Aaron Katz on ClickHouse, AI Revenue Durability and Enterprise Data

Harry Stebbings welcomes ClickHouse co-founder and CEO Aaron Katz at Fulham Football Club. Katz describes ClickHouse as an open-source analytical database used for a broad set of workloads. Stebbings discloses that he is an investor in the company.

Asked where the AI industry is in its cycle, Katz says he thinks it is still early. Compared with the internet, mobile and social transitions he experienced, he sees agentic products and company revenue growing more quickly; that is his market assessment.

Stebbings asks whether AI companies will be stuck with lower gross margins than traditional software firms. Katz says rapid growth and capital availability can allow companies to operate with lower margins for a period, provided they can demonstrate a credible path to improvement.

For an investor, Katz identifies revenue durability as a major risk. In his view, infrastructure products can have high switching costs, while some agentic applications may be easier to replace as model providers and products leapfrog one another.

Stebbings asks whether a prominent coding product could face that same risk. Katz says it belongs in the category he is watching, even though ClickHouse’s own use of Anthropic tools has risen sharply. He reports that the company’s spending increased dramatically from an initially small base.

Katz says he sent an internal ‘AI awakening’ message because he thought ClickHouse was adopting coding tools too slowly. The team increased usage and is also exploring open-weight models, although he says it applies different tools to review and production use cases.

On token budgets, he says he is willing to spend while revenue grows and the company ships features faster. He expects model costs and efficiency to improve and believes spending can be controlled later; these are his operating judgments, not guarantees of returns or price declines.

Looking back, Katz says he might have hired more sales capacity. He describes roughly 100 quota-carrying sellers covering many use cases while competing with far larger sales organizations. The productivity and comparison figures are company-reported observations.

He initially wanted pressure on product and engineering and preferred a self-service, developer-led approach. Over time, he says, a company also needs to add enterprise sales. He contrasts product-led distribution with the sales-heavy approach of some data-warehouse providers.

From his years at Salesforce, Katz recalls the lesson that leaders can overestimate what they will do in one year and underestimate what they can do in five. He credits Marc Benioff with communicating an ambitious vision before the product had every enterprise feature.

Katz says ClickHouse is entering some product categories roughly two years earlier than expected through internal development and acquisitions. He reports six acquisitions over four years and cites a recent acquisition in agent observability; the timing and impact are his company’s account.

His build-versus-buy test starts with whether the internal engineering team can enter an area organically. If founders are already building on ClickHouse in a category he expects the company to need, an acquisition may make sense, provided the partnership is a good fit.

For agentic workloads, Katz expects query patterns to become more exploratory and less tied to a fixed human persona. Agents may query several systems in parallel, making responsiveness and the slowest part of a workflow important to the user experience.

Stebbings asks whether greater query volume forces prices down. Katz says he does not assume agents will automatically manage costs like humans. He expects infrastructure providers to reconsider consumption and pricing as workloads expand.

Katz imagines agents eventually selecting components of an application’s infrastructure. He wants ClickHouse to be a default database in that process, but this is a company ambition and forecast, not evidence that agents will make all purchasing decisions.

He emphasizes latency, ingestion and storage efficiency and cites a large customer’s event volume as an example. His claim that ClickHouse alone can meet a particular requirement is a competitive assertion, not an independent comparison of all database technologies.

Enterprise security and compliance, Katz says, require a choice of deployment models: managed cloud services as well as environments controlled by the customer. Supporting multiple options complicates the product roadmap but may serve customers with different requirements.

Looking several years ahead, he expects agents to need identities, authorization and budgets before they can provision services independently. He acknowledges that human oversight currently controls spending and access; the timing and degree of future autonomy remain uncertain.

Stebbings asks whether companies will use millions of specialized models or mostly frontier models. Katz expects both: specialized systems for some domains and major labs remaining important providers. He does not claim to know which individual application companies will win.

On ClickHouse’s customers, Katz says revenue has grown slightly faster than customer counts because existing users expand into more use cases. He reports net dollar retention above 200% and a wide range of customer spend; these are company statements, not independently verified figures.

Asked about competition, he says he worries most about the technology not yet in the market. ClickHouse itself began as an open-source project that incumbents could have dismissed, and Katz wants the company to anticipate a similar challenger from behind.

Katz offers an even split between open and proprietary approaches as a rough prediction, pointing to different business models in data and observability. He distinguishes open-source software from open-weight models and says enterprise adoption of the latter may follow different dynamics.

He would not dismiss frontier labs eventually offering more infrastructure themselves, though he says he does not see them as ClickHouse competitors today. This is a speculative future risk based on the breadth of their current ambitions.

Katz believes some enterprises will value indemnification and other protections when choosing models and that this could limit certain open-weight deployments. The availability of contractual and technical protections varies by provider and deployment; he is expressing a view, not a universal rule.

Stebbings raises the opposite concern: some customers may distrust sending sensitive data to frontier providers and prefer models they can host. Katz acknowledges internal concerns about code and output, and says ClickHouse limits some uses. The appropriate choice depends on the data, model and controls.

Katz says he sees security concerns around some open-weight models but declines to speculate that using software from China necessarily implies espionage. He expects concerns and available protections to evolve; the conversation does not establish a backdoor or any specific incident.

Katz says he favors open-source and open-weight technology and does not favor government measures limiting their distribution. He also says he would not personally direct capital toward building a new open-weight model; these are his policy and investment opinions.

He reports that more enterprise customers, including some digital-native companies, are discussing on-premises deployment. That is an observation from ClickHouse’s customer conversations, not evidence that all enterprises are leaving cloud services.

Katz says some large financial firms are adopting new technology more quickly than they historically did. Open source and self-service products can shorten evaluation by allowing teams to try software before a traditional sales process, although deployment and procurement still vary.

Investors often ask whether a cloud provider can take an open-source database and sell a managed version. Katz acknowledges that risk and argues that differentiated managed services and proprietary features can help defend a business; he says few open-source companies execute that balance well.

Even if agents select parts of an application stack, he expects enterprise infrastructure purchases to involve human engineers and budget owners for years. He therefore does not plan to stop investing in developer relations or brand awareness.

Katz describes concerts around technology conferences and a football sponsorship as ways to build awareness and relationships. He says hospitality can bring existing and prospective customers together, while recognizing that attributing subsequent sales to a sponsorship is difficult.

He describes measuring account spending and top-of-funnel indicators such as website visits and trials, but says a direct causal link to the sponsorship is hard to isolate. The discussion of sports-asset prices is a personal market opinion, not evidence of a particular return.

On fundraising, Katz credits technical co-founders and says he evaluates investors for customer introductions, advocacy, recruiting and how they work with other people. He checks references with founders who have already worked with prospective investors.

He argues that a temporary valuation step-up matters less than building a durable company with appropriate long-term investors. Fundraising can provide capital, recruiting benefits and employee liquidity, but he wants to judge those benefits against dilution and the company’s longer horizon.

Katz says ClickHouse’s priority is new customer acquisition and reports high gross and net retention. He predicts the company could reach $1 billion in annual recurring revenue before the end of 2027; that is a target and forecast, not an achieved result.

Stebbings asks whether startups can compete for talent against well-funded frontier labs. Katz says the difficulty depends on the role: AI startups in San Francisco may face especially intense competition, while an infrastructure company can seek distributed-systems specialists in many countries.

He reports that ClickHouse employs people across 27 countries, has low attrition and has used structured secondary transactions. Those figures describe the company as he presents it at the time, not a general claim about the hiring market.

On remote work, Katz says he has mixed feelings. Office work at Salesforce helped his own early career, but ClickHouse began during the pandemic with founders in different locations and found specialized engineers beyond a single city.

He says the company is adding office options without requiring everyone to return. His forecast of more offices reflects demand he sees among staff; the number and timing are plans rather than confirmed future outcomes.

Katz says international customers and cloud partnerships require people across time zones and languages. In his view, that makes a single central location impractical even if in-person collaboration remains valuable.

The most uncomfortable period was the first six months after product launch, he says, when database adoption took time and revenue was slow. ClickHouse had raised substantial capital before it had a mature product, leaving the team under pressure to deliver.

He recalls an earlier funding round that felt expensive because the company had few employees and no revenue or product. He says it took about a year to build the offering and he did not anticipate the scale of the later AI wave.

Katz describes price and performance as key database attributes and believes ClickHouse was well placed for increasing demand. His claim of superior efficiency and performance is the CEO’s competitive view, not an independently established ranking.

Stebbings asks whether going from zero to $100 million in a year is now the standard. Katz says rapid growth should be weighed against revenue durability and the ease with which customers can move to a competitor.

He recounts ClickHouse’s own reported revenue progression and forecasts more than $500 million for the year. He says AI-native companies represent less than 12% of revenue. These are company-reported metrics and projections, not verified financial statements.

Katz regards concentration as an operating concern. When a customer, industry or category becomes more than about 10% of revenue, he wants to understand the exposure because the loss of one relationship or segment could affect predictability and growth.

Retaining customers, he says, depends on continuing to deliver value relative to alternatives. He focuses on return on investment and total cost of ownership, including what a customer spends to run ClickHouse compared with another service.

Asked whether a company needs an internal enemy to motivate people, Katz says he prefers not to manufacture adversity. External competition and personal challenges are already plentiful; he wants to hire competitive people and improve employees’ working lives.

He says ClickHouse tends to recruit engineers with distributed-systems experience and pairs less experienced employees with veterans. His account reflects the company’s hiring preferences rather than a claim that one age or background is best for all teams.

On college, Katz says he would encourage his daughters to attend for the life experience while acknowledging that the value of a degree depends on the field. Stebbings questions whether curricula keep up with fast-changing technology.

The discussion turns to medicine and AI. Katz says that for a serious medical or legal issue he personally wants an expert; Stebbings describes using AI for some medical questions. These personal views and an anecdote about a procedure are not evidence that AI can safely replace professional care.

Katz predicts that organizations may develop a finance role focused on AI consumption and budgets, and speculates that software could later automate parts of that role. Both the job’s prevalence and the pace of automation are uncertain.

Asked about technology companies and the chip market, Katz mentions commercial relationships with several firms and expects cloud providers and frontier labs to become more relevant chip designers. His comments are market predictions, not investment recommendations.

He disagrees with the view that AI is merely an overblown bubble, while acknowledging he cannot identify every winner or loser. Stebbings raises debt and market concentration; Katz distinguishes private investment risks from the disclosures required of public companies.

Katz accepts that asset values can compress but does not predict a return to prices from a decade earlier. The discussion includes broad claims about market concentration that are not independently verified in the interview; neither participant can know future returns.

Katz praises board member Peter Fenton’s pattern recognition and help with recruiting. He says he tells a board member whether a candidate conversation is meant to assess the person or persuade them to join, since those require different approaches.

His recurring concern is an unexpected competitor that changes the market in the way he believes ClickHouse did. He wants the company to keep reinventing itself rather than focus only on rivals already visible today.

Asked to choose between removing two competitors, he instead says he respects Databricks and sees both overlap and distinct areas in the markets they serve. His description is a competitive perspective, not a comprehensive market analysis.

On balancing leadership and parenting, Katz emphasizes commitment, asking for advice and learning from people with experience. In marriage, he values accepting that progress is not linear, recognizing differences and maintaining a shared purpose.

Stebbings mentions a family member’s chronic illness and his hope for medical advances. Katz agrees that future medical progress would be important, while saying that forecasting five years ahead is difficult. Their hopes are not medical predictions or advice.

Katz says ClickHouse could consider a public listing but feels no rush and wants the company to outlive him. He presents possible timing as a choice, not an announced offering or assurance that a listing will occur.

He sees costs to public ownership, including volatile stock prices affecting employee morale and the presence of short sellers. Structured secondary transactions can provide some employee liquidity while a company remains private, although terms and access differ.

The speakers mention reported large acquisitions by private companies as examples of possible private-market financing. The interview does not verify the transactions or their terms, and they should not be treated as established facts solely on that basis.

Katz also lists possible benefits of an IPO: broader investor access, financing, awareness and a shared milestone for employees. He believes public markets can offer better long-term price discovery, though prices can still be volatile and the timing is uncertain.

Stebbings thanks Katz and refers back to their wager on the $1 billion ARR target. The wager reflects their expectations, not a financial result; Katz closes by thanking him for the conversation at Fulham.