The Twenty Minute VCClickHouse CEO: AI Margins Need to Improve | Revenue Concentration Should be a Concern
CHAPTERS
- 0:00 – 1:53
ClickHouse in 2026: open-source OLAP at $350M+ ARR and aiming for $1B
Harry and Aaron set the context: ClickHouse as a lightning-fast, resource-efficient analytics database used by many AI-native leaders. Aaron frames the company’s growth trajectory, ambition to reach $1B ARR quickly, and why the business is positioned as core infrastructure for the AI wave.
- •ClickHouse’s value prop: fast queries + efficient storage at massive scale
- •Adoption by AI-native companies (e.g., Anthropic/OpenAI mentioned)
- •ARR scale and rapid growth expectations; confidence in IPO readiness
- •Vision: becoming a default data layer for modern/agentic applications
- 1:53 – 3:50
Are we early in AI? Why this cycle feels faster than internet/mobile/social
Aaron argues the AI cycle is still in the early innings, but unlike prior platform shifts it’s compressing timelines dramatically. He highlights unprecedented company growth rates and the unique infrastructure demands created by agentic products.
- •AI adoption and revenue scaling are accelerating faster than prior tech waves
- •Agentic experiences are maturing quickly, stressing systems in new ways
- •Builder perspective vs investor “bubble” narratives
- •Infrastructure requirements are shifting with explosive query volume and complexity
- 3:50 – 4:47
AI investing’s biggest risk: revenue durability in low-switching-cost apps
Aaron identifies durability of revenue as the key risk, especially for agentic applications where switching costs can be low. He contrasts infrastructure stickiness with rapidly changing model/provider landscapes that can erase app-level advantage quickly.
- •Infrastructure software tends to have high switching costs; apps often don’t
- •Model providers leapfrog each other quickly, destabilizing downstream revenue
- •Investor focus should shift from growth alone to durability of that growth
- •Even dominant products can be vulnerable if switching is easy
- 4:47 – 6:16
Anthropic spend, ‘AI Awakening,’ and model routing: frontier + open weights
Aaron describes ClickHouse’s internal push to adopt AI coding tools and the resulting surge in Anthropic usage. They discuss independence vs being tied to a single harness provider, and why teams may route different tasks to different model types.
- •ClickHouse’s internal mandate to accelerate AI tool adoption
- •Anthropic spend increasing massively; practical usage driving conviction
- •Routing: open-weight models for some workflows (e.g., review) vs frontier for production
- •Concern areas: inference quality, safety, and suitability for production code
- 6:16 – 7:25
Token spend and AI ROI: why roadmaps are being pulled forward by years
The conversation turns to budgeting and ROI when usage can explode. Aaron argues that if revenue growth is strong and tokens get cheaper, it’s rational to over-invest, then rein in later—especially because AI tooling accelerates delivery and category expansion.
- •Primary internal yardstick: sustained revenue growth, not perfect token efficiency
- •Token costs trending down; productivity and roadmap speed trending up
- •AI tooling accelerates feature velocity and surface-area coverage
- •Control lever: rein in consumption later if needed
- 7:25 – 9:09
Scaling go-to-market: Datadog vs Snowflake playbooks and sales capacity regrets
Aaron explains why ClickHouse initially prioritized product/engineering pressure and favored a PLG motion, then later layered enterprise sales. He reflects that increasing sales capacity earlier is the main thing he would change given competitive ‘armies’ of sellers.
- •Early emphasis on product-led growth; later need for enterprise motion
- •Competitors have thousands of sellers vs ClickHouse’s ~100 quota carriers
- •Rep productivity is high, but capacity constrained against enterprise incumbents
- •Lesson: timing and sequencing of PLG + enterprise is crucial
- 9:09 – 10:29
Salesforce lessons and the power of long horizons + perception shaping
Drawing from 12 years with Marc Benioff, Aaron shares a core operating lesson: overestimate one year, underestimate five. He links this to roadmap storytelling—building conviction before the full product exists—and delivering into that narrative over time.
- •Benioff lesson: long-term compounding beats short-term planning
- •Marketing and narrative can create permission to win big accounts early
- •Roadmap credibility becomes a strategic asset
- •Execution closes the gap between perception and product reality
- 10:29 – 11:52
Build vs buy: acquisitions, new categories, and agent observability (Langfuse)
Aaron describes ClickHouse’s mix of organic innovation and inorganic moves when founders build compelling businesses on top of ClickHouse. He calls out agent observability as a coming enterprise requirement, using Langfuse as an example of strategic expansion.
- •Decision rule: build if internal teams can win; buy if founders are already scaling on ClickHouse
- •Six acquisitions used to accelerate entry into adjacent categories
- •Agent observability positioned as inevitable for enterprises
- •Concept of ClickHouse evolving toward a broader data platform
- 11:52 – 16:12
When AI agents become the customer: latency, exploration, and stack selection
Aaron explains how software changes when agents—not humans—drive query patterns: no personas, unpredictable exploration, and many parallel queries across systems. He predicts agents will increasingly select underlying infrastructure, forcing vendors to compete on latency and efficiency.
- •Agents traverse many tools; experience is limited by the slowest link
- •Agent query patterns: exploratory, bursty, parallel, and low-latency dependent
- •Pricing and consumption models must adapt to new usage dynamics
- •Future: agents choose the infrastructure stack, not just humans
- 16:12 – 18:00
Agents need identity + budgets: governance as the missing layer of autonomy
They explore what it takes for agents to provision resources safely: identity, authorization, and spend controls. Aaron argues most organizations won’t allow unconstrained autonomous provisioning, so governance and oversight become foundational for the next wave.
- •Agents will need identity, permissions, and budget constraints
- •Provisioning stacks requires authorization and auditable governance
- •Today humans monitor spend; future points toward autonomous controls
- •Investment implication: look for enablers of safe autonomous software building
- 18:00 – 28:01
Enterprise model landscape: specialized models, frontier dominance, and open debates
Aaron predicts a mixed world: specialized models for narrow domains, but frontier labs remain dominant for enterprises. They discuss the difference between open source and open weights, and why indemnification, security assurances, and output concerns affect adoption—especially for Chinese open-weight models.
- •Both specialized and frontier models will coexist; frontier likely dominates enterprise
- •Key enterprise requirements: indemnification, protections, and trust guarantees
- •Clarification: open source software vs open weights models are not the same
- •Security/data-retention skepticism shapes enterprise deployment choices
- 28:01 – 29:30
Deployment shifts and enterprise acceleration: on-prem resurgence + shorter cycles
Aaron notes a surprising trend: even digital-native companies are reconsidering on-prem, driven by privacy, compliance, and control. He also observes enterprises adopting faster than historically, aided by open source and PLG evaluation paths that compress sales cycles.
- •On-prem/VPC demand rising, including from Silicon Valley ‘cloud-first’ companies
- •Multiple deployment models required to serve regulated enterprises
- •Sales cycle compression: from years to quarters in some cases
- •Open source + PLG reduces friction for evaluation and adoption
- 29:30 – 36:26
Moats, hyperscaler risk, and brand in an agentic world (incl. Fulham sponsorship)
Aaron addresses common investor concerns about open-source defensibility and hyperscalers re-packaging products. He argues differentiation comes from cloud execution and hard-to-replicate proprietary capabilities, while brand and awareness still matter because humans control budgets—hence investments like major event marketing and sports sponsorship.
- •Investor misconception: ‘Where’s the moat?’ for open-source businesses
- •Risk: hyperscalers offering managed versions; defense via superior cloud/proprietary features
- •Agents may influence choices, but humans still approve architecture and budgets
- •Brand-building tactics: conferences, flagship events, and Fulham sponsorship ROI logic
- 36:26 – 53:45
Fundraising, scaling to $1B ARR, talent competition, and the concentration question
Aaron explains how he evaluates investors (reference-driven) and why long-term durability matters more than short-term valuation optics. He shares ClickHouse metrics (high retention/NDR), a target timeline for $1B ARR, hiring realities vs frontier labs, and why revenue concentration remains an operator-grade risk even in AI.
- •Investor selection: references, concrete value-add (customers, recruiting, advocacy)
- •Metrics: ~99%+ gross retention, 200%+ NDR, rapid customer adds; $1B ARR ‘under 2027’ bet
- •Talent: frontier labs pay aggressively; ClickHouse recruits different profiles globally
- •Revenue concentration: >10% in a customer/sector is a real risk; ClickHouse’s AI cohort <12%
- 53:45 – 1:05:56
Quick-fire and closing: chips, bubbles, boards, leadership, and public markets
In rapid-fire, Aaron weighs in on AI ‘hype’ skepticism, chip ecosystem distribution, and lessons from board members like Peter Fenton. The conversation closes with thoughts on parenting/partnership parallels to leadership, plus why staying private can be rational despite IPO readiness.
- •Contrarian take: AI isn’t overblown; ‘we’re just getting started’
- •Chips: hyperscalers/frontier labs remain major players alongside new entrants
- •Board dynamics: pattern recognition, recruiting pull, and ‘buy vs sell’ framing
- •Public vs private: volatility, employee morale, and reasons an IPO still matters
