Aaron Katz on ClickHouse's Explosive Revenue Growth

The interview took place inside Craven Cottage, the home stadium of Fulham Football Club, because the company being discussed had just bought the shirt sponsorship. That fact alone tells you something about the current moment in technology: a database company, the kind of business that used to advertise in trade magazines, is now paying to put its name on a Premier League jersey. But the real spectacle that day was not the pitch. It was the numbers.

Aaron Katz is the co-founder and CEO of ClickHouse, an open source database built for fast analytical queries over huge volumes of data. Anthropic, OpenAI, Tesla, Netflix, and Microsoft all use it. The company has just crossed $350 million in annual recurring revenue, or ARR, the standard measure of a subscription business's yearly income run rate (00:50), and it is valued at $15 billion (44:17). Katz has lived through three technology cycles as a builder, first at Salesforce for twelve years, then at Elastic, and now at ClickHouse. So when he says the current AI wave does not resemble anything he has seen before, it is worth taking seriously.

"This seems to be accelerating at a unprecedented pace," he told Harry Stebbings, host of the 20VC podcast. "We haven't seen revenue growth like this in our lifetime" (05:10). ClickHouse's own trajectory backs this up. The company went from zero revenue to $12 million, then $50 million, then $200 million, and will finish this year north of $500 million.

"We went zero, 12, 50, 200, and we'll finish this year north of 500." — Aaron Katz [45:24]

That is not a smooth SaaS curve of tripling and then doubling. It is a jump that Katz says outpaces even the fastest historical infrastructure companies he can name, including Datadog and Snowflake, the two he studied most closely before building his own go-to-market plan (10:08).

The Real Risk Is Not Margins

The obvious objection to all this growth is gross margin, the percentage of revenue left after paying for the cost of delivering a service. Some AI-native companies run gross margins in the 30 percent range, far below the 70 to 80 percent that public software investors expect. Katz is not worried about that, as long as a company can show a credible path toward margin expansion over a few years while growing this fast (06:06). What worries him instead is durability: whether the revenue sticks around.

"The switching costs can be very low for agentic applications, and we're seeing that with these model providers that are leapfrogging one another." — Aaron Katz [06:45]

Infrastructure software like ClickHouse is deeply embedded in a customer's systems, so switching away is expensive and slow. An application built on top of a large language model can be rebuilt around a different model in weeks if a competitor ships something better. That gap, Katz argues, is the honest risk in today's AI investing environment, more than the margin story that gets most of the attention.

Some of that discipline traces back to his years at Salesforce, where he worked under founder Marc Benioff for twelve years starting when the company was a three-year-old startup selling what was essentially a glorified contact manager. Benioff insisted on chasing giants like Siebel, SAP, and Oracle long before the product could compete with them, and he was right on a longer timeline than anyone expected.

"That you can overestimate what you can achieve in one year and underestimate what you can achieve in five." — Aaron Katz [11:17]

Katz has carried that patience into his own fundraising. ClickHouse did not need its most recent round of capital; the company already had a billion dollars on its balance sheet. He raised anyway, in part to bring in strategic investors, and in part to fund things like the Fulham sponsorship, which he frames less as sentiment and more as brand infrastructure aimed at the humans who still control enterprise technology budgets (31:03). He argues that sports and live experience are becoming more valuable, not less, precisely because so much of work and life is moving into software (34:42). He also predicted that top sports franchises will eventually sell for $20 billion, pointing to the Los Angeles Lakers' recent $12.5 billion sale as a signal of where prices are headed (34:05).

When Software Buyers Are Machines, Not People

The most striking part of the conversation is Katz's picture of who, or what, will be choosing infrastructure a few years from now. Today, a human engineer decides whether a company uses ClickHouse or Snowflake or Databricks. Katz thinks that decision is starting to move to the AI agents themselves.

"I'm thinking about a future where the agents actually select the infrastructure stack behind the application." — Aaron Katz [15:24]

He already sees an early version of this: when Anthropic's engineers wanted to pick a database for a new observability feature, they asked their own AI model, Claude, which one to use, and Claude suggested ClickHouse (17:52). If that pattern generalizes, then the winning technology in any category becomes whatever an AI model is most likely to recommend, which changes how companies think about developer relations and community.

That agentic future has a very specific technical requirement, in Katz's view: speed. Human users tolerate a slow dashboard. An AI agent running dozens of database queries at once, to answer a single question, cannot. "What's the number one requirement for agentic query patterns? Low latency," he said, describing systems whose experience "is gonna be defined by the slowest point in that chain" (14:09). As proof of what this scale looks like in practice, he pointed to Tesla, which is now ingesting a billion events per second into ClickHouse (15:44).

That same shift toward machine buyers raises a trust problem, and here Katz offers a genuinely counterintuitive claim. The conventional wisdom in AI circles holds that closed, expensive frontier models like those from Anthropic and OpenAI will be reserved for a handful of extremely high-value problems, like curing cancer, while cheaper open weight models handle the rest of enterprise work. Katz disagrees, especially for large companies (23:59). Open weight models, particularly those built in China, cannot yet offer the legal indemnification, meaning a formal guarantee against liability, that enterprises want before they trust a model with sensitive code or data (25:50). Some of his own peers, he says, have gotten this backwards, treating a frontier lab's promise of "zero data retention" as untrustworthy while feeling comfortable sending sensitive material to an open Chinese model, when the actual risk profile runs the other way (25:26). ClickHouse itself limits open models to lower-stakes tasks like code review while keeping frontier models for anything that touches production systems or customer data.

None of this optimism blinds Katz to concentration risk, the danger that a company depends too heavily on one customer, category, or trend. He treats any revenue segment that crosses 10 percent of total revenue as a real concern worth active management (46:41). Right now, the entire basket of AI-native customers on ClickHouse, including Anthropic, OpenAI, Harvey, Sierra, and Decagon, adds up to just under 12 percent of revenue (45:45), a number he is comfortable with because ClickHouse's customer base is broad enough that no single wave, even the AI wave itself, can sink the business if some of those names fade.

That comfort with uncertainty extends to the biggest question hanging over the whole industry: is this a bubble? Katz's answer, delivered near the end of the conversation, is a bet rather than a certainty. He does not think AI is overblown, and he is willing to put a number behind ClickHouse's own ambitions, wagering that the company hits $1 billion in ARR before December 2027 (37:57), well ahead of the pace of Salesforce, which went public in 2004 at a $1 billion market cap after years of building. Whether ClickHouse goes public on that timeline or later, Katz says, is almost beside the point. "I'm hoping this company outlives me," he said, "in which case whether or not we go public next year or in five years is pretty irrelevant" (58:19). It is the kind of statement that only makes sense from someone who has already watched a five-year plan get blown apart by events nobody saw coming, and decided the only sane response is to build for a horizon further out than the current panic, or the current hype, can reach.