20VC: The AI Bubble Is Wrong | AI Margins Need to Improve | Revenue Concentration Should be a Concern | Why People Over-Estimate Open Models But Enterprises Still Fear Frontier Models with Aaron Katz, ClickHouse

Summary

Aaron Katz frames ClickHouse as an infrastructure company benefiting from the fastest AI adoption cycle he has ever seen, while warning that AI application revenue can be fragile because switching costs are often low. He says gross margins matter less than sustaining growth and proving a path to margin expansion, and he is most focused on revenue durability and customer value. On the product side, he expects agents to create new demands around low latency, unpredictable query patterns, identity, budgets, authorization, and governance, and he believes agents may eventually select infrastructure stacks themselves. He also argues enterprises will still prefer frontier labs over open-weight models for sensitive workloads because they want indemnification, legal protections, and stronger output guarantees. Beyond product, Katz describes ClickHouse’s rapid scaling, distributed hiring, global deployment footprint, and balanced capital strategy, and he says there is no rush to go public because structured private-market liquidity has reduced the traditional IPO advantage.

Notes

• ClickHouse is an open source database positioned as infrastructure for AI-native companies, with emphasis on query speed and storage efficiency.
• Katz says AI-company growth is accelerating at an unprecedented pace.
• He argues the market is still early, not in a late-cycle AI bubble.
• He is less worried about gross margins than about multi-year margin expansion and sustained growth.
• For AI applications, Katz sees revenue durability as the central risk.
• He says switching costs are low for agentic apps because model providers can leapfrog quickly.
• ClickHouse’s own Anthropic spend is up 100x from the start of the year.
• The company internally pushed an “AI awakening” memo to accelerate adoption.
• Open-weight models are useful for code review but not necessarily for shipping production code.
• He says token expense can be managed if revenue growth continues to outpace it.
• ClickHouse has about 100 quota-carrying salespeople and Katz wishes he had scaled sales earlier.
• ClickHouse used a Datadog-style PLG motion before layering enterprise sales on top.
• The company is shipping faster and entering new categories two years ahead of plan.
• ClickHouse uses acquisitions as a founder-partnership mechanism, not just a buyout process.
• For agentic query workloads, low latency and unpredictable patterns are the core infrastructure requirements.
• He says agents will force pricing-model changes because query volumes are exploding.
• He imagines agents choosing the underlying infrastructure stack for applications.
• Tesla is an extreme-scale example: Katz says it ingests a billion events per second into ClickHouse.
• Enterprise deployments need cloud, on-prem, VPC, and firewall flexibility.
• Agents will need identity, budgets, and authorization before they can autonomously consume services.
• Today, humans still monitor agents, set controls, and prevent unauthorized spend or data access.
• He expects specialized models and frontier labs to coexist, with frontier providers still dominant.
• ClickHouse positions itself as infrastructure, not a business that picks category winners.
• The customer base expands rapidly, with net dollar retention over 200% and midpoint spend around $100,000.
• He says the biggest competitive threat is the next company that has not launched yet.
• He predicts a roughly even split between open and proprietary software over time.
• He does not rule out frontier labs becoming infrastructure providers.
• Enterprise buyers want indemnification and output guarantees, which open-weight models may not provide.
• He explicitly distinguishes open-weight models from open source software.
• Companies worry about sending source code to frontier labs because of trust and retention concerns.
• Open models may be fine for review workflows but not for pushing code into production environments.
• Sensitive enterprise workloads still point toward frontier labs because of legal protection and security concerns.
• Katz thinks open source and open-weight models are the future, but he would not deploy capital to build one himself.
• He says some digital-native companies are even considering moving back on-prem.
• Large financial-services firms are adopting new technology faster, with sales cycles compressing from years to quarters.
• Open source and PLG reduce evaluation friction and help customers deploy without a traditional enterprise-sales process.
• Investors worry about redistribution risk if hyperscalers offer managed versions of ClickHouse.
• Agents may eventually pick the underlying database and infrastructure stack themselves.
• Brand awareness still matters because budgets are controlled by people, not agents.
• ClickHouse raised capital strategically and did not need the cash because it already had $1B on the balance sheet.
• Sports sponsorships are justified through awareness and hospitality, especially with enterprise budget owners.
• He believes sports assets are structurally valuable because the experience is hard to replicate with technology.
• ClickHouse can measure sponsorship ROI through account-level spend, site traffic, and trials.
• Katz values investor relationships mainly for customer introductions, advocacy, and recruiting help.
• He says fundraising should be evaluated over a 20-year horizon, not a two-year valuation cycle.
• ClickHouse’s key operating goals are durable company building, rapid ARR growth, hundreds of customers per month, and very high retention.
• He targets $1B ARR by December 2027 or earlier.
• Katz says competition for talent differs sharply between AI startups and infrastructure companies.
• ClickHouse hires globally, with people in 27 countries and single-digit attrition.
• ClickHouse is moving toward a distributed office model rather than strict return-to-office.
• More than half of ClickHouse revenue and customers are outside North America, and the company operates across 36 hyperscaler regions.
• The product launch overlapped with ChatGPT, and database adoption moved slowly at first.
• He says venture pricing is about where the company will be in 12 to 18 months.
• The Series B felt expensive in hindsight because the company had no revenue, product, customers, or meaningful headcount.
• He says ClickHouse was built to win on price and performance regardless of the next AI trend.
• He says the current funding round is the cheapest in hindsight.
• ClickHouse revenue ramped gradually: 0, 12, 50, 200, then north of 500 for the year.
• Revenue concentration above 10% in any category or customer base is a real operator concern.
• Customer value must stay ahead of competitors on ROI and total cost of ownership.
• ClickHouse avoids internal conflict and prefers to hire highly competitive people.
• The engineering org skews experienced, with an average age in the mid-to-late 30s.
• He thinks software engineering may become a very uncertain labor market within four years.
• For serious medical or legal questions, Katz still prefers the best human expert.
• He thinks robotics will be more disruptive than chatbots in medicine.
• Katz predicts an AI finance function devoted to tracking AI consumption.
• He thinks that AI finance role will eventually disappear once agents govern themselves.
• ClickHouse does business with Meta, Microsoft, Nvidia, and Cerebras, and Microsoft is a major analytical workload.
• He expects hyperscalers and frontier labs to remain important in the chip ecosystem.
• He says ClickHouse could go public next year if it wanted, but there is no rush.
• He says public-company stock can fall 40% to 50% on a slight quarterly miss, which hurts morale.
• Structured tenders have reduced the employee-liquidity argument for going public.
• He still sees public markets as better for long-term price discovery.