为什么 Anthropic、Meta 和 Tesla 都选择了同一个数据库｜Aaron Katz，ClickHouse

All right. Well, Aaron, when we first met, you told me the story of ClickHouse and how you got involved. I remember it was one of the most fascinating stories of the genesis of any startup. Tell me the whole story.

I remember you and I first met down in Santa Barbara, and we went on a walk along the beach. We told each other our respective founding stories. It feels like that was yesterday, but I think it was probably 4 years ago.

ClickHouse was developed by one of my co-founders, Alexey Milovidov, when he was working at Yandex. He was looking for a database that would be able to ingest and store petabytes of streaming data from Yandex’s various services for web analytics. Nothing off the shelf was able to scale and provide the performance that was needed, so he developed ClickHouse.

It’s short for clickstream data warehouse. He was thinking about the data warehouse use case back in 2009 when he developed the database. It took off inside Yandex for a variety of different use cases, and he had the courage to convince the Yandex leadership team to open-source it in 2016.

Seven years later, they had formed a team around ClickHouse. It was one of the fastest-growing databases after it was open-sourced in terms of the number of contributors and the number of contributions, and it still is.

That’s when I first discovered it. When I was at a previous company, we started to bump into ClickHouse in the market. We actually started to lose a lot of observability workloads specifically to ClickHouse. Uber was migrating its logging infrastructure to ClickHouse. We lost a big metrics project at eBay. Disney, Comcast, Deutsche Bank, and others were moving to it.

It was on my radar, but there were a lot of popular open-source OLAP engines at the time. Again, this was 10 years ago. During COVID, I stepped out and started thinking about what I was going to do next, and an investor asked if I’d be interested in talking to the Yandex leadership team about ClickHouse. I jumped on it.

This was early 2021. In the thick of COVID, everything was done virtually at the time. I met the Yandex leadership team and Alexey, and they said, “We realize that ClickHouse is very popular, and we think that it’s got a lot of potential. We’d be curious to know if you were interested in doing something with it.”

1. Building ClickHouse Cloud & Raising $300M

I said, “Absolutely. I would love to partner with Alexey and form a company around it. It’s going to have to look and smell like a Silicon Valley startup, with a venture-backed Delaware corporation. I think we could really unleash the potential of this technology by building a fully managed service—a serverless offering with compute-storage separation that could automatically scale and idle, because these analytical workloads are very bursty.

“We could take a lot of these companies that are running this technology themselves, either in the cloud or on-premises, and give them all of these amazing benefits of a hosted offering called ClickHouse Cloud.”

I reached out to 2 venture capitalists I’d known in the industry: Mike Volpi, who at the time was at Index Ventures, and Peter Fenton at Benchmark. In the world of open-source infrastructure, Lukas, as you know better than anyone, there’s really a handful of investors who you’d want to partner with—people who have experience with these business models and understand all of the nuances around open source.

It took us about 9 months to pull off because, at the time, Yandex was a publicly traded company with, I think, a $30 billion market cap. When you take intellectual property out of a public company, it comes with a lot of regulatory concern and consideration. There were a lot of lawyers involved, as you could imagine.

I convinced an executive at Google named Yuri Izrailevsky. I’d met Yuri when he was at Netflix, where he ran platform engineering. At the time, they were AWS’s largest customer, I believe, and he really led their migration to the cloud. He’d been building distributed systems around open source for 20 years.

I convinced him, along with others, to leave Google and join Alexey and me to form the company. We had 3 co-founders: Yuri running product and engineering, Alexey as the creator of ClickHouse and our chief technology officer, and myself as the CEO.

We got it done in August 2021 with a $50 million seed round. Maybe it was even a pre-seed round—possibly the largest pre-seed round in the history of enterprise software. I didn’t have a pitch deck. We had no product, no customers, and no revenue. But it was the peak ZIRP period, and there was a lot of capital getting put to work.

We quickly followed it up with another $250 million round that was jointly led by Coatue, where you and I met, and Altimeter. There’s a lot more to the story, but that’s the origin story of ClickHouse Inc., the company. You’re looking at the headquarters here in Silicon Valley, and Alexey and his team are in Amsterdam. We’ve got a really strong engineering hub in the Netherlands.

We have over 3,000 customers on ClickHouse Cloud, we add hundreds every month.

Where are we 5 years later from when we formed the company? When you were doing this initial pre-seed round, which was absolutely enormous at that time, people might not realize what an outlier it was. Pre-seeds have gotten astonishingly big in some AI sectors, but when that was happening, what gave you and your investors the confidence to do that if there was no revenue and not even a product launched yet?

We had thousands of companies using the open-source ClickHouse distribution for a wide array of use cases, from analyzing clickstream data to data warehousing to observability and as a cyber backend. I think Microsoft is on record that some of its largest analytical workloads run on ClickHouse, like Microsoft Titan and Microsoft Clarity. I mentioned some of the other companies that were migrating to ClickHouse.

We had a lot of conviction that the technology itself was highly differentiated to start with. I feel like we had a bit of a head start, frankly, compared with most startups, because we had this very feature-rich database and the core team of committers.

As you know, in open source, you can have hundreds or thousands of contributors, but to really control the roadmap and the direction of the project, you need to employ the committers. We’ve seen in the past what happens if an open-source company does not do that.

We felt like we had the right team: Yuri, having had such success building distributed systems around open-source databases; myself on the distribution side, having studied go-to-market for more than 20 years at companies like Salesforce and others. I felt like we had the right team assembled, and I think the investors agreed with that thesis.

We had a great group of companies as design partners early on. We had 50 companies that were using open-source ClickHouse saying, “These are the characteristics of a cloud service that would compel us to move from self-managing the database to a fully hosted service.”

We had a lot of input from early adopters of ClickHouse, and we saw a ton of market pull. As you know, because we cover such a broad surface area, there are a lot of competitive dynamics that we need to navigate.

On the data warehousing side, you’ve got Snowflake, Redshift, BigQuery, and so on. On observability, it’s a very crowded market, as you know. But ClickHouse really shines in terms of low-latency analytics that are customer-facing, like the Weights & Biases Weave product, which is built on ClickHouse Cloud.

And so, we felt like we had a huge TAM to go after. We had the right team to develop a differentiated service, and many of us had been part of open-source companies in the past. So, we had the unfortunate experience of knowing what to avoid, because getting it right with open source, as you know, is very tricky. Only a handful of companies have been able to achieve it. And that's why we prioritized this cloud service before everything else.

2. Growing Up Around Xerox PARC

I actually wanted to ask you. I saw that your father worked at Xerox PARC in the really early days of Silicon Valley, I believe. I wonder how growing up in that environment might have shaped your perspective on technology.

Well, he worked initially at Xerox out of college in the '60s, selling copy machines in the Bay Area. I think his largest customer was Stanford University. This was when Xerox and IBM were widely recognized as the most innovative technology companies in the world at the time.

He then went to Xerox Document Systems, which was co-located at PARC. For those listening, PARC stands for Palo Alto Research Center, and it's where a lot of the innovations that we use every day were created and invented. Things like the graphical user interface and the concept of the mouse. It was just revolutionary.

So, I was a kid, Lukas. Honestly, I just turned 50 yesterday. I was born in '76, so we're talking about the mid-'80s, when my dad was working at Xerox Document Systems—maybe the late '80s. Obviously, I wasn't paying too much attention to it all.

I think there's the benefit of just osmosis and being around it. But I do remember going to PARC with my dad and sitting in his office, doing homework, reading a book, doing LEGOs, or God knows what. I think just being around it and seeing the energy was very appealing to me.

So, before I even started college, when I went to UC Davis, it was clear to me that I was going to go into technology. I grew up around it. I saw the lifestyle that it provided. I saw the joy that my dad got by building and leading teams, securing a customer win, or helping a customer solve a problem.

I don't know. For me, it was just the same way that if you grow up and your mother or your father is a doctor, there's probably a decent chance that may be a path you pursue because you're just around it. You live in it.

3. Salesforce, Mark Benioff & the Dot-Com Bust

Totally. And then you went to Salesforce, I think, in the first dot-com bubble era. I was wondering if that shaped you, and if seeing that big bubble gives you a perspective on the possible AI bubble that we're in now.

Of course it shaped me. It was transformational. The 12 years I spent at Salesforce were so pivotal in my life for so many different reasons.

The other reason I went to Salesforce back in early 2002—24 years ago—was because I didn't get into business school. The plan was to go get my MBA. Coming out of the dot-com bust in '01 and early '02, I applied to a handful of what many would perceive to be elite business schools: Harvard, Stanford, and so on. I didn't get into any of them. I got waitlisted at MIT Sloan.

So, I was unemployed because the startup that I was working at had failed, and I was living up in Lake Tahoe, waiting tables. I needed a job, frankly. So, I came back to San Francisco, and I was able to navigate the interview process to meet with Marc Benioff.

I remember that interview vividly because he asked me one question. He said, “Why should I hire you?” Somehow, I convinced him that it was a good decision. At the time, it was a 150-person startup.

How did you answer that question? I've got to know.

I said, “Because I'm going to outwork and be better than everyone else, and I'm going to be your top-performing sales engineer.” I was a sales engineer, which now people call solution architects. Maybe they call them forward-deployed engineers. I don't know, but you get it: technical pre-sales.

I said, “I'm going to work harder, I'm going to be smarter, and I'm going to out-execute my peers.” I knew that he was very competitive because he was a disciple of Larry Ellison at Oracle. So, I knew that he would respond favorably to somebody who had a competitive mindset.

I'd gone through the gauntlet of interviews to get to him, so I suspect I wouldn't have gotten to him if the team didn't think I was going to be a good hire.

Nevertheless, how does that correlate to the AI bubble that we're in today? I do think there are a lot of parallels.

Salesforce—software as a service was not a new thing. It had been tried and failed during the dot-com bust. But Salesforce was the one that made it a reality, created the category, and spawned so many other companies: ServiceNow, Workday, and we could go on and on.

4. Cloud Skeptics vs. AI Skeptics | History Repeating

At the time, if you recall, our competitors—Siebel, SAP, Oracle, and Microsoft—were basically on record saying, “Cloud computing is a fad. No one's going to move their customer data to the cloud. It's not secure. It's not reliable. It's not durable.”

We were on this mission to create this category and refute all of these baseless claims that it wasn't going to work, that all software was going to be run in data centers on-premises.

If you apply that to the current AI themes, people say, “We're never going to realize superintelligence. AI models are unpredictable. You cannot trust them. You don't want to send your enterprise data to an LLM provider because it's your competitive edge.” All these things are going to be debunked, if they haven't been already.

We're all going to be using AI the same way that we use cloud services every day. Even in large, regulated industries, they're still using cloud services in their companies.

I see AI as no different from the early skeptics or incumbents who have their heads in the sand. They just don't want to accept reality—the same way that Siebel didn't want to accept the reality that their core business was, frankly, in decline. They ultimately got acquired by Oracle.

It was a good company. Salesforce replicated a lot of the Siebel feature set from a CRM standpoint. But there was this paralysis that set in, where they were unwilling to accept that cloud computing was going to be the way enterprise software was delivered in the future.

I think you can either accept the reality that we're all going to be using AI models much more than we are today, in a much shorter period of time. It's not going to take 20 years. It took us—I joined in '02, and the company went public in '04. We still only had a handful of enterprise customers when Salesforce went public.

AI is getting adopted 100 times faster than software as a service was 20 years ago.

5. Building a Modern Go-To-Market Playbook

When you were creating the go-to-market motion for ClickHouse, it seems like you built it in a really modern way. One of the things that I really admired about ClickHouse is that when we bought it at Weights & Biases, it felt like you met us where we were at. It felt like a very technical sale to a technical audience.

There wasn't a lot of pomp and circumstance. Nobody was asking us to go play golf or get dinner. It was just fixing our problems, getting us up and running, waiting to charge us until we were up and running, and being incredibly effective at giving us timely technical support. I really admired that. I went to my team and said, “We have to learn from this ClickHouse team.”

It's interesting to learn about your background. You have this—you’re probably deeper in Silicon Valley go-to-market motions than anyone. Your dad was doing it in the '80s. What do you think you took from that early experience with Salesforce and before that? And what did you leave behind as you were building out the ClickHouse go-to-market team and motion?

Well, when I was starting the company, as I mentioned, the strategy was going to be to take this very popular open-source database and build a managed service. I looked at the other open-source companies at the time that had gone down this path. From where I sat, MongoDB was the best of the bunch with its Atlas service. It was a broad database platform that satisfied a lot of different use cases, so that was the first design principle.

But that doesn't mean that you're going to be successful in building a service that distributes itself to a highly technical audience. So, I looked at the next 2 proprietary infrastructure companies at the time, Datadog and Snowflake, and how they went to market.

They were very different. They were both very successful, but through different means. Datadog built an offering that a developer could evaluate, deploy, and scale in a frictionless way. People call it PLG now, as if it's some sort of new thing. At Salesforce, 20-plus years ago, we had a free trial. Was that PLG? Of course.

But in terms of infrastructure, the goal was to develop a service that you or another engineer could stand up, evaluate, test, and even push into production without ever talking to someone in sales. And if they needed to talk to someone in sales, they weren't going to talk to a salesperson. They were going to talk to an engineer because they wanted a peer discussion.

Nobody wants to be sold to. People like buying things, but nobody likes being sold to.

Snowflake, by comparison, invested heavily in sales and marketing, targeting the enterprise buyer. They were very successful in doing so. It took them a lot longer, and it required a lot more capital.

I just thought the Datadog path was going to be a lot easier, frankly, and offer a faster time to market. That was the design principle behind ClickHouse Cloud.

When somebody like you, or somebody at LangChain, Vercel, Lovable, Decagon, Sierra, Anthropic, OpenAI, or anybody else using ClickHouse wanted help, they weren't going to get sold to. We were going to establish a joint Slack channel with our top engineers, including Alexey and our committers, and make them successful in such a short period of time that they would be in production before our competitors had even scoped the project.

We weren't going to charge for migration services. Let's say you're migrating off of Postgres, Snowflake, or BigQuery. We were going to help you do that without any investment of capital on your part. You were investing your time to evaluate our service, and we were going to make you successful as soon as we possibly could. That was the spirit, and 4 years later, it still seems to be working.

6. The SaaS Crash, Agents & the Future of Infrastructure

Another topic that I love to ask you about is the SaaS apocalypse, or the cratering of SaaS company valuations, which on one side seems very extreme and has hurt a lot of good friends of mine. On the other hand, even at Weights & Biases, we see more of our customers engaging with our products like a database, using agents to log into our software instead of humans. I'm curious where you stand on that.

One of the few benefits of age is experience. I've been through a lot of cycles in our industry, from 9/11 to the financial crisis of 2008 and 2009 to the SaaS sell-off during COVID.

This is significant. I don't think we're going to get back to the multiples that people have enjoyed in the past, so I think this is a new normal in terms of what companies are going to trade at. I think it's a little overblown, and I think there are companies like Datadog and Snowflake that are being incorrectly associated with SaaS companies like Salesforce, ServiceNow, and Workday. I think they're very different.

I do think valuations go up and down. I try to tell our employees that all the time. I've seen companies where employees get so fixated on the stock value, valuation, and stock price that they get drunk on it, and the hangover is rough when all of a sudden the valuation gets cut in half. You've got to have a long view.

Not everything is always up and to the right. It's been great for ClickHouse over the last 4 years, but we will be tested and tried like every great company over the course of decades. That's the time horizon you have to have in mind.

I do think serving at the infrastructure layer is a very good place to be right now. If you simplified it, it's picks and shovels for the gold rush. Everyone's going to need a database, and what are the characteristics of that database? It always comes down to price and performance.

Obviously, when you have a hosted service, reliability is P0, along with durability, security, scalability, and all those things. But when somebody makes the purchasing decision, it's going to come down to price and performance, assuming everything else is table stakes.

I don't think we're going to get back to the levels that we all enjoyed in the past from a multiple basis—you can just call it a multiple on forward revenue. I think these companies are being valued on their free cash flow dynamics more so today than they ever have been in the past, and on the terminal value of the company in the future, more so than simply expecting that these companies are going to generate 30% to 50% year-over-year growth.

I think it's a really good opportunity for companies like ours that are well-capitalized and private, and we're seeing that with Databricks, for example. They're reaccelerating growth on a $4 billion revenue base across a variety of different product lines. I think their data warehousing product offering has passed $1 billion of revenue. Their AI-related workloads are now over $1 billion of revenue, and they're accelerating growth.

I think Databricks, ClickHouse, and companies that are designing for agents, not humans, are going to get a lot of lift over the next number of years. I'm thinking about a world where these agents are actually selecting and provisioning the infrastructure behind an application.

Anthropic presented at our user conference last year with you, Lukas, when you were on stage talking about why you chose ClickHouse. Anthropic, OpenAI, and Tesla were also on stage presenting. Anthropic asked Claude at the time what they should be using for a specific use case, and Claude suggested ClickHouse.

I was talking to the CEO of one of the largest fintech companies in Europe a few weeks ago, and he said, “Every LLM I ask what I should be using to replatform our company suggests ClickHouse.”

That's great, but it still requires a human to make that request. I'm thinking about a world where these agents are actually selecting and provisioning the infrastructure behind an application. You say, “Build me an application that needs to observe telemetry,” and it's going to go and not just recommend ClickHouse, but provision a service and stand up the stack.

It's going to have not only an analytical database, but transactions that need to be stored. It's going to provision a Postgres service. We're developing a managed Postgres service like many others, but we've been thinking about this for years. It's in private preview today, it'll be in public beta in a few months, and it'll be generally available by the end of the year.

We're going to have this unified data stack where you can have both an analytical workload running on ClickHouse and a transactional experience built on Postgres. It's not going to be yet another managed Postgres service. It's going to be the world's fastest Postgres service in the cloud.

7. The Datadog Love-Hate Story

But I guess when I look at Datadog, I'm also a fan. I've been a fan for a long time, and I admire the team over there. But that seems like a harder position, honestly, doesn't it?

They sit at this layer where developers love them because developers are logging in and enjoying the thoughtful interfaces they've built. But if it's mostly agents logging in, that seems like a massive disruption to their business model.

I do see more and more people just using ClickHouse directly as their observability layer rather than going through something like Datadog. Now you can easily build these custom interfaces for your particular application.

I think that's a valid way to look at it. As you know, it's impossible to look out 3 to 5 years right now in our industry. It's changing faster than I've ever experienced in my career.

If you look at a 1- to 3-year time horizon, which is really the only realistic outlook, you can see Datadog still being a very meaningful supplier in the industry. It's such a premium product. People really underestimate how difficult it is to compete with Datadog and develop a comparable service.

We tried at my previous company, and I would not say we were successful. I think we have a reasonable view on how formidable a competitor Datadog is and how high-quality a product it is.

I'll give you an example. Lukas, we used Datadog internally in the past. When we were developing our service, it was the obvious choice because we had this incredibly compressed time frame to launch ClickHouse Cloud. Standing up a Datadog service for APM, logging, and metrics was the obvious choice.

It grew like wildfire, as it does in many companies. All of a sudden, I was spending 7 figures on Datadog before I had any revenue. There was a period when I was spending more on Datadog than I was generating in revenue as a company.

We also knew that you could use ClickHouse for observability, as you just mentioned. We didn't have HyperDX yet, so you had to rely on something like Grafana, which is another great product and an alternative to Datadog.

We went to the company—we were about 100 people at the time, and we're now about 500—and said, “We have to migrate off Datadog. We have to use our own technology.”

The concept of dogfooding, or drinking your own champagne, whatever you want to call it, is really important. We can't say that we're a meaningful supplier in observability if we don't even use our own technology.

It was almost like we had a mutiny in the company. Our developers said, “You can't take Datadog away from me.” It was like I was taking heroin away from a drug addict. They said, “You can't take my Datadog.” I said, “We have to migrate off Datadog. There's no question that we need to do this. It's just a function of how quickly.”

It took us longer than I would have liked, but eventually we achieved it. We published a blog about what life is like on the other side of Datadog in terms of the cost savings and performance improvements that we received.

I would encourage anybody who's considering evaluating ClickHouse as an alternative to read it. It's deeply technical and provides a bit of a framework for how to go through that process.

I see Datadog being a very meaningful supplier. You saw their recent quarterly results and some of the big wins they were able to secure at leading AI companies, so I do think they're going to survive this and come out on the other side successfully.

But again, like many companies, I don't think they're going to trade at the same multiple they have in the past.

Are you feeling an impact of AI on your hiring plans, both on the engineering side, where we're seeing a lot of automation, and on the go-to-market side, where I feel like there's early energy around automation and it seems to have slowed down, at least at this moment in history in 2026?

Yes and no. No, in the sense that we're 500 employees today, and we're going to end this year closer to 1,000. So we're going to double the size of the company in 12 months. That headcount plan has not come down as a result of AI. If anything, it may accelerate, because our engineers are going to be 10 times more productive by embracing AI applications to make themselves and their peers more successful and to develop code at a much faster pace.

There are functions in the company that we are automating that we didn't even have beforehand. For example, the concept of an SDR: when I started the company, I said we were never going to have SDRs, or CSMs, because I think it's clumsy, and I think they're functions that mask other issues around product quality and sales efficacy. I always think about the customer's experience and what it's like to be handed off from a junior inside sales rep to a quota-carrying rep to a renewals manager who's described as a customer success manager, and so on.

But I think we can deploy AI agents that can serve a lot of those functions around lead qualification and demand generation, as well as enhancing the customer experience by giving customers the information they need at that point in time, providing contextual help in the product, AI assistants, and so on. The ability to query your data using a natural-language interface through just a text prompt—for example, to say, "Hey, build me a stacked bar chart that shows revenue growth over the last 12 months and break it down by region"—means I don't need to go to a data analyst anymore and have that individual create that chart.

I can just ask LibreChat, which is an open-source chat interface that we acquired last year and are integrating with InfluxDB's model and our MCP server into a ClickHouse Cloud service. Lukas, I would highly encourage you to check this out for your own implementation of ClickHouse, where you can query your production data in real time and do ad hoc analysis. You can do situational-awareness planning and scenario planning. You can say, "Build me 3 different possible outcomes: optimistic, pessimistic, and realistic," and it will query your data. The insights you're able to get from it are staggering.

That's how we are using AI internally and how we're building it into the product itself, which we've been thinking about basically since before ChatGPT. But it's not having a negative impact on headcount, to answer your question.

How is it affecting your engineering roadmaps? I feel like infrastructure is the place where AI engineering has had the slowest adoption. Do you allow, for example, AIs to automatically submit code into your codebase?

There is typically a human review of code that goes into the codebase. You have to remember, we're building a database. We're not building a mobile app over the weekend, like a dating app. "Move fast and break things" doesn't apply to a database. If we don't have a highly reliable, durable, and secure service, then we don't have a company, because our customers won't trust us.

It's a little bit different when you're developing this type of mission-critical application to blindly commit code that an agent developed without some sort of human review. But the amount of contributions and commits to the codebase that are AI-assisted is significant and is growing every day. I would anticipate that if that number is 50% today, within 6 months it will be 80%.

Here's a question I've been wanting to ask you. You've been having what looks like an incredibly charmed run. But as I've gotten closer to CEOs who look like they're on this hockey-stick run, I've always found that there are tough things internally, these kinds of "oh" moments. I wanted to hear from you: what's been the biggest challenge or the scariest moment in the life cycle of ClickHouse so far?

It hasn't been a straight line. I'll start there. It may have been from the outside looking in.

It looks like a straight line from the outside, Aaron, I have to tell you.

Yandex is commonly referred to as the Google of Russia, which back in 2021 was not a bad moniker. It's not a great one today. A condition of forming the company was that I had no ties to Russia. I didn't have the foresight that Russia was going to invade Ukraine, but I knew enough to understand that they were going to be developing a cloud service, and I couldn't have engineers coding on a Russian network. I couldn't have anybody in Russia.

So I convinced Alexey and the team to move from Moscow to Amsterdam. That was not an easy achievement, convincing these engineers to pick up their lives and relocate to a new country. Then Russia invaded Ukraine a month later. They moved in January of 2022, and Russia invaded Ukraine on February 24, 2022—6 weeks after they moved.

We then had this association and relationship with Yandex that we needed to untangle. That was not an easy process. We celebrate the heritage of the software and the genius of Alexey, Sasha, Nikolai, Nikita, Ksenia, and these incredible engineers who are still in our company. But we also sympathize with the fact that they basically had to leave their home country and relocate and have not gone back. Culturally, that was very challenging.

I then put $300 million with Silicon Valley Bank and got a call that Silicon Valley Bank was going to default and I was going to lose all my money. I said, "Oh, shit." I'm pretty sure $200 million of that was off its balance sheet and being custodied at U.S. Bank. I'm not entirely sure, but I'm pretty sure, because Silicon Valley Bank was still the asset manager. But nobody could give me a clear answer on that Thursday.

I had a high degree of confidence that $200 million of the $300 million was not at risk, but there was $100 million that was on their balance sheet. I had this SWAT team—myself, my head of finance, and my general counsel—trying to figure out, like the rest of the industry, what the fuck should we do? Should we just wait it out and hope that they don't default and we don't lose $100 million, or do we contribute to a bank run and wire the money somewhere?

Fortunately, we had banking relationships set up in Europe, which many startups didn't. They didn't have a path out, so founders were wiring themselves money, which is not a great look. I made the decision mid-morning that Thursday. I called my head of Europe, Arno, and said, "Wake up the bankers at our bank in the Netherlands, because there's a $100 million wire that's coming through in the next 30 minutes, and it needs to clear because I think there's going to be a massive run and the system's going to freeze."

8. The Origin Story: From Yandex to ClickHouse Inc.

We got the $100 million out. I called a friend of mine at Silicon Valley Bank and said, "Man, I'm sorry. I have to do what's best for my company, my employees, and our customers, and I can't afford to lose this money." We wired $100 million, and it cleared. Thirty minutes later, SVB's banking system went down, and people could not get their money out.

The government ended up backstopping SVB on Sunday, as we all know, but there were 72 hours when nobody could tell me whether or not the remaining $200 million was safe. So that was a little bit of a tricky time.

As you know, being a longtime ClickHouse Cloud customer, these services have issues. We haven't had anything catastrophic, but around reliability and durability, we've had to turn inward and ask, "Are we doing everything that we can to make this the most resilient service in the market?" That has not been a straight line.

I love where we are today in our security posture and our reliability and durability, but it's not like you just wake up and have the most secure, durable database service. It requires a lot of sleepless nights and a lot of engineering work to make sure that you're protecting your customers' data and making the service available, especially for Weights & Biases. You're building a customer-facing application on it. You're betting your company on our service. There's a lot of responsibility that goes with that.

I've been very fortunate with the team that we've assembled. For the most part, my leadership team is intact from when I formed it, when we established the company, and we've added to it. So that's been a total joy.

Those would be some of the trickier things that we've had to navigate in terms of company building, I would say. We've also, like Weights & Biases, got great investors. We've added to the roster, so we're well capitalized. But I don't think our company has really been tested in terms of how strong our resolve is, and it's going to come, because the competition is waking up to the threat that we pose. They're not just going to lie down and accept that ClickHouse is going to capture this market. These are very well-organized, very well-capitalized companies with incredible leadership.

Who do you see as waking up and coming after you?

It depends on the use case. Let's talk about data warehousing. Who are the big suppliers in that market? Snowflake is obvious. Google BigQuery is a great product with a big customer base that we would love to eat into. Amazon Redshift. Increasingly, while we partner very well with Databricks, we do see some competitive overlap. Those 4—let's start there.

With observability, I mentioned Datadog, where there is some competitive overlap. I mentioned Grafana and Elastic. Again, these are great companies with a very strong customer base, a lot of capital, and very capable teams.

I feel like, in terms of real-time analytics, which is what ClickHouse was originally designed for, we are the market leader in that category. Whether it's Vercel, Lovable, Sierra, Decagon, Ramp, Vantage, Klaviyo, or Attentive, you name it, these are customer-facing B2B SaaS applications. Weights & Biases and LangChain are also customer-facing B2B SaaS applications, and I don't think there's a strong alternative to ClickHouse for those use cases.

9. Hardest Moments: Russia, SVB & Sleepless Nights

All right, let's have a good spot to end. Thanks.

Thanks, Aaron. Fun interview.

10. Outro

All right. That was good. Awesome.