

Aaron Katz

>> It's a pretty unique company in that sense, because when we started the company, we did it with $50 million, which is not your typical seed or series A round. And then we quickly followed it up with another $250 million. This was back in 2021, so kind of peak Zerp period.

Dave Vellante

>> Well done.

Aaron Katz

>> And we didn't have really a commercial product at the time. It took us a year and a half to design ClickHouse Cloud. That's now been in the market for two and a half years and is growing very quickly. And it supports workloads on AWS, Google Cloud, and Azure. And we've got a partnership in China with Alibaba to make sure that we support those customers in Asia that want to deploy on other cloud providers. And so, we launched the product and we've got now over 2,000 customers. We did a series C financing, which was led by Khosla Ventures in May of this year. That was a $350 million round. And we had participation from other investors, like IVP, and Bond, and Bessemer, and Battery, and others, and a lot of insider participation. And we just recently announced that we extended that round and brought in a great group of new investors, like Citibank, for example, and a couple prominent individuals, like Nico Rosberg, the Formula One champion, and Brock Purdy, and Christian McCaffrey, Kyle Juszczyk. So, I'm trying to build this really diverse investor base. We're an enterprise software, so it's not the most glamorous industry in the world. And so, if you can make it a little bit more interesting by getting a broader group of people helping you solve these problems, I find that to be effective.

Dave Vellante

>> I always like to ask entrepreneurs, and founders, and CEOs, when you decided to scale your go-to market, and how did you determine that you had product market fit? What were the signals? And then how did you scale your go-to market?

Aaron Katz

>> Well, this is the beauty of open source, is that we had thousands of companies already using the software to help us design a managed service. In terms of if you were to go from self-managing open source to a cloud service, what are the characteristics of that cloud service? And it was really around price and performance. And so, it took us a year and a half to design the system to enable what's called the separation of compute and storage, which you and I were talking about earlier today. And what does that mean? It means your data's stored in object storage, so something like S3 or GCS or Azure Blob storage, and then your compute nodes are stateless. And that enables auto-scaling and idling, which means you only pay for what you use. So, the performance is significantly faster, but at a lower price. And those are really the two characteristics of a database that people are looking for. And so, that was the original design behind the service when we launched it two and a half years ago. And we've been able to have a lot of the AI companies that I mentioned deploy it. You can run it in our managed service. We have what's called bring your own cloud, where the data plane is decoupled from the control plane. That can run in your own VPC. We've got a private binary that you can run yourselves, like Tesla does. Tesla now ingests a billion events per second into ClickHouse. All observability data is being ingested into ClickHouse. They've got a table with a quadrillion rows. And so, there's very few databases that can support that type of load.

Dave Vellante

>> Okay. And you do this on object. You don't need block to get the performance. And I'm sure that's why our developers in part anyway chose it, because it's much more efficient from a cost standpoint.

Aaron Katz

>> It's absolutely more efficient. But what you also want is you want the performance of SSDs. And that's the beauty of ClickHouse, where you can get the cost benefits of object storage, but the performance as if it's running on local disks.

Dave Vellante

>> And then so, aspirations for the company. Where do you see it going? What's the vision?

Aaron Katz

>> I was very fortunate to join Salesforce when it was a small startup in '02. It went public in '04. I spent 10 more years there. And then I joined Elasticsearch when it was a small startup. And we took it public here at the New York Stock Exchange six years ago, seven years ago in October of 2018. And so, I think it's healthy for companies to enter the public markets. I'll start there. And so, I know it's in vogue for companies to stay private longer, and I understand the reasons as to why they do that. They have this asymmetric advantage, where they can have access to a lot of private capital, grow their company very aggressively, and not deal with the constraints of being a public company. I'm a little more traditional in the sense that I think a technology company, when it reaches a certain scale, and has a lot of predictability in the business, and a lot of all of the unit economics are strong, and you can predict where you're going to land the plane for the next few years, that you should go public. And it's part of the deal that we sign up for, is my belief. But I'm biased, obviously, with these two great experiences of taking these companies public. In the case of Elastic, where I was an officer, in the case of Salesforce, where I was a mid-level manager at the time. And so, that's the objective here. This market opportunity is enormous. Our revenue's growing extremely quickly. And you asked a question about when we were thinking about the company, what was kind of the distribution model that we were thinking about? And I looked at two very successful companies at the time, Datadog and Snowflake, and in infrastructure as a service. And where I sat, they approached go to market a little bit differently. Datadog was very developer-led, self-service, PLG, whatever you want to call it these days. And Snowflake I think was a little bit more traditional, in terms of investing heavily in sales, and marketing, and going after the enterprise. And they were both successful with those strategies, but I thought it was going to be a lot easier if we follow the Datadog approach, where somebody can come, experience the service, load their data, run their queries, add their credit card, and they don't ever have to talk to anybody in sales, for example. And that's proven to be the case. We add over a 100 customers a month to our managed service. Last month, we added over 200 customers to our managed service. The vast majority of those never talk to anybody in sales. It's this frictionless experience. And then when they're ready and they say, "Hey, we want to make a longer term commitment, this is more strategic, and we want to have a multi-year agreement, we need some discounts, we need some SLAs that you offer to your committed spend customers," we engage with them in that sales dialogue.

Dave Vellante

>> Well, and it means that you're more efficient with your capital, and you can put more into R&D, and invest there, versus having to charge more to pay for your sales and marketing.

Aaron Katz

>> I think that's the right way to look at it. We're a very efficient company. Our burn multiple is... Some people would actually say I'm probably too efficient and I need to invest more heavily in sales and marketing, which we're doing. We're going to double the sales capacity over the next six months from where we were. But I really didn't want just a bunch of salespeople to be a crutch for product quality. I've seen that in my career. And I really wanted the pressure to be on engineering and product to deliver a service that delivers real value without a salesperson needing to foster that evaluation.

Dave Vellante

>> And on your IPO comments, we're the same. We're biased because the retail investor can participate in IPOs. It's harder for them to participate in private markets. They get edged out or they have to pay massive premiums. So, we love public markets. So, hopefully, we can cover you ringing the bell here. I really appreciate your time.

Aaron Katz

>> All right. Well, I hope we can get to that stage. Thanks for having me.

Dave Vellante

>> Thank you so much for your time. And thank you for watching our Mixture of Experts Series. This is Dave Vellante for NYSE Wired in theCUBE. We'll be right back, right after the short break.