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

ClickHouse创始人兼首席执行官亚伦·卡茨（Aaron Katz）在富勒姆足球俱乐部（Fulham Football Club）的克拉文农场球场（Craven Cottage）录制的一期20VC节目中，与哈里·斯特宾斯（Harry Stebbings）进行了交流。这个独特的地点反映了ClickHouse近期对该俱乐部球衣胸前的赞助。卡茨介绍ClickHouse是全球最受欢迎的开源分析型数据库，以闪电般快速的查询和资源效率著称，被Anthropic、OpenAI、特斯拉（Tesla）和Netflix等领先的AI原生公司所使用。ClickHouse拥有惊人的营收轨迹：从零起步，达到1200万、5000万、2亿美元，并预计本年度将以超过5亿美元的年度经常性收入（ARR）收官，目标是在2027年12月之前达到10亿美元ARR。 卡茨分享了他对当前AI周期的看法，指出与以往的技术变革（互联网、移动、社交）相比，其加速达到了“前所未有的速度”。这种快速增长转化为对系统的新要求以及“我们有生之年从未见过”的营收规模化。尽管承认一些AI公司毛利率较低的担忧，但他认为，考虑到快速增长和健康的资产负债表，只要有明确的利润率扩张路径，目前这是可以接受的。对于投资者，他强调营收的*持久性*，将ClickHouse等基础设施软件的高转换成本，与模型提供商快速相互超越的“代理应用”可能存在的低转换成本进行了对比。 在内部，ClickHouse经历了“AI觉醒”，导致对Anthropic的支出增加了100倍。卡茨优先考虑持续的营收增长而非严格的token预算，指出token成本正在下降，并且AI工具正在加速他们的路线图，使他们能够更快地推出产品，并提前数年进入新类别。他反思ClickHouse曾“过于高效”，尤其后悔未能更早地在销售能力上进行更多投资。他最初的策略受到Datadog的PLG（产品驱动增长）方法启发，专注于产品和工程，但他现在看到了引入企业销售模式的必要性，并借鉴了他在马克·贝尼奥夫（Marc Benioff）领导下在Salesforce工作12年的经验。 展望未来，卡茨强调了从为人构建到为自主代理构建的转变。代理没有特定人格，要求低延迟，并表现出不可预测的探索性查询模式。他设想，未来代理需要身份、预算和授权才能消费服务，并且它们最终将为自己构建的应用程序选择底层基础设施堆栈（如ClickHouse）。他预言，尽管一些专业模型将蓬勃发展（例如法律科技），但大型前沿实验室仍将保持主导地位。他有争议地提出，企业采用应在开源权重模型和前沿模型之间各占50%，理由是对开源权重模型的法律保护和赔偿存在担忧，尤其是那些源自中国的模型。 卡茨指出，包括数字原生公司在内的企业，由于隐私、合规性和安全担忧，正在考虑重新转向本地部署的趋势。这要求ClickHouse支持多种部署模式。他最大的竞争担忧是“下一个ClickHouse”——一种从“后视镜”中出现的颠覆性技术，正如ClickHouse本身颠覆了现有企业一样。 关于融资，卡茨透露ClickHouse资产负债表上拥有10亿美元，并不*需要*最近的资本注入，但他很重视大卫·萨克斯（David Sacks）、迈克尔·戴尔（Michael Dell）和摩根大通（JP Morgan）等战略投资者。他将融资视为建立一家持久、长寿、世代相传的公司的一种手段，而不仅仅是为了短期估值飙升。他预计将在“未来两年内，甚至更早”达到10亿美元的ARR。 在AI领域的人才招聘充满挑战，但ClickHouse利用其在27个国家的全球布局来招聘经验丰富的数据库工程师，并保持个位数的流失率。尽管承认面对面协作的价值，但该公司分布式特性是由支持不同时区和语言客户的需求驱动的，这使得集中式枢纽不切实际。 卡茨对医疗进步，尤其是在慢性病方面的进展感到兴奋。关于ClickHouse的未来，他认为没有上市的必要，相信公司在未来几年内无论上市与否都将蓬勃发展。他指出，公开市场目前“不理性”，保持私有可以保护公司免受波动大的股价和卖空者的影响，而结构化二级市场交易则为员工提供了流动性。他承认上市带来的公众认知度和用于并购的可交易货币的好处，但他认为Stripe和OpenRouter等大型私有公司现在正在证明，即使不上市，也可能进行重大收购。

Aaron Katz, founder and CEO of ClickHouse, joined Harry Stebbings for a 20VC episode recorded at Craven Cottage, Fulham Football Club – a unique setting reflecting ClickHouse's recent front-of-shirt sponsorship. Katz introduced ClickHouse as the world's most popular open-source analytical database, known for lightning-fast queries and resource efficiency, used by leading AI-native companies like Anthropic, OpenAI, Tesla, and Netflix. ClickHouse boasts an astounding revenue trajectory: zero, 12, 50, 200, and finishing the current year north of $500 million ARR, aiming for $1 billion ARR before December 2027. Katz shared his perspective on the current AI cycle, noting its "unprecedented pace" of acceleration compared to previous tech shifts (internet, mobile, social). This rapid growth translates into new demands on systems and revenue scaling "unlike anything we've seen in our lifetime." While acknowledging concerns about lower gross margins in some AI companies, he believes they are acceptable for now, given the rapid growth and healthy balance sheets, provided there's a clear path to margin expansion. For investors, he emphasizes revenue *durability*, contrasting the high switching costs of infrastructure software like ClickHouse with the potentially low switching costs of "agentic applications" where model providers are rapidly leapfrogging each other. Internally, ClickHouse experienced an "AI awakening," leading to a 100x increase in Anthropic spending. Katz prioritizes sustained revenue growth over strict token budgeting, noting that token costs are declining, and AI tooling is accelerating their roadmap, allowing them to ship products faster and enter new categories years ahead of schedule. He reflected that ClickHouse has been "too efficient," specifically regretting not investing more in sales capacity earlier. His initial strategy, inspired by Datadog's PLG approach, focused on product and engineering, but he now sees the necessity of layering in an enterprise sales motion, drawing lessons from his 12 years at Salesforce under Marc Benioff. Looking to the future, Katz highlighted the shift from building for humans to building for autonomous agents. Agents don't have personas, demand low latency, and exhibit unpredictable, exploratory query patterns. He envisions a future where agents need identity, budget, and authorization to consume services, and where they will ultimately select the underlying infrastructure stack (like ClickHouse) for applications they build. He predicts that while some specialized models will thrive (e.g., legal tech), large frontier labs will remain dominant. He controversially suggests a 50-50 split between open-weight and frontier models for enterprise adoption, citing concerns about legal protections and indemnification with open-weight models, particularly those originating from China. Katz noted a trend of enterprises, including digital-native companies, considering moving back to on-prem deployments due to privacy, compliance, and security concerns. This requires ClickHouse to support multiple deployment models. His biggest competitive fear is "the next ClickHouse" – a disruptive technology emerging from the "rearview mirror," as ClickHouse itself disrupted incumbents. On fundraising, Katz revealed that ClickHouse had a billion dollars on its balance sheet and didn't *need* the recent capital infusion but valued the strategic investors like David Sacks, Michael Dell, and JP Morgan. He views fundraising as a means to build a durable, long-lasting, generational company, not just for short-term valuation bumps. He anticipates hitting $1 billion ARR "within the next two years, if not sooner." Talent acquisition in the AI world is challenging, but ClickHouse leverages its global distribution across 27 countries to hire experienced database engineers, maintaining single-digit attrition. While acknowledging the value of in-person collaboration, his company's distributed nature is driven by the need to support customers across diverse time zones and languages, making centralized hubs impractical. Katz is excited about medical advancements, particularly in chronic conditions. Regarding ClickHouse's future, he sees no rush to go public, believing the company will thrive regardless of its public status in the next few years. He noted that public markets are currently "irrational," and being private offers protection from volatile stock prices and short-sellers, while structured secondaries provide employee liquidity. He acknowledges the benefits of public awareness and a tradable currency for M&A but argues that large private companies like Stripe and OpenRouter are now demonstrating that significant acquisitions are possible without being public.