Overview of 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
In this episode of 20VC, Harry Stebbings sits down with Aaron Katz, CEO and co-founder of ClickHouse, for a wide-ranging discussion on AI infrastructure, enterprise software, open vs. frontier models, and the future of agentic applications. Katz argues that the AI wave is not a bubble, but rather an unprecedented acceleration in product adoption and revenue growth. He explains why he believes ClickHouse is becoming a core database layer for AI-native companies, why enterprise buyers still trust frontier labs more than open-weight models, and why the next major battleground will be agent identity, budgeting, governance, and infrastructure selection.
Key Themes and Takeaways
AI is not a bubble — it is still early
Katz pushes back hard on the idea that AI is overhyped.
- He says the pace of revenue growth in AI is unlike anything he has seen in his career.
- He believes the current cycle is accelerating faster than previous waves like internet, mobile, or social.
- His core view: winners will dramatically outperform losers, even if some companies fail.
Margin pressure is real, but not the main concern
On AI company margins, Katz is pragmatic:
- Gross margins may be lower in the short term than traditional SaaS.
- He expects margins to improve over time as infrastructure gets more efficient.
- He cares more about revenue durability than near-term margin purity.
ClickHouse’s Growth and Positioning
ClickHouse is becoming foundational AI infrastructure
Katz frames ClickHouse as a high-performance, resource-efficient open source database used by leading AI and software companies.
- Customers mentioned include Microsoft, Anthropic, OpenAI, Tesla, and Netflix.
- He emphasizes that the best companies increasingly use ClickHouse for analytics, observability, and AI workloads.
- The company recently crossed $350M ARR and is targeting $1B ARR as quickly as possible.
Revenue growth has been explosive
He describes ClickHouse’s growth as:
- 0 → 12 → 50 → 200 → 500+ million ARR progression
- A trajectory that he says is faster than most infrastructure software companies have ever seen
- Sustained by broad customer adoption and strong expansion within existing accounts
Revenue concentration is a valid concern — but manageable
Katz acknowledges concentration risk as an operator concern, especially in AI.
- He says no single customer or category should dominate revenue.
- AI companies represent less than 12% of ClickHouse revenue, despite being highly visible.
- He sees diversification as important for durability, even if some clusters grow very fast.
Enterprise Software in the AI Era
AI workloads are changing system requirements
Katz argues that agentic applications have very different infrastructure needs:
- Low latency is critical because agents execute many queries in parallel.
- Query patterns are less predictable than human-driven BI/reporting workflows.
- Systems need to handle extreme throughput, efficiency, and responsiveness.
Agents will eventually select infrastructure
One of the most forward-looking ideas in the conversation is that agents themselves may choose the infrastructure stack.
- Today, humans still make architecture and procurement decisions.
- In the future, agents may provision compute, storage, databases, and networking directly.
- That future creates new requirements around identity, authorization, budgets, and governance.
Open Models vs Frontier Labs
Enterprises still fear frontier labs more than many open-weight models
Katz’s view is nuanced and somewhat contrarian:
- He thinks many enterprises still prefer frontier labs like OpenAI and Anthropic for legal and operational protection.
- He also notes that open-weight models, especially some from China, raise concerns around indemnification, output inference, and security.
- However, he expects open models to become more capable and more widely adopted over time.
He expects a roughly 50/50 split between open and frontier
This was one of the more controversial predictions:
- Katz believes open and frontier models may each capture about half the market in the long run.
- He rejects the idea that frontier models will be used only for rare high-stakes tasks.
- He argues that enterprise trust, compliance, and deployment flexibility will keep frontier labs highly relevant.
ClickHouse GTM, Sales, and Expansion
He wishes he had scaled sales earlier
Katz admits one of his biggest mistakes was underinvesting in sales capacity sooner.
- ClickHouse initially focused on product and engineering, following a more Datadog-like PLG motion.
- He now believes the company should have added more enterprise sales sooner.
- The company has around 100 quota-carrying reps, which he feels is small relative to the opportunity.
PLG + enterprise sales is the winning combo
He sees the best model as:
- Product-led adoption to get started quickly
- Enterprise sales layered on top once usage expands
- Broad use-case coverage that helps one customer expand across multiple workflows
He’s very focused on roadmap and execution
A recurring theme is the importance of long-term roadmap planning:
- He credits Mark Benioff with the lesson:
“You overestimate what you can achieve in one year and underestimate what you can achieve in five.” - ClickHouse is already shipping ahead of schedule and entering new product categories faster than expected.
- He says the company is building for a 20-year outcome, not a two-year valuation cycle.
Product, Acquisitions, and Competitive Moat
Build vs buy is driven by strategic fit
Katz explained ClickHouse’s acquisition strategy:
- Build organically where the team has confidence and differentiation.
- Acquire when founders are building in areas that fit the long-term platform vision.
- He cited Langfuse as a recent example in agent observability.
The real competitor is the company not yet in the market
A key defensive mindset for him:
- He worries less about existing competitors than about the next disruptive entrant.
- ClickHouse itself was once the “company behind the open source project” that no one expected to become so large.
- His goal is to keep reinventing the company before someone else does.
Brand, Sponsorship, and International Growth
Brand still matters in an agentic world
Even if agents influence more technical choices, Katz says humans will remain in the loop for a long time.
- Budgets are still controlled by people.
- Enterprise architecture decisions still involve human judgment.
- That means developer relations, community, and brand remain important.
The Fulham sponsorship is a strategic awareness play
Since the episode was filmed at Craven Cottage, the conversation also covered ClickHouse’s Fulham sponsorship.
- Katz sees the sponsorship as both brand awareness and hospitality.
- He uses events and brand activations to reach senior enterprise buyers and engineers.
- He views sports sponsorships as valuable because they create memorable, high-touch experiences.
Hiring, Work Culture, and Remote Work
ClickHouse is distributed, but increasingly hybrid
Katz described the company as globally distributed by necessity.
- The company operates across 27 countries.
- Over half of revenue comes from outside the U.S.
- ClickHouse is expanding office presence in multiple regions, not forcing a single HQ model.
He is pro-experience, not pro-youth for its own sake
His hiring philosophy favors:
- Engineers with deep distributed systems experience
- People who can operate in complex infrastructure environments
- Mentorship and pairing younger hires with experienced builders
On college, he’s mixed but skeptical of some ROI
He thinks college is still valuable for many reasons, but:
- The curriculum often lags the market
- Software engineering may be especially uncertain by the time a student graduates
- He still believes medicine and some specialized fields justify formal education more strongly
Favorite Forward-Looking Predictions
Jobs that will emerge
He expects a new role to become common soon:
- An AI finance function dedicated to monitoring AI consumption, tokens, budgets, and infrastructure spend
What will matter in 3–5 years
Katz highlighted several future shifts:
- Agents will need identity and authorization
- Enterprises will demand more on-prem and private deployment options
- Specialized models will exist, but frontier labs will remain dominant
- Infrastructure selection may increasingly be agent-driven
Final Thoughts
Aaron Katz’s worldview is highly optimistic about AI, but grounded in infrastructure realities. He believes the market is expanding faster than skeptics admit, that enterprise adoption is still early, and that the real winners will be the companies that combine strong product, durable infrastructure, and disciplined long-term execution.
His biggest messages:
- AI is not a bubble; it is a massive platform shift.
- Revenue durability matters more than vanity metrics.
- Enterprises still want trust, governance, and deployment flexibility.
- Agents will reshape how software is built, bought, and operated.
- ClickHouse is positioning itself to be the database layer for that future.