ClickHouse Passes $350M ARR as OpenAI's Usage Grows 10x: AI Agents Are Rewriting the Database Market

The loudest numbers in AI this year have belonged to chips and frontier models. But one of the clearest signals about how AI is actually being deployed arrived this week from a layer of the stack most users never see: the database. ClickHouse, the company behind the open-source columnar database of the same name, has surpassed $350 million in annual recurring revenue, up roughly 40% in just three months — and the growth is being driven overwhelmingly by AI agents, including a reported ~10x increase in usage by OpenAI over the past year.

The milestone was first reported by The Information on August 25, 2026, and quickly rippled through markets — in part because ClickHouse has become a quiet bellwether for real AI workload growth, and in part because of who owns a large chunk of it.

The numbers

ClickHouse’s revenue curve has steepened dramatically in 2026:

• $250 million ARR as of late May 2026 — already more than triple year-over-year, with over 4,000 customers, announced at the company’s Open House event
• $350+ million ARR as of late August 2026 — up about 40% in three months
• Growth is being fueled by what The Information described as demand from AI agents for database and observability infrastructure

For context, a 40% quarterly jump on a base of a quarter-billion dollars is the kind of acceleration that venture investors usually only see at the very inflection point of a platform shift. ClickHouse’s leadership has been explicit that AI is the cause: agents need somewhere to write their telemetry, traces, evaluation data, and high-volume operational logs — and they need it queryable in real time.

Why agents love a columnar database

The technical story is straightforward. Modern AI agents are prolific data producers. Every agent run emits traces, tool-call logs, token accounting records, evaluation scores, and state snapshots. Traditional transactional databases and batch-oriented warehouses choke on this profile: extremely high write throughput, wide tables, and queries that need answers in milliseconds — “how did agent X fail yesterday,” “which tool calls are burning budget,” “what did the model see at step 42.”

Columnar engines like ClickHouse were built for exactly this — high-concurrency scans over append-heavy event data. That’s why the database became a default in observability (it powers telemetry backends across the industry), and why agent frameworks increasingly treat it as the system of record for everything an agent does. ClickHouse itself argued in a March 2026 essay that “AI workloads demand high-concurrency, real-time query processing, and full-fidelity data at scale” — and the revenue line now validates the thesis.

The headline customer datapoint: OpenAI’s usage of ClickHouse has reportedly grown about 10x over the past year. When the most valuable AI company in the world multiplies its consumption of your infrastructure tenfold, that says something about where inference-time compute and agent deployments are heading. The Information’s report noted that OpenAI’s surging use is a meaningful component of ClickHouse’s accelerating ARR.

From Yandex spinoff to $15B AI infrastructure player

ClickHouse’s journey is one of the stranger origin stories in enterprise software. The database engine was originally developed inside Yandex — Russia’s search giant — to power real-time analytics at web scale, and was open-sourced in 2016 under the leadership of creator Alexey Milovidov. In 2021 it was spun out as ClickHouse Inc., co-founded by Milovidov (CTO) and Aaron Katz (CEO, a former Snowflake sales chief), with a $250 million Series B at a $2 billion valuation.

The capital history since then reads like a company riding a wave it didn’t fully predict:

• May 2025: $350 million Series C led by Khosla Ventures at a $6.35 billion valuation
• October 2025: Series C extension adding Citi Ventures, Insight Partners, Peak XV, and others
• January 2026: $400 million Series D led by Dragoneer at a $15 billion valuation — a 2.4x markup in eight months — announced alongside the acquisition of LLM observability platform Langfuse and the launch of a Postgres-compatible engine

At its May 2026 Open House, the company also launched ClickHouse Agents, natural-language database agents powered by Anthropic’s Claude — an AI-native interface layered on top of the very infrastructure AI is filling up. That deal flow matters for interpretation: ClickHouse isn’t just selling to AI companies, it’s operationalizing AI inside its own product.

The Nebius angle

Part of why the ARR number traveled so fast beyond database circles: Nebius Group (NASDAQ: NBIS) holds an approximately 28% stake in ClickHouse, a legacy position from the Yandex spinoff era. At the $15 billion Series D valuation, that stake is worth roughly $4.2 billion — and every revenue milestone ClickHouse clears makes the asset (and a potential future IPO, which TechCrunch reported the company was charting toward in May) more concrete. Market-watchers on X flagged the connection within hours of the report.

What it means

Three takeaways worth holding onto:

1. The AI buildout is broader than GPUs. Training clusters and inference chips dominate headlines, but agents are quietly rebuilding the entire data stack beneath them. ClickHouse’s 40%-a-quarter growth is a lagging indicator of agent deployment volume, not model releases.
2. Observability is becoming a first-order AI cost. Langfuse’s acquisition into ClickHouse and the “database + observability” framing in The Information’s report suggest the next battleground is making agents debuggable and accountable at scale — which is a data problem before it is a modeling problem.
3. Open-weight infrastructure companies can outgrow the labs. A database company spun out of Yandex five years ago is now growing at rates that many foundation-model companies would envy, with gross margins software investors dream about.

The $350 million figure also resets expectations for ClickHouse’s next act. With a $15 billion valuation to grow into, an IPO path already being discussed, and the single hottest workload class in computing writing terabytes into its tables daily, the company has converted a decade-old open-source engine into the ledger of the agent economy — one 10x customer at a time.