A database built to answer fast, not store safely
ClickHouse is a column-oriented database: it stores each column of a table together on disk rather than each row, which is slower for looking up one record and much faster for scanning millions of rows to compute a sum or a count. That trade is built for analytics, not transactions. Alexey Milovidov started the project inside Yandex in 2009 to build real-time reports out of raw web traffic data, it went into production for Yandex.Metrica in 2012, and Yandex open sourced it under Apache 2.0 in 2016. [1][2]
The company that commercializes it, ClickHouse, Inc., incorporated in Delaware in September 2021, headquartered in the San Francisco Bay Area, with Milovidov continuing as CTO alongside CEO Aaron Katz and President Yury Izrailevsky. It raised a $50 million Series A from Index Ventures and Benchmark that year, followed by a $250 million Series B at a $2 billion valuation later in 2021, a $350 million Series C in May 2025 at roughly $6.35 billion, and a $400 million Series D in January 2026 announced alongside the Langfuse acquisition. [1][2]
How an analytics database ends up in the AI stack
ClickHouse's relevance to AI is mostly indirect: teams running LLM applications in production generate large volumes of trace data, one row per model call, with prompts, outputs, latency and cost attached, and that is the same kind of write-heavy, scan-heavy workload ClickHouse was built to answer quickly. It is used as a backing store inside AI infrastructure rather than as a model or an agent itself, alongside long-standing customers like Uber, Comcast, eBay and CERN's LHCb experiment that have nothing to do with AI. [1]
Atlas interpretation: That makes ClickHouse a picks-and-shovels company for the current AI buildout in the same sense a cloud provider or an observability vendor is: it profits from AI applications existing and generating data, whether or not any particular one succeeds. [1]
Buying the observability layer instead of building it
On January 16, 2026, ClickHouse acquired Langfuse, an open source platform teams use to trace, evaluate and monitor LLM applications. Langfuse's own announcement said the two teams had already been working together closely since Langfuse adopted ClickHouse as its underlying data layer, and that its founders had term sheets in hand for a Series A round before deciding to join ClickHouse instead. Both companies said Langfuse stays open source and self-hostable, with no planned licensing changes and Langfuse Cloud continuing to run as before. [3]
Atlas interpretation: The deal reads less like ClickHouse entering a new market and more like it acquiring a product that was already one of its own reference customers, folding a tool that generates ClickHouse workloads into the company that runs the database underneath it. [3]
Sources
- Our Story | ClickHouse
ClickHouse · Aug 20, 2026
- ClickHouse
Wikipedia · Sep 9, 2026
- Langfuse joins ClickHouse
Langfuse · Jan 16, 2026