Why Anthropic, Meta, and Tesla All Chose the Same Database | Aaron Katz, ClickHouse

Summary

• ClickHouse entered venture formation with product-market evidence that most startups only earn later. Alexey Milovidov built it inside Yandex in 2009 for petabyte-scale streaming analytics, open-sourced it in 2016, and enterprises were already moving logging, metrics, and warehouse workloads onto it. In August 2021, Katz raised a $50 million seed/pre-seed with “no pitch deck…no product…no customers…and no revenue,” followed quickly by $250 million.
• Its commercial moat is as much distribution design as database performance. ClickHouse Cloud was built serverless with compute-storage separation for bursty workloads, then sold through self-service evaluation, free migration help, and engineer-to-engineer Slack channels intended to put customers in production “before our competitors have even scoped the project.” Katz now reports more than 3,000 cloud customers and hundreds added monthly.
• Katz sees AI adoption as SaaS history replaying at radically higher speed. Salesforce’s on-prem incumbents called cloud a fad; he expects objections to model reliability and enterprise data sharing to be debunked and says AI is being adopted “100 times faster” than SaaS was 20 years ago.
• The investor call is lower software multiples but durable infrastructure demand. Katz does not expect old forward-revenue valuations to return; free cash flow and terminal value now carry more weight, while databases remain “picks and shovels” purchased on price and performance once reliability, durability, security, and scalability are table stakes. He separates Datadog and Snowflake from application SaaS and cites Databricks at $4 billion of revenue, with warehousing and AI workloads each above $1 billion.
• Agents could become the buyers and operators of databases, not merely their recommendation layer. Claude recommended ClickHouse to Anthropic, a European fintech CEO said every LLM he queried did likewise, and Katz imagines agents provisioning ClickHouse plus Postgres automatically. Managed Postgres is in private preview, due for public beta in a few months and general availability by year-end; Katz says the unified stack will be “the world’s fastest Postgres service in the cloud.”
• Datadog faces interface disintermediation, but Katz thinks its near-term product moat is underestimated. Biewald argued agent users and easy custom UIs favor direct ClickHouse; Katz called the logic valid yet described Datadog as premium and formidable over a realistic one-to-three-year horizon. ClickHouse itself once spent seven figures on Datadog before earning revenue and faced an internal “mutiny” when migrating away.
• AI is expanding ClickHouse’s operating ambition without relaxing database-grade controls. Headcount is planned to rise from about 500 to nearly 1,000 this year, AI-assisted code contributions from an estimated 50% to 80% within six months, and agents are aimed at SDR- and CSM-like work. Human review stays because “move fast and break things doesn’t apply to a database.”
• The smooth external trajectory conceals concentrated execution risks. The core team moved from Moscow to Amsterdam six weeks before Russia’s February 24, 2022 invasion. Katz later says he believed $200 million was custodied at U.S. Bank while $100 million was on SVB’s balance sheet; he pulled the $100 million 30 minutes before SVB’s system went down, while the remaining $200 million stayed uncertain for 72 hours. Reliability is P0, and well-capitalized warehouse and observability incumbents are “waking up to the threat.”

Deep dive

1. ClickHouse had product-market proof before ClickHouse Inc. existed

• Katz traces ClickHouse to 2009, when Alexey Milovidov at Yandex could not find an off-the-shelf database able to ingest and store petabytes of streaming service data at the required performance. He built the “clickstream data warehouse”; it spread across Yandex, then was open-sourced in 2016 after seven years.

• Katz encountered it at his prior company through lost workloads: Uber was moving logging to ClickHouse, eBay chose it for a large metrics project, and Disney, Comcast, and Deutsche Bank were adopting it. In early 2021, an investor connected him with Yandex leadership, who asked whether he was interested in building something around a project already drawing contributors and use cases.

• The carve-out took nine months because Yandex was public—Katz recalls a roughly $30 billion market cap—and transferring IP demanded regulatory work. He paired creator Alexey with Yuri Izrailevsky, formerly of Netflix and Google, and himself; in August 2021 they raised $50 million with “no pitch deck…no product…no customers…and no revenue,” during the peak ZIRP period, then another $250 million led by Coatue and Altimeter.

2. A serverless cloud and engineer-led sale converted open-source pull

• The product thesis was specific: fully managed ClickHouse Cloud, serverless, with compute-storage separation so bursty analytical workloads could scale automatically and idle. Thousands already ran the open-source software, while 50 design partners specified what would make them stop self-managing; employing core committers protected control of the roadmap.

• Katz used MongoDB Atlas as a design reference for how an open-source database could become a broad cloud platform, then contrasted Snowflake’s capital-intensive enterprise sales engine with Datadog’s frictionless developer adoption. He chose the latter because an engineer should be able to “evaluate, deploy, scale” and even enter production without sales.

• When help was needed, ClickHouse offered a peer discussion, not a conventional pitch: joint Slack channels with top engineers, Alexey, and committers. The operating promise was to make users successful “before our competitors have even scoped the project,” including migration help from Postgres, Snowflake, or BigQuery without charging services fees.

• Katz now reports more than 3,000 ClickHouse Cloud customers and “hundreds every month.” The sharpest wedge is low-latency, customer-facing analytics such as Weights & Biases’ Weave, while the database also reaches warehousing, observability, cyber back ends, and clickstream analysis—a large surface area crowded with specialized incumbents.

3. AI favors infrastructure demand even as it resets software multiples

• Katz’s analogy comes from 12 years at Salesforce, which he joined in 2002 after the dot-com bust, failed MBA applications, and waiting tables in Lake Tahoe. Incumbents called cloud insecure and unreliable; he sees today’s claims that models are untrustworthy or enterprise data cannot go to an LLM provider as the same “head in the sand” paralysis.

• The timing is the differentiator: Salesforce went public in 2004 with only a handful of enterprise customers, while Katz says AI is being adopted “100 times faster” than SaaS was 20 years ago. His categorical forecast is that the current objections “are going to be debunked if they’re not already.”

• On public markets, Katz does not expect a return to prior forward-revenue multiples; free-cash-flow dynamics and terminal value now matter more than assumed 30%-50% annual growth. He calls the sell-off “a little overblown,” separates Datadog and Snowflake from application SaaS names, and cites Databricks reaccelerating on $4 billion of revenue, with warehousing and AI workloads each above $1 billion.

• Distribution is shifting too: Anthropic asked Claude what infrastructure to use and got ClickHouse, while a European fintech CEO told Katz, “Every LLM I ask” suggests it. Katz’s envisioned endpoint is agents provisioning ClickHouse alongside Postgres; the latter is in private preview, with public beta due in a few months and general availability by year-end. He says the unified data stack will be “the world’s fastest Postgres service in the cloud.”

4. Datadog’s interface moat is vulnerable, but its product moat is real

• Biewald’s pushback—worth keeping—is that if agents rather than developers log into observability products, Datadog’s thoughtful interfaces may matter less; teams can query ClickHouse directly and generate custom interfaces. Katz called that view “valid,” but said three-to-five-year forecasting is impossible and retained a one-to-three-year case for Datadog as a meaningful supplier.

• The defense rests on firsthand failure: Katz’s previous company tried to build a comparable service and, he says, was not successful. At ClickHouse, Datadog was the obvious choice for APM, logs, and metrics under a compressed launch schedule, then “grew like wildfire” until pre-revenue spending reached seven figures—temporarily exceeding ClickHouse’s revenue.

• Replacing it triggered “almost like we had a mutiny”; Katz likened taking Datadog from developers to taking heroin from an addict. The migration was also a dogfooding imperative: he said ClickHouse could not call itself a meaningful observability supplier while using Datadog. ClickHouse eventually migrated to its own stack, reporting cost savings and performance gains, yet Katz still expects Datadog to “survive this” and win AI customers—just not regain its former multiple.

5. AI expands ClickHouse’s headcount plan while shrinking handoffs

• ClickHouse has about 500 employees and plans to finish the year closer to 1,000; AI has not reduced that plan and “may accelerate” it if engineers become 10 times more productive. Katz never wanted SDRs or CSMs, calling the handoffs clumsy and often symptomatic; agents can instead qualify leads, generate demand, and deliver contextual product help.

• Product-side, natural-language requests can generate a regional revenue chart or optimistic, pessimistic, and realistic scenarios from live data. Katz describes using LibreChat, an open-source chat interface ClickHouse acquired last year, with an InfluxDB model and ClickHouse’s MCP server. Engineering-side, he estimates 50% of code contributions are AI-assisted today and expects 80% within six months, but human review remains standard: “Move fast and break things doesn’t apply to a database.”

6. The hockey stick has already met geopolitical, bank, and reliability shocks

• Katz’s condition for the Yandex carve-out was no ties to Russia, including no coding on Russian networks. He persuaded Alexey’s Moscow team to relocate to Amsterdam in January 2022; Russia invaded Ukraine on February 24, roughly six weeks later, forcing the company to untangle Yandex while supporting engineers who had left home and not returned.

• The liquidity scare was sharper: Katz had placed $300 million with Silicon Valley Bank. He was pretty sure $200 million was custodied at U.S. Bank, with SVB still the asset manager, but could not confirm it; he believed $100 million was on SVB’s balance sheet. He sent that amount to the Netherlands; it cleared, and he says SVB’s banking system went down 30 minutes later.

• The eventual government backstop did not erase the 72 hours when the remaining $200 million’s status was unclear. Nor does Katz call the service journey flawless: there have been issues, though “nothing catastrophic,” and making security, reliability, and durability P0 requires sleepless engineering because customer-facing products are “betting your company on our service.”

• The next test is competitive. In warehousing he names Snowflake, BigQuery, Redshift, and increasingly Databricks; in observability, Datadog, Grafana, and Elastic. Katz says he feels ClickHouse is the market leader in real-time analytics, citing customer-facing applications at Vercel, Lovable, Sierra, Decagon, Ramp, Klaviyo, Attentive, Weights & Biases, and LangChain—and says rivals are now “waking up to the threat.”