ClickHouse
$15 billion on $250 million of ARR — the fastest-compounding database in the market, priced at 60x revenue, with nothing on file at the SEC.
ClickHouse is the best growth story in data infrastructure and the least mature IPO candidate on this list. Annualized revenue tripled to $250M by May 2026, ClickHouse Cloud ARR grew more than 250% year over year, and the customer list — Meta, Anthropic, Tesla, Sony, Cursor, Capital One — reads like the AI buildout's own invoice book. Dragoneer led a $400M Series D in January 2026 at a reported $15B. That is where the discipline has to start: $15B on $250M of ARR is 60x revenue, roughly 2.4x what the market pays Snowflake, and the company has filed nothing with the SEC — no S-1, no confidential draft, not even a Form D. There is an IPO thesis here. There is not yet an IPO.
The Numbers at a Glance
What ClickHouse Actually Is
A column-oriented database built for one thing: answering analytical questions over enormous tables in under a second. It was written inside Yandex roughly seventeen years ago by Alexey Milovidov to power web analytics at a scale the available tools could not touch, open-sourced in 2016, and spun out as an independent company in 2021. Milovidov is still CTO.
The commercial model is open core done properly. The database is free, widely deployed and genuinely loved by engineers — which is the distribution channel. The money is in ClickHouse Cloud, the managed service, plus the surrounding machinery: ClickPipes for managed ingestion and ClickStack for observability. That is the classic Elastic and MongoDB shape, and it carries the classic risk: the free tier is also the competition.
What makes the current moment different is why the workloads are arriving. AI applications generate exhaust — traces, evaluations, token-level logs, agent steps — at volumes and write rates that behave far more like observability than like a warehouse. ClickHouse is architecturally good at exactly that, which is why Anthropic, Cursor, Lovable and Decagon show up beside Meta and Tesla in the customer list. The AI boom is not a story ClickHouse tells; it is a workload that happens to suit the engine.
Six acquisitions, and the newest one is an LLM observability company
ClickHouse has bought six startups, most recently Langfuse — an open-source LLM observability project with over 20,000 GitHub stars at the end of 2025 and 26M+ SDK installs per month. Alongside the Series D it also announced a native Postgres service with Ubicloud, with change-data-capture into ClickHouse and a claim of up to 100x faster analytics on synced data. Read together, these are not adjacent features. They are an attempt to own the AI application's entire data path — transactional store, replication, analytical engine and the observability layer on top — before Datadog or Snowflake get there.
Where the Moat Is — and Where It Leaks
The durable advantage is performance at a price point, plus the developer gravity that comes from a decade of open-source adoption. When a team benchmarks ClickHouse against a general-purpose warehouse on a high-cardinality, high-write-rate workload, the result is usually not close — and the engineers who ran that benchmark become the buyers when the workload outgrows self-hosting.
That conversion path is the business. Free open source builds the installed base; operational pain converts a slice of it to Cloud. It is cheap, credible distribution that Snowflake has to buy with a sales force.
The leak is in the same pipe. Everything ClickHouse sells on top of, anyone can self-host for free. Every large customer that gets good at running it themselves is revenue that never arrives, and the most sophisticated users — exactly the Metas and Teslas on the logo wall — are the most capable of doing so. Meanwhile DuckDB has taken the small-to-mid analytical workload in-process, StarRocks and Apache Druid contest the real-time tier, and Snowflake and Databricks are both closing the latency gap from above with their own fast-query layers.
Open core is the cheapest customer acquisition in software and the hardest revenue to defend. The same property does both.
The Financials — and the Size of What Is Missing
This is a private company with no filing obligation, so what follows is what management has chosen to say, dated and attributed. It is not an income statement.
| Metric | Figure | As of / source |
|---|---|---|
| Annualized revenue run rate | $250M | May 2026, Yury Izrailevsky via TechCrunch |
| ARR growth | ~3x YoY | Same; company separately cites Cloud ARR +250%+ |
| ClickHouse Cloud customers | 3,000+ | Company, Series D announcement |
| Total customers | 4,000+ | TechCrunch, May 2026 |
| Series D | $400M | January 2026, led by Dragoneer |
| Reported valuation | ~$15B | Press; not stated in the company's own announcement |
| Total raised | $1B+ | Press, post-Series D |
| GAAP revenue | — | Not disclosed |
| Gross margin | — | Not disclosed |
| Profitability / burn | — | Not disclosed |
| Net revenue retention | — | Not disclosed |
Three cautions on the top line. ARR is not revenue — it is a run-rate snapshot that flatters any business growing this fast, and for consumption-priced infrastructure it is unusually sensitive to a handful of large accounts. Gross margin is unknown, and for a managed service running someone else's compute it is the number that decides whether 60x is a growth multiple or a fantasy; Snowflake's product gross margin sits in the mid-70s, and a cloud data service materially below that is a different company. Nothing about burn is public — $1B raised against $250M of ARR is consistent with either disciplined scaling or heavy subsidy, and from outside there is no way to tell.
Management expects revenue in "the high-nine digits by the end of the year"
That is Izrailevsky in May 2026, via TechCrunch — a figure approaching $1 billion, from $250M seven months earlier. Taken literally it implies roughly another tripling inside a single fiscal year at a scale where almost no software company has ever done it. Treat it as management ambition on an unaudited metric, not guidance, and note what it would mean if it were even half right: at $500M of ARR, the $15B mark is 30x, not 60x, and this brief's central valuation objection substantially weakens. The next disclosed ARR print is the most important number in this story.
IPO Status — Preparation Without a Filing
We checked EDGAR directly rather than trusting coverage. A company search for ClickHouse returns no registrant. There is no Form S-1, no confidential draft registration statement, and no Form D — meaning even the private rounds were placed without a Reg D notice filed under that name. On the public record, ClickHouse has not begun.
| Signal | Status | What it tells you |
|---|---|---|
| S-1 or DRS on file | None | No SEC filing of any kind exists. A listing is at minimum several quarters out. |
| CFO hire | Done | Jimmy Sexton, previously of Snowflake, is CFO — the standard first structural move toward a listing. |
| Management statement | Soft | Izrailevsky, May 2026: positioned for an IPO "within the next few years." No timetable. |
| Crossover investors on the cap table | Yes | T. Rowe Price accounts, GIC, WCM and Dragoneer in the Series D — the investor set that typically precedes a listing by 12–24 months. |
| Audited financials | Unknown | Not public. No margin, retention or profitability disclosure exists to underwrite against. |
The Series D cap table is the most informative item there. Dragoneer led, with Bessemer Venture Partners, GIC, Index Ventures, Khosla Ventures, Lightspeed, accounts advised by T. Rowe Price and WCM Investment Management. Sovereign wealth and mutual-fund money does not enter at $15B for a five-year hold; it enters expecting a public mark. That is the strongest IPO signal available — and it is an inference about investor intent, not a company commitment.
Competitive Landscape
| Competitor | Type | Scale | QL Signal | Where it collides |
|---|---|---|---|---|
| Snowflake | Public · SNOW | $116.1B cap | Underweight | The warehouse incumbent, pushing down into lower-latency query |
| Datadog | Public · DDOG | $79.4B cap | Neutral | Owns observability — precisely where ClickStack and Langfuse point |
| MongoDB | Public · MDB | $29.2B cap | Underweight | The open-core-to-cloud comparable; the template and the warning |
| Elastic | Public · ESTC | $8.8B cap | Underweight | Same model, same search/analytics adjacency, repriced to 5x revenue |
| Oracle | Public · ORCL | $432.9B cap | Underweight | Enterprise incumbency and the AI-infrastructure land grab |
| Databricks | Private | Mega-cap private | n/a | The other side of the lakehouse fight; far broader platform |
| DuckDB | Open source | In-process | n/a | Eats the small-and-mid analytical workload before it ever becomes revenue |
| StarRocks / Druid | Open source | Real-time OLAP | n/a | Direct architectural substitutes in the same tier |
Note the tape ClickHouse would be listing into: Snowflake, MongoDB, Elastic and Oracle all carry Underweight QL signals, and Datadog is Neutral. Every listed comparable in this category is currently out of favour. A company priced at 60x private would be arriving in a market that has spent two years compressing the multiples of the companies it most resembles.
Valuation Framework
| Company | Value | Revenue | Multiple | Growth |
|---|---|---|---|---|
| ClickHouse — at reported $15B | $15.0B | $250M ARR | 60.0x | ~200% |
| Snowflake | $116.1B | $4.68B | 24.8x | ~26% |
| Datadog | $79.4B | $3.43B | 23.1x | ~25% |
| MongoDB | $29.2B | $2.46B | 11.9x | ~19% |
| Elastic | $8.8B | $1.74B | 5.1x | ~15% |
Growth-adjusted, the premium is arguable rather than absurd. ClickHouse at 60x on roughly 200% growth is 0.30x per point of growth; Snowflake at 24.8x on ~26% is 0.95x; Elastic at 5.1x on ~15% is 0.34x. On that crude measure ClickHouse is priced below Snowflake and in line with Elastic — which is the bull case, and it is a real one.
The bear case is that the denominator is the fragile part. Growth of 200% does not survive contact with $1B of revenue; the question is whether it lands at 60%, 40% or 25%, and how fast. At $500M of ARR the multiple halves to 30x on its own. At $250M with growth decelerating to Snowflake's rate, the fair multiple is Snowflake's — which implies roughly $6B, a 60% markdown from $15B. The entire bull case rests on the next two ARR prints, and there is no audited statement to check them against.
Elastic is the same business model, ten years on, at 5.1x revenue
Elastic ran the identical playbook — beloved open-source engine, managed cloud on top, search-and-analytics adjacency — and the market now pays it 5.1x revenue with an Underweight signal. MongoDB, the best-executed open-core-to-cloud story in the public market, gets 11.9x. Neither is a failed company; both are the mature form of exactly what ClickHouse is becoming. If ClickHouse is worth 60x and MongoDB is worth 11.9x, the difference has to be growth that persists for years, not quarters.
Risk Matrix
| Risk | Severity | Detail |
|---|---|---|
| Valuation at 60x ARR | High | 2.4x Snowflake's multiple on an unaudited run-rate metric. Needs multi-year hypergrowth to be retrospectively cheap; a single decelerating print reprices it hard. |
| No SEC filing, no audited financials | High | No S-1, no DRS, no Form D. No margin, retention, burn or GAAP revenue disclosure exists. Every number here is management-sourced. |
| Open-source substitution | High | The product that drives adoption is free. Sophisticated large users can and do self-host, and DuckDB has taken the entry tier in-process. |
| AI-workload concentration | Medium | The customer list leans heavily on AI-native companies whose own spending is venture-funded. A funding-cycle turn hits ClickHouse's fastest-growing cohort first. |
| Consumption pricing | Medium | Revenue falls when customers optimise. Snowflake's own multiple compression began with exactly this. |
| Big-platform encroachment | Medium | Snowflake and Databricks are both building down into low-latency query; Datadog owns the observability budget ClickStack targets. |
| Acquisition integration | Medium | Six acquisitions in under five years, with Langfuse pushing into a new category. Integration risk against an incumbent that does it natively. |
| Key-person and origin | Low | The engine's author is CTO — a strength with concentration attached. The Yandex lineage is long since severed but appears in diligence. |
How to Get Exposure
With nothing on file, there is no near-term listing to underwrite. The realistic routes are private, indirect, or patient.
| Route | Availability | Notes |
|---|---|---|
| Secondary marketplaces | Limited | Priced against the January 2026 Series D. Expect a premium to the $15B primary mark and minimum sizes that exclude most buyers. |
| Crossover funds | Indirect | T. Rowe Price accounts, GIC and WCM hold positions. Exposure through a fund is diluted to near-invisibility. |
| Public proxies | Imperfect | SNOW, DDOG, MDB and ESTC trade the same theme with published financials — and all but DDOG currently carry Underweight signals. |
| The IPO itself | Not yet | No filing exists. The first observable milestone would be a DRS appearing on EDGAR, which is confidential until the company chooses to go public with it. |
Bottom Line
ClickHouse is compounding faster than any listed company in its category and is priced accordingly. $250M of ARR tripling year over year, 3,000+ cloud customers, Meta and Anthropic and Tesla on the logo wall, and a Series D cap table stacked with the crossover investors who show up before listings — the quality of the business is not the question. The question is the number. 60x an unaudited run-rate metric, against Snowflake at 24.8x and MongoDB at 11.9x, in a tape that currently marks every public comparable Underweight, is a price that requires hypergrowth to persist for years rather than quarters. And there is no S-1, no draft, not even a Form D — so there is no audited margin, no retention figure and no burn number to test the story against. This is a watchlist name of the highest quality and a pre-IPO position only for investors who can hold through a repricing. The two things that would change the analysis are a disclosed ARR print near management's own "high-nine digits" ambition, which would halve the multiple on its own, and a DRS appearing on EDGAR. Neither has happened yet.