Aug 31, 2026 · 1h 5m · news
ClickHouse CEO: AI Margins Need to Improve | Revenue Concentration Should be a Concern
0:00 / 0:00
⌖ your search result is the highlighted band (23:13–23:46). Playback starts there
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this 20VC interview hosted by Harry Stebbings at Craven Cottage, ClickHouse Co-Founder and CEO Aaron Katz examines software economics, database architecture for autonomous AI agents, revenue durability, and the strategic milestones driving ClickHouse toward $1 billion in ARR.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Harry holds 28.4% of the talking time here. How this is scored →
Harry as informed peer 5.2 Guest teaching 4.3 Guest disagreement 1.9 Harry pushing back 3.2
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 23:15 Aaron dismisses 90/10 open model forecast
Aaron forcefully rejects Harry's assertion of conventional wisdom that open models will take 90% of token volume, citing enterprise compliance and indemnification.
Hardest push from Harry ▶ 52:00 Harry attacks the value of university curriculumHarry refuses Aaron's polite endorsement of college education, insisting that university syllabi are obsolete by the time they are taught.
Biggest teaching moment ▶ 12:02 Aaron re-architects the concept of agent query loadsAaron provides a masterclass on how agentic queries eliminate predictable human user personas, forcing databases to solve for multi-directional low-latency SQL execution.
Harry holds his own ▶ 1:04:00 Harry challenges the modern rationale for IPOsHarry cites major private liquidity events and massive multi-billion-dollar private acquisitions to challenge why high-performing companies need public markets at all.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| Defining ClickHouse and the Velocity of the AI Cycle | 6 | 5 | 2 | 4 | Harry queries the macro cycle and pushes on gross margin compression in AI companies citing Fireworks AI. Aaron provides historical perspective across tech cycles, explaining why top-line growth and durability outweigh near-term SaaS gross margins. | |
| Internal AI Awakening, Token Economics, and Development Speed | 5 | 4 | 2 | 4 | Harry asks about internal token budgeting and questions relying on single model providers. Aaron explains his internal memo, the AI awakening, and his willingness to absorb token costs to maximize development velocity. | |
| PLG vs Enterprise Sales and Lessons from Marc Benioff | 5 | 5 | 1 | 2 | Aaron reflects on admitting he under-invested in enterprise sales capacity compared to Datadog and Snowflake models. He shares foundational lessons from 12 years under Marc Benioff at Salesforce regarding ambitious roadmaps. | |
| Accelerated Product Roadmaps and Strategic M&A Expansion | 6 | 6 | 2 | 4 | Harry asks deep technical and product architecture questions regarding agentic workloads and M&A integration. Aaron details how agents break traditional human persona query patterns, requiring unprecedented low latency and flexible deployment models. | |
| Specialized Domain Models vs Frontier AI Labs | 6 | 4 | 3 | 4 | Harry tests whether specialized fine-tuned models will beat frontier labs, citing his investment in Lagora. Aaron bluntly clarifies that as an infrastructure provider ClickHouse remains neutral and measures value by compute consumption. | |
| Competitive Moats, Open Architectures, and Enterprise Security | 6 | 5 | 3 | 5 | Harry pushes on the conventional wisdom that 90% of tokens will go through open-source models versus frontier labs. Aaron directly rejects that premise, arguing enterprise indemnification and legal concerns will keep frontier labs dominant. | |
| Open Source Licensing, Open Weights, and US Tech Competitiveness | 5 | 6 | 3 | 4 | Harry highlights reports that enterprises use Chinese open weights for sensitive tasks over US frontier labs. Aaron rejects espionage conspiracy claims while explaining genuine data retention and indemnification tradeoffs. | |
| Enterprise On-Premises Migration and Open-Source Moats | 5 | 5 | 1 | 3 | Harry inquires about on-premises migration and how investors misunderstand open-source moats. Aaron explains why enterprise sales cycles are compressing and how ClickHouse defends against cloud hyperscaler duplication. | |
| Developer Brand Strategy and Grassroots Marketing Campaigns | 5 | 4 | 1 | 2 | Harry questions if developer relations and brand matter when agents start selecting software infrastructure stacks. Aaron explains that human budget owners still control procurement and details unconventional marketing like Chainsmokers concerts. | |
| Sports Sponsorships, Craven Cottage Hospitality, and Franchise Value | 5 | 4 | 1 | 2 | Harry and Aaron explore the economics and experiential ROI of Premier League sponsorships and private box hospitality at Craven Cottage compared to typical corporate marketing channels. | |
| Venture Capital Dynamics, Global Hiring, and the Path to $1B ARR | 6 | 5 | 2 | 4 | Harry presses Aaron on talent acquisition and his timeline to $1B ARR, setting up a friendly bet. Aaron explains why ClickHouse hires globally across 27 countries rather than solely competing against SF-based frontier labs. | |
| Global Distribution vs In-Office Mandates | 5 | 5 | 3 | 4 | Harry challenges Aaron on remote work, asserting his vocal stance for mandatory in-office work. Aaron counters by explaining ClickHouse's international customer footprint requires distributed regional hub coverage. | |
| Overcoming Early Growth Plateaus and Justifying High Valuations | 6 | 4 | 2 | 3 | Harry probes the hardest moments of scaling and justifying high valuations ahead of revenue. Aaron recounts the stress following their Series B and product launch concurrent with ChatGPT's release. | |
| Revenue Durability and Concentration Risk | 6 | 4 | 2 | 3 | Harry questions whether customer concentration risk matters anymore given Nvidia's success. Aaron defends the operator view that no single customer or niche should represent over 10% of revenue. | |
| Switching Costs and Total Cost of Ownership | 5 | 5 | 2 | 3 | Harry asks whether companies really replace software at scale and whether internal enemies create competitive drive. Aaron rejects the concept of internal enemies, emphasizing total cost of ownership and innate employee competitiveness. | |
| The Value of College and AI's Impact on Professions | 6 | 4 | 3 | 5 | Harry strongly pushes back on Aaron's support for college degrees, arguing modern curricula are completely outdated. Aaron acknowledges software engineering uncertainty but defends clinical and human domains like medicine and law. | |
| Quick-Fire Round: Future Jobs, Tech Giants, and Chips | 6 | 4 | 2 | 4 | Harry runs a rapid-fire session covering future roles, big tech stocks, and concentration in public markets. Aaron rejects the narrative that AI is a bubble, reiterating that value creation is just getting started. | |
| Board Dynamics, Future Disruptors, and Databricks White Space | 6 | 4 | 2 | 3 | Harry asks about board governance with Peter Fenton and Mike Volpi, as well as head-to-head competition with Databricks and Snowflake. Aaron notes respect for Databricks while identifying substantial white space in analytics. | |
| Leadership Lessons: Parenting, Marriage, and Vision | 4 | 4 | 1 | 1 | Harry asks personal questions on balancing executive leadership with marriage and parenting. Aaron draws parallels between alignment in a 24-year marriage and aligning company vision. | |
| Timing the IPO and the Reality of Public Markets | 6 | 4 | 2 | 4 | Harry questions why any enterprise should go public given private market mega-acquisitions like Stripe/PayPal. Aaron outlines the genuine benefits and price discovery discipline of public capital markets. | |
| 20VC End Screen | 0 | 0 | 0 | 0 | Brief outro and show credits. |
Statements from this episode (40)
Katz: Anthropic, OpenAI, and Weights & Biases all use ClickHouse
“It's the world's most popular open source database, and it satisfies a very broad Array of use cases, and it's used by nearly every AI native company, Anthropic, OpenAI, Weights and Biases.”
Stebbings: Fireworks AI operates at 30% to 35% gross margins
“Lynn was on the show and she was like, yeah, we're at 30, 35%. We hope to be more over time.”
Katz: Low AI gross margins matter less given rapid growth
“As long as they can demonstrate a path to margin expansion over the course of the next few years, while still growing at these unprecedented levels with very healthy balance sheets, I worry less about gross margins like we did five years ago in traditional ent…”
Katz: Revenue durability is the single biggest risk for AI applications
“If you were to come down to one, if you were to say this, what's the single biggest risk would be durability of revenue because the switching costs that you and I talked about are very high for infrastructure software. The switching costs can be very low for a…”
Katz: ClickHouse's Anthropic spend is up 100x this year
“Our anthropic spend is up a hundred times from what it was at the beginning of the year.”
Katz: Anthropic and OpenAI Outpace Open Weights on Application Harnesses
“Anthropic specifically and open AI have the benefit of being both a model provider and the harness provider that the open weights models right now are behind in.”
Katz: ClickHouse generates over $100M with about 100 quota-carrying reps
“We've got about a hundred quota caring salespeople, for example, and I think, you know, our revenue scale is significantly more than a hundred million. And so our average rep productivity is quite high relative to the industry average.”
Katz: Following Datadog's PLG playbook was easier than Snowflake's enterprise sales
“Datadog had this PLG self-service developer-led motion, so you could get started, deploy an agent, instrument your application, never talk to anybody in sales. Snowflake went heavy after the enterprise through, you know, very expensive sales and marketing. I j…”
Katz: Tesla ingests one billion events per second into ClickHouse
“Tesla, for example, is ingesting a billion events per second into ClickHouse.”
Katz: Anthropic chose ClickHouse after asking Claude for a recommendation
“If you ask Anthropic how they chose to use ClickHouse, they'll tell you they asked Claude, what technology should we use for this specific observability use case? And Claude suggested ClickHouse.”
Katz: AI agents will be fully autonomous within three to five years
“If you look out three to five years, those agents are going to be fully autonomous.”
Stebbings: Every company will have an internal AI model
“I think every company will have their own model, you know, trained on their own data. They'll supplement their own data with additional data.”
Katz: Frontier AI labs will remain dominant over specialized models
“I think you're going to have specialized models for specific use case, like legal tech. If you look at Harvey, for example, I know they're, I think, leading the category in terms of specialization, but I think the large frontier labs are still going to be the …”
Katz: ClickHouse net dollar retention is over 200%
“You know, we had over 130% net dollar retention at previous companies. We're over 200% because these use cases expand.”
Katz: ClickHouse's midpoint customer spend is approximately $100K
“Average customer spend. I mean, we've got customers that spend tens of millions of dollars with us every year. We've got customers that spend thousands of dollars with us every month and everything in between. So averages can be a little bit misleading. If I w…”
Katz: AI token volume will split 50-50 between open and frontier models
“Well, the easy answer is fifty-fifty in the same way that what percentage of enterprise software today is open source versus proprietary. I think it's a pretty even distribution.”
Katz: Enterprise indemnification requirements will limit open-weight model adoption
“I don't, because I think it, especially in the enterprise, they want provisions and protections that potentially open weight models, especially those that come out of China, cannot provide. Around indemnification, for example, and output inference, and it's go…”
Katz: Open-weight AI security concerns will be addressed in 1-2 years
“Today, I think it is. If I look out a year or two from now, I think a lot of these concerns will be addressed.”
Katz: Top Silicon Valley startups are considering moving workloads on-prem
“And even companies that I thought would never be going back on prem are talking about going back on prem. Like some of the most innovative digital native companies in Silicon Valley are now thinking about moving their stack from, you know, one of the hyperscal…”
Katz: Major financial enterprises are adopting technology faster than ever
“A lot of the times, historically, the sales cycles into these big firms would be measured in years, not quarters, right? They're now adopting technologies much faster than they ever have before.”
Katz: Very few open-source companies successfully defend their moats
“I think if you can maintain A competitive advantage with your cloud offering or your proprietary features that are very, very difficult to replicate, then you can maintain that moat. And I think very few open source companies get that right.”
Katz: ClickHouse extended its financing despite having $1B on its balance sheet
“And so we did a financing earlier this year, and then we extended it and brought in some strategic investors like yourself and David Sachs and Michael Dell and JP Morgan. We didn't need the capital. We had a billion dollars on the balance sheet.”
Katz: Sports franchise valuations will reach $20 billion
“I do. I mean, we just saw that with the Lakers. I think it was the most expensive sports transaction in history.”
Katz: ClickHouse will surpass $1B in ARR before December 2027
“Well, I'd put the over under at December, 20, 27. And I would take the under.”
Katz: ClickHouse has almost 800 employees, will reach 1,000 by year-end
“We're almost 800 employees. We'll be a thousand by the end of the year.”
Katz: Over half of ClickHouse revenue and customers are outside North America
“Over half of our revenue comes from outside of the US. 40% here in EMEA, 10% in Asia. Over half of our customers are outside of North America.”
Katz: ClickHouse raised at a $2B valuation with zero revenue
“The Series B that Cotu and Altimeter jointly led at two billion felt expensive. We had no revenue. We had no product. We had control of an open source database. We had no customers. We had 15 employees.”
Katz: AI customers represent less than 12% of ClickHouse's revenue
“The basket of AI companies that's using us in nearly every AI companies built on ClickHouse from Harvey, Sierra, Decagon, Anthropic, OpenAI, et cetera, represents less than 12% of revenue.”
Katz: Having any customer or sector exceed 10% of revenue creates risk
“If I've got one customer that or one category or one industry that accounts for more than 10% of revenue, I spend a lot of time thinking about it. And that seems to be kind of the industry standard, that threshold, that if some dimension of your revenue base a…”
Katz: Companies Should Not Manufacture Internal Adversity or Enemies
“I worry about creating adversity inside of a company. Well, you have so much adversity outside of your company, right? You've got to deal with competitors, disparaging yourself and the company. You've got to think about how you stay ahead of the competitive la…”
Katz: ClickHouse Engineering Team Average Age Is Mid-to-Late Thirties
“The average age of our Engineering organization is in the mid to late thirties, which is a little bit different, I would imagine, than a typical venture backed company that's only had a product for three years.”
Katz: Software Engineering Graduates Will Face an Uncertain Market in Four Years
“If you're studying software engineering, I think you're going to be entering a market that's very uncertain in four years time.”
Stebbings: I Use ChatGPT for Most Medical Queries Instead of Seeing Doctors
“Like, I use ChatGPT for most of my medical queries now, and it prevents me from seeing a lot of doctors.”
Katz: Autonomous Robotics Will Be More Disruptive to Medicine Than Pure AI
“I think robotics would be probably more disruptive, frankly, to the medical industry. We had a family member have a procedure that was done entirely autonomously with just a doctor overseeing it, but didn't actually touch an instrument. And I think that will b…”
Katz: Dedicated AI token-management finance roles will be common within 5 years
“An AI finance function. Solely dedicated on AI consumption inside of an organization. That's all they wake up thinking about.”
Katz: AI token management jobs will be made irrelevant by self-governing agents
“And then that job will be made irrelevant five years from then, because AI agents will govern themselves.”
Katz: ClickHouse powers Microsoft's largest analytical workloads
“Microsoft's one of our power users. We power the largest analytical workloads at Microsoft”
Katz: Hyperscalers and Frontier Labs will be relevant chip providers
“I think a lot of the hyperscalers in Frontier Labs are going to be very relevant providers in the chip ecosystem.”