Aaron Katz

1. Key Themes

AI Revenue Growth Is Unprecedented — And This Is Not a Bubble

Aaron argues that the current AI cycle is categorically different from prior tech waves. Unlike internet, mobile, or social, the pace of both product maturity and revenue growth is faster than anything he has witnessed across a 25+ year career.

"Those cycles, in my experience, were much more gradual. This seems to be accelerating at an unprecedented pace in terms of how quickly these agentic experiences are maturing and how quickly these companies are growing. I mean, we haven't seen revenue growth like this in our lifetime." 00:04:57

"A widely held belief around AI that's generally agreed upon but is wrong is that it's overblown and that we're in a hype cycle, that we're in a bubble... We're just getting started." 00:52:31

Revenue Durability and Switching Costs Are the Most Undervalued Investment Metrics

Aaron repeatedly returns to the idea that speed of revenue growth is less important than durability. For infrastructure, switching costs are high; for agentic applications, they are dangerously low. He frames this as the single biggest risk for investors in AI application companies today.

"If you were to say, what's the single biggest risk — it 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 agentic applications." 00:06:42

"For any category that goes from zero to a hundred million in a year, I worry what's the competitive moat that they have to preserve that hundred million from that customer going to something else." 00:45:02

Agents Are the New Buyers of Infrastructure — Identity, Budget, and Authorization Are the Missing Primitive

Aaron lays out a specific vision: in three to five years, agents will autonomously provision entire infrastructure stacks. The critical gap today is that agents lack identity, budget, and authorization mechanisms. He sees this as a massive unsolved problem and implicit investment opportunity.

"Anthropic, if you ask them 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. I'm thinking about a future where they say, hey, we need to build an application. Provision the underlying stack... And the agent's actually making that selection process. But they need to have authorization. They need an identity. They need to have a budget to be able to consume those services." 00:17:34

"If I were in your shoes, I'd be thinking about what companies are best positioned to give that agent everything they need to build a software application." 00:18:02

Enterprise Enterprises Are Adopting Faster Than Ever — Sales Cycles Are Compressing

Historically, selling into large financial services firms took years. Aaron is seeing that compress to quarters, aided by open source and product-led growth, and validated by conversations at Canary Wharf with some of Europe's largest banks.

"I was at Canary Wharf yesterday meeting with some of the largest financial services companies in the world. They are leading the way in terms of their adopting new technologies in a way that I've never seen in the past. A lot of the times, historically, the sales cycles into these big firms would be measured in years, not quarters." 00:28:28

The Frontier vs. Open Weight Model Debate Is Misunderstood — Enterprise Will Not Defect to Open

The popular thesis is that 90% of tokens will flow through open models and frontier models will be reserved for rare, high-stakes tasks. Aaron rejects this — especially for enterprise — citing indemnification concerns, output inference risk, and distrust of Chinese open weight models.

"I don't. 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." 00:23:59

"I think when you need the legal protection that most enterprises do, you're going to want to work with one of the frontier lab providers. I think there's too much security concern around some of these open weight models." 00:26:28

The On-Prem Renaissance: Even Digital Natives Are Moving Back

Contrary to the prevailing cloud-first narrative, Aaron is seeing a meaningful reversal — innovative Silicon Valley digital natives are seriously considering moving workloads back on-prem, driven by data sovereignty, compliance, and trust concerns.

"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 one of the hyperscalers to an on-prem environment." 00:27:53

PLG + Enterprise Sales Layering Is the Winning GTM Formula — But Timing Matters

Aaron explicitly benchmarked ClickHouse's GTM against Datadog (PLG) and Snowflake (enterprise sales) and chose the Datadog path first. His regret is not layering on enterprise sales capacity sooner. He now views the sequencing as PLG-first, then enterprise overlay.

"I just thought it was going to be a lot easier to follow the Datadog playbook than the Snowflake playbook... But at some point, you need to layer in an enterprise sales motion on top of some sort of PLG distribution." 00:10:46

"The one thing, and if I look back over the last two years, that I wish I had done differently was increasing sales capacity." 00:10:01

Sports Sponsorships as Enterprise GTM — Hospitality Drives Pipeline

Aaron frames the Fulham shirt sponsorship not primarily as brand awareness but as a measurable pipeline generation tool. Twenty C-level executives at a Michelin-level dinner, half customers, half prospects — directly attributable to that event.

"We had 20 executives last night attend an intimate Michelin grade dinner. You know, the C-level executive came from Paris, one of the largest banks in Europe, just to experience that. Those types of relationships are extremely important, especially as we move up market." 00:32:47

The Private Company Advantage Is Real and Growing — The Short Seller Absence Matters

Aaron makes a specific, underappreciated case for staying private: the absence of short sellers materially changes the management environment. Employee morale is protected, and structured tenders now replicate the liquidity benefit of being public.

"There's really two dimensions that don't apply. You don't have your employees looking at your stock price every day and you don't have anybody shorting your company. Those are two material impacts of being a public company versus being a private company." 00:58:58

Distributed Engineering at Scale Is a Competitive Advantage, Not a Liability

ClickHouse operates across 27 countries, 36 cloud regions, with over half of revenue outside the US. Aaron argues this is structurally required when you support customers globally, and is a genuine talent and market access advantage rather than a cultural weakness.

"Over half of our revenue comes from outside of the US. 40% here in EMEA, 10% in Asia... There's no way that you can centrally manage that from one location. You need to have people in every single time zone." 00:40:53

2. Contrarian Perspectives

The 90/10 Open-vs-Frontier Token Split Is Wrong — It Will Be Closer to 50/50

The prevailing venture and media consensus is that open models will dominate token volume, with frontier models reserved for a handful of critical tasks. Aaron explicitly rejects this and draws an analogy to the open source vs. proprietary software split in enterprise, which has settled roughly even.

"The easy answer is 50-50. In the same way that what percentage of enterprise software today is open source versus proprietary? I think it's a pretty even distribution." 00:22:30

"I don't. I don't. 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." 00:23:59

Revenue Concentration Above 10% in Any Single Customer, Category, or Sector Is a Hard Red Line — Even for NVIDIA-Like Stories

The prevailing investor sentiment accepts high concentration in high-conviction winners (e.g., "NVIDIA is 90% data center — that's fine"). Aaron takes the opposite operating stance: if any single dimension exceeds 10% of revenue, it is a material concern, not a feature.

"If I've got one customer, or one category, or one industry that accounts for more than 10% of revenue, I spend a lot of time thinking about it... If some dimension of your revenue base accounts for more than 10% of your revenue, you've got exposure. And I want to limit exposure." 00:46:21

Anthropic and OpenAI Could Become Core Infrastructure Providers — Not Just Model Companies

Conventional wisdom treats the frontier labs as model vendors and assumes infrastructure categories (databases, networking, compute) remain with specialists. Aaron explicitly does not dismiss Anthropic and OpenAI entering infrastructure.

"I wouldn't dismiss Anthropic and OpenAI as core infrastructure providers. Really? Yeah, not at all. I don't see them as competition today. But you see how they're entering new categories so quickly. Just imagine the surface area they're going to cover in three years. They're building their own chips." 00:22:59

The Biggest Competitive Threat Is the Company Not Yet in the Market

Most founders obsess over known competitors. Aaron's real concern is the unknown disruptor — the next ClickHouse — because ClickHouse itself was the thing nobody saw coming.

"What I worry about is the technology coming from the rearview mirror. I worry about the next ClickHouse. Like people really didn't see this technology coming... I worry about what's the company that's going to disrupt us in the same way that we're disrupting the competitors in front of us." 00:21:20

Large Enterprises Using Open Chinese Models for Their Most Sensitive Data — And That's a Real Phenomenon

Harry expected this to be the reverse, but Aaron confirms it: some large enterprises are using Anthropic for less sensitive workloads, and open source Chinese models for more sensitive data because they can run it on-prem behind their own firewall and don't have to trust a vendor's "zero data retention" claim.

"I think a lot of companies worry about sending their source code, for example, to a frontier lab... You worry about the output from that code generation. You worry about third party indemnification... So what do you do in that case? You then go to an open model. You limit the use cases." 00:25:26

3. Companies Identified

ClickHouse

World's most popular open source OLAP database, used by nearly every major AI-native company. Known for extreme query speed, resource efficiency, and the ability to ingest a billion events per second (Tesla use case). Currently at $350M+ ARR, growing to $500M+ this year, targeting $1B by December 2027 or sooner. 200%+ net dollar retention, 99%+ gross retention, 4,000+ customers, valued at $15B+.

"We went zero, 12, 50, 200, and we'll finish this year north of 500, which in the database world is faster growth than we've ever seen, including all of the competitive companies." 00:45:02

"Tesla, for example, is ingesting a billion events per second into ClickHouse. That throughput is unprecedented." 00:15:53

Anthropic

Frontier AI lab, customer of ClickHouse. Notable detail: Anthropic chose ClickHouse for observability by asking Claude, which recommended it. ClickHouse's Anthropic spend is up 100x from the beginning of the year.

"Our Anthropic spend is up 100 times from what it was at the beginning of the year." 00:07:15

"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." 00:17:34

Datadog

Cited as the gold standard for PLG-driven infrastructure GTM. Aaron benchmarked ClickHouse's go-to-market against Datadog's developer-led, self-serve model. Positioned as likely winning the observability category over open source alternatives.

"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." 00:10:15

Databricks

Cited as executing "extraordinarily well" and positioned as the likely winner in data warehousing over Snowflake due to its open source foundation. Aaron has significant respect for CEO Ali Ghodsi.

"It's well documented that Databricks is executing extraordinarily well in the market. I've got a ton of respect for Ollie and the company." 00:55:42

Harvey

Named as the leading specialized legal AI company. Cited as example of a category where specialized models make sense, while Aaron believes frontier labs will still dominate most enterprise use cases.

"If you look at Harvey, for example, I know they're, I think, leading the category in terms of specialization." 00:19:30

Langfuse

Berlin-based agent observability startup with three co-founders. Acquired by ClickHouse earlier in 2025. Called out as entering a category — agent observability — that every enterprise in the world will eventually need.

"Our most recent one earlier this year was Langfuse out of Berlin, which is three incredible founders entering agent observability. Every enterprise in the world is going to need this technology." 00:13:21

Fireworks AI

Inference provider at ~30-35% gross margins. CEO Lin mentioned as a mutual contact. Cited in the context of whether the AI infrastructure world will permanently operate at lower gross margins.

"We both know Lin at Fireworks... yeah, we're 30%, 35%. We hope to be more over time." 00:19:02

Snowflake

Named as the primary comparison point for enterprise-first GTM (heavy sales, high cost). Aaron chose not to follow their playbook initially. Still a significant competitor in data warehousing.

"Snowflake went heavy after the enterprise through very expensive sales and marketing." 00:10:15

Salesforce

Where Aaron spent 12 years under Mark Benioff. Used as primary framework for enterprise software GTM, vision-setting, and the lesson of overestimating one-year outcomes while underestimating five-year outcomes.

"You can overestimate what you can achieve in one year and underestimate what you can achieve in five." 00:11:03

OpenAI

Frontier lab and ClickHouse customer. Aaron does not dismiss OpenAI as a future infrastructure provider and sees them entering new categories at rapid pace.

"I wouldn't dismiss Anthropic and OpenAI as core infrastructure providers... They're building their own chips." 00:22:59

Weights & Biases

Named as a ClickHouse customer and AI-native company using the platform.

"It's used by nearly every AI native company, Anthropic, OpenAI, Weights and Biases." 00:04:14

Sierra

Named as an AI company built on ClickHouse, part of the basket of AI customers representing under 12% of revenue.

"...from Harvey, Sierra, Decagon, Anthropic, OpenAI, etc. represents less than 12% of revenue." 00:45:30

Decagon

Named as an AI company built on ClickHouse. Same context as Sierra above.

"...from Harvey, Sierra, Decagon, Anthropic, OpenAI, etc. represents less than 12% of revenue." 00:45:30

Stripe

Cited as a private company conducting extremely large M&A — acquiring a business for $50-60 billion — demonstrating that the traditional public company currency advantage for acquisitions has largely been neutralized.

"Stripe is buying PayPal for 50 to 60 billion dollars as a private company." 00:59:27

Cerebras

Named as a chip manufacturer that ClickHouse does business with, alongside NVIDIA. Cited in the context of a more distributed chip ecosystem emerging.

"We do business with both NVIDIA and Cerebras and a lot of the chip manufacturers." 00:51:44

Elastic

Aaron's previous company as a leading executive. Used as context for his understanding of open source infrastructure GTM and the risk of hyperscalers redistributing your open source as a managed service.

"You said about kind of switching costs there... especially given your time with Elastic." 00:47:21

Alibaba

Named as a cloud partner. Aaron personally flew to China to launch a partnership with them, underscoring the importance of in-region relationships.

"I flew to China to launch a partnership with Alibaba." 00:41:22

4. People Identified

Co-founder and CEO of ClickHouse. Previously 12 years at Salesforce under Benioff, then senior executive at Elastic. Built ClickHouse from 0 to $350M+ ARR with ~800 employees across 27 countries. Praised by investors for consistently outperforming targets. Notable for disciplined capital allocation, cutting back investors from rounds, and maintaining 99%+ gross retention and 200%+ NRR.

"We need to get to a billion dollars of ARR as quickly as possible. I'd put the over-under at December 2027, and I would take the under." 00:00:00

Mark Benioff

Founder and CEO of Salesforce. Aaron worked under him for 12 years. Described as a visionary marketer who created enterprise perception ahead of product reality, then delivered on the roadmap. Key lesson cited: overestimating one-year outcomes, underestimating five-year outcomes.

"Mark had this bigger vision. And he said, we're going after Siebel, SAP, Oracle, Microsoft. We didn't have the product set to go after those competitors. But he was such an incredible marketer that he created this perception in the industry that some of the largest companies in the world could adopt this technology." 00:11:32

Yuri and Alexei (ClickHouse Co-Founders)

Engineers and co-founders of ClickHouse. Aaron describes them as the best engineers he has worked with in his career by a wide margin.

"The credit really goes to Yuri and Alexei. My two co-founders are spectacular. The best engineers I've worked with in my career by a very wide margin." 00:35:42

Peter Fenton

Board member at ClickHouse (and prior company). Described as a career venture capitalist with exceptional pattern recognition and talent magnetism. Aaron uses him selectively — either to recruit or to critically evaluate candidates — and briefs him on which mode to operate in.

"Peter is a career venture capitalist, and he's helped shape and form some of the most influential and impactful companies in technology. And so he has this amazing pattern recognition... he's an incredible talent magnet." 00:53:39

Mike Volpe

Board member at ClickHouse. Former Cisco operator who ran corporate development and completed over 100 acquisitions. Contrasted with Peter Fenton as the operator vs. investor board archetype.

"Mike was an operator, worked at Cisco for a long time, ran CorpDev. I think he did over 100 acquisitions." 00:53:39

Ali Ghodsi (Ollie)

CEO of Databricks. Aaron expresses significant respect for his execution. Cited a dinner conversation where Ali described running Databricks like a public company internally — no stock price distraction, no short sellers — which influenced Aaron's thinking about staying private.

"It's well documented that Databricks is executing extraordinarily well in the market. I've got a ton of respect for Ollie and the company." 00:55:42

"I asked him this question. I'm like, it feels like you guys are ready to go public. And he said, I kind of basically run a public company, and he walked me through that." 00:58:28

Martin Casado

Partner at Andreessen Horowitz. Named as a friend of Aaron's who deeply understands what ClickHouse does technically. Was conflicted at the time of ClickHouse's early fundraise and could not invest. Aaron describes this as a missed opportunity for both sides.

"I met with Martin Casado at Andreessen who's a friend of mine and I would have loved to have him involved because I think he understands what we do in a very unique way technically. Why is he not involved? He was conflicted at the time." 00:54:42

David Sachs

Named as a strategic investor in ClickHouse's recent financing round. Aaron values the affiliation.

"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." 00:31:45

Michael Dell

Named as a strategic investor in ClickHouse's recent financing round alongside David Sachs.

"We extended it and brought in some strategic investors like yourself and David Sachs and Michael Dell and JP Morgan." 00:31:45

Satya Nadella

CEO of Microsoft, which Aaron describes as one of ClickHouse's most important enterprise customers — powering the largest analytical workloads at Microsoft. Aaron met Satya a few months prior and was impressed.

"Microsoft's one of our power users. We power the largest analytical workloads at Microsoft... I had the opportunity to meet Satya a couple months ago. It was quite impressive." 00:51:44

Lin (CEO of Fireworks AI)

CEO of Fireworks AI. Mentioned in the context of AI infrastructure gross margin profiles. Appeared previously on the 20VC show discussing their 30-35% gross margins.

"We both know Lin at Fireworks. I think every company will have their own model trained on their own data." 00:19:02

Vinod Khosla

Named in the context of acquiring a sports franchise (the Seattle Seahawks) — cited as a savvy businessman making what Aaron believes will be a good investment.

"Well, Vinod's a very savvy businessman, so I'm sure it's going to be a good investment for him." 00:33:48

Jason Lemkin

Co-host of a separate weekly show Harry does. Cited for the quote that if you are not well into your 2027 product roadmap already, you are behind. Used to prompt Aaron on AI-driven development acceleration.

"Jason said last week that if you are not well into your 2027 roadmap already, you are behind." 00:11:53

5. Operating Insights

The "AI Awakening" Email as a Cultural Reset Tool

When Aaron noticed his company was not adopting AI coding tools fast enough, he sent a company-wide email with the subject line "The AI Awakening" — explicitly naming the gap and calling the company to act. The result was an explosive uptake internally. This is a replicable leadership tactic: name the lag explicitly, frame it as a company-level imperative, and let the organization rally. The signal he tracked was Anthropic spend, which went up 100x within the year.

"I sent an email to the company and the subject was the AI awakening. And I said, I think we're moving too slowly and I'm not seeing the adoption of these coding applications, for example, that I would expect to see for a leading database provider like ClickHouse. And the company rallied to the call." 00:07:15

Brief Investors Before Board Interactions — Buy or Sell Mode

Aaron has a specific protocol for using board members in talent conversations: he tells Peter Fenton before any call whether the goal is to recruit (sell) or evaluate (critically assess) the candidate. This prevents a talented board member from unconsciously sandbagging a recruiting effort or over-selling a candidate who needs rigorous vetting.

"I typically tell him whether or not he's buying or selling, whether or not he's trying to convince this person to join the company or really evaluating this person critically. And that's going to influence, I think, how he approaches that conversation." 00:54:08

Reference Investors the Same Way You Reference Candidates

Rather than relying on VC pitches, Aaron references investors through their portfolio companies — asking specifically about customer introductions made, social media advocacy, recruiting help, and whether co-investors view them positively. He requires unanimous positive answers before proceeding.

"I reference them like you would in any other relationship. So I talk to the companies that they've invested in in the past. I say, what's it like to work with Harry? What customer relationships has he made that have been valuable? How has he helped with awareness from his social presence? Has he helped with recruiting?... Those all need to be a unanimous yes before we start working together." 00:36:09

10% Revenue Concentration Rule — Applied to Any Dimension

Aaron runs a hard internal rule: if any single customer, category, or industry exceeds 10% of revenue, it demands active management. Even a category as fast-growing as AI companies collectively represents under 12% of ClickHouse revenue, by design. This is the operational expression of his durability-first philosophy.

"If I've got one customer, or one category, or one industry that accounts for more than 10% of revenue, I spend a lot of time thinking about it... If some dimension of your revenue base accounts for more than 10% of your revenue, you've got exposure." 00:46:21

Enterprise Events at Sports Venues: Structure Them as Intimate Pipeline Events, Not Just Awareness

The Fulham sponsorship is not primarily measured as a brand impression play. Aaron structured the matchday as a deliberate pipeline event — 20 C-suite executives, half existing customers and half prospects, at a Michelin-level dinner. The mix is intentional: existing customers validate, prospects experience. Attribution to sponsorship is still imprecise, but the pipeline basket is directly measurable.

"We had 20 guests last night, budget owners from some of the largest companies in the world. Half of those are customers. Half of those are prospective customers. So I can very easily measure the spend that I can gather from that basket of accounts over the next 12 months." 00:35:01

6. Overlooked Insights

Agent Identity, Budget, and Authorization Is the Next Major Infrastructure Category — and Nobody Is Building It Yet

This was mentioned briefly as a throwaway observation, but it is actually a precise infrastructure gap description with massive investment implications. Aaron is articulating that as agents move from human-supervised to fully autonomous over the next three to five years, there must be a layer that gives agents an identity (who is this agent?), a budget (how much can it spend?), and authorization (what is it permitted to do and access?). He explicitly says "we're not there yet today" and frames it as an investment question. No specific company was named as solving this — meaning it is an open category.

"Agents need to have an identity and a budget... Most companies aren't just going to let their agents run wild and build whatever they want and consume as many resources as the agent deems fit. There's going to need to be some sort of governance and oversight with that consumption... If you look out three to five years, those agents are going to be fully autonomous." 00:17:34

"If I were in your shoes, I'd be thinking about what companies are best positioned to give that agent everything they need to build a software application." 00:18:02

The New Job That Will Exist in Five Years: AI Finance Function Dedicated Solely to Token Consumption

Aaron named this in the quick-fire section and it passed without elaboration — but it is a significant organizational design prediction. He envisions a dedicated finance function inside enterprises whose entire mandate is managing AI token consumption and resource budgets across the organization. Critically, he also predicts this role becomes automated by agents within a decade. For investors and operators, this implies near-term opportunity in AI cost management and FinOps tooling for LLM consumption — a category that barely exists today.

"An AI finance function solely dedicated on AI consumption inside of an organization. That's all they wake up thinking about. In terms of token resource management. Correct. And then that job will be made irrelevant five years from then because AI agents will govern themselves." 00:50:57