Databricks

IWhat it does

Databricks is a private company. Its shares do not trade on any exchange, and as of 2 November 2026 I found no public IPO filing for it⁶. The company sells software that lets large organisations store, clean, analyse and build artificial intelligence (AI) systems on their own data, all in one place. Customers are mostly big companies and public bodies. The company says more than 20,000 organisations use it¹.

The core idea is called the lakehouse. Older systems kept raw files in a cheap pile (a data lake) and tidy tables in an expensive, separate system (a data warehouse). A lakehouse tries to do both on one copy of the data. The company grew out of a UC Berkeley research project and was founded in 2013 by seven people, including CEO Ali Ghodsi¹¹⁴. It is built on Apache Spark, an open-source engine that came out of the same lab¹⁴. The company is headquartered in San Francisco¹⁶.

Databricks has widened the product well beyond the first idea. The main pieces today are data warehousing (the company calls it Lakehouse and says it passed a $1.5B run-rate), Lakebase (a serverless Postgres database for applications and AI agents, with a run-rate above $100M), Genie (an assistant business users can ask questions of their data), Agent Bricks (a toolkit for building AI agents) and Unity Catalog (a layer that controls who can see which data)¹². Lakebase is built on technology from Neon, a serverless Postgres company¹⁹. Databricks announced in May 2025 that it would buy Neon¹⁸, and CNBC reported the price as about $1B, which is press-reported²⁰.

Why would a company pick this product? Most large organisations hold data in many places: operational databases, files in cloud storage, streams of events, spreadsheets. To train or run an AI system on that data, someone has to collect it, clean it and control access. Databricks sells that work as a managed service, so the buyer does not have to stitch together many tools. The company says more than 1,000 customers each consume at over $1M a year, and more than 100 at over $10M, so the revenue is concentrated in large accounts¹.

A run-rate is the latest period’s revenue scaled up to a year. It is not audited revenue, and companies choose how to calculate it. Databricks has not published audited accounts, so every revenue figure on this page is a number the company chose to announce or a third party estimated.

IIHow it makes money

Databricks mostly charges for usage. A customer buys capacity in advance or pays as it goes, and the bill rises with the amount of data stored and the amount of computing done on it. Usage pricing helps when customers do more with the platform, which is why the company can grow quickly inside accounts it already has. It also means revenue can fall if customers cut workloads, and the company runs on top of the big cloud providers, which take their own share of the underlying cost.

The company does not publish gross margin, operating margin or net income. Sacra, a research firm, cites a gross margin near 80% as of mid-2024 and notes that heavy AI use may be pressing it¹³. That is a third-party figure, not a filed one. The only profit signal from the company is a statement that adjusted free cash flow was positive over the last twelve months, with no dollar amount¹. Adjusted means the company chose what to leave out, and stock pay is the usual item.

Net retention is the share of last year’s revenue that existing customers still pay this year, including expansion. The company said in December 2025 that it was above 140%². It did not repeat the figure in the August 2026 release¹. For comparison, Snowflake reported 126% in its latest quarter⁵.

The company breaks out a few product lines. At the December 2025 announcement, AI products and data warehousing had each passed a $1B run-rate². By August 2026, data warehousing was above $1.5B and growing over 100% year on year, and Lakebase was above $100M¹. Sacra estimates AI products were about $1.4B, or 26% of revenue, at the end of 2025¹³. Together these named lines are a minority of a $7B total, so most revenue still comes from the core data engineering and analytics work.

Company-announced figures

IIIThe market and the competition

Databricks sells into the market for data platforms and AI tooling. Its closest public rival is Snowflake, which started as a cloud data warehouse and is moving toward AI. In its quarter ended 31 July 2026, Snowflake reported product revenue of $1.49B, up 37%, and guided to $6.07B for its full fiscal year⁵. Databricks says its own run-rate is above $7B and growing over 80%¹. The two are not measured the same way, since one is a run-rate and the other is booked revenue, but the gap in growth is large.

The big cloud providers are both partners and rivals. Amazon, Microsoft and Google each sell their own data warehouses and AI services, and each also sells Databricks through its marketplace¹³. Microsoft is also a Databricks investor, and the two have worked on making Azure Databricks work with Microsoft Fabric, Microsoft’s own data suite¹³. That makes the relationship useful but dependent: a cloud provider could favour its own product in sales or pricing.

Other competitors come from the edges. Palantir sells AI software layered over a customer’s data, and it grew revenue 96% in its latest quarter to $2.31B³. Datadog, a monitoring company, is moving into security and data tools and reported 36% growth to $1.12B in the June 2026 quarter⁴. Sacra also lists Splunk, Microsoft Sentinel and CrowdStrike as targets for Databricks’ newer security product¹³. I cannot judge how much of the market each of these could take. The sources I opened do not give market share.

Snowflake’s figures give a useful yardstick. It had 828 customers with over $1M of trailing product revenue, a net retention rate of 126% and remaining performance obligations (contracted future revenue) of $9.0B⁵. Databricks has not disclosed a comparable contracted-revenue figure, so I cannot compare the quality of the two revenue bases. Palantir, in contrast, reports non-GAAP profitability on a scale few software firms reach: a 62% adjusted operating margin in Q3 2026 and guidance for $4.8B to $5.0B of adjusted free cash flow this year³. That tells me public investors are paying very high prices only where growth and profit arrive together.

What drives demand is simple to state. Companies want AI that works on their own private data, and that data has to be gathered, cleaned and governed first. If AI spending slows, or if companies decide the cloud providers’ bundled tools are good enough, usage-priced revenue is where it would show first.

IVThe moat and how it could be disrupted

My rating is contestable. The advantages are real, but a well-funded rival or a standards change could erode them within a few years.

What protects it. First, switching costs. Once a company has put years of data pipelines, tables, permissions and dashboards on a platform, moving them is slow, risky and dull work. Second, scale and breadth: Databricks sells data engineering, warehousing, databases, governance and AI on one bill, and more than 1,000 customers each spend over $1M a year at run-rate¹. Third, momentum in talent and trust: it has open-source roots in Spark, Delta Lake and MLflow, and developers know the tools¹⁶.

What weakens it. Open formats cut both ways. Databricks pushed open table formats and open sharing so that customers would not feel locked in, and Snowflake has since added support for the same sharing standard¹³. If data can be read by any engine, the engine becomes easier to swap. Usage-priced software is also easy to shrink: a customer can move a single workload without leaving the platform. And the cloud providers control the underlying machines and the sales relationship.

How it could be disrupted. A cloud provider could bundle a good-enough lakehouse at a low price. AI agents could write and move code between engines cheaply, which lowers the cost of switching. A large AI lab could sell its own data tools directly to enterprises. Or growth in AI workloads could slow before margins recover. Databricks has also moved into a crowded field with Lakebase, where it competes with long-established database vendors, and the company’s own benchmark that Genie answers questions better than other agents is unverified¹³.

I treat the moat as strong in the existing customer base and weak in new workloads. That split is why the rating is not higher.

VThe numbers we can see

All revenue figures below are company-announced run-rates unless marked. There are no audited accounts available to the public, so I cannot check cost of revenue, stock pay, cash or debt.

Growth has sped up. The run-rate was above $4.8B in Q3 2025 with growth over 55%², $5.4B with growth of 65% in Q4⁷, and above $7B with growth over 80% in Q2 2026¹. Growing faster at larger scale is unusual. The company did not say how much of the jump came from acquisitions, price changes or a few very large customers.

On funding, the company announced a Series L at a $134B valuation in December 2025, described as raising more than $4B². In February 2026 coverage said it had raised $7B in total, made up of $5B of equity and $2B of debt, with the valuation unchanged at $134B⁷. Other outlets give different debt figures: one reports $1.8B of debt from JPMorgan and private lenders in January¹⁰, and another describes the $2B as debt capacity rather than drawn debt, so the amounts actually borrowed are unclear.

In July 2026 the company said it had signed a term sheet for a round at $188B led by Coatue¹¹. Reuters, via DealStreetAsia, reported it on 17 July without an amount¹¹. The Wall Street Journal, as relayed by Trending Topics, put it at about $3B, which is press-reported¹². The round closed on 13 August 2026 at $5B and a $190B valuation¹. The press release says valuation but not whether it is before or after the new money, and CNBC does not say either¹²¹. I treat it as post-money, which is the usual convention, but I could not confirm that from the primary source or from the news coverage I opened. TechTarget puts total financing above $32B, including debt⁹.

A word on reliability. The August 2026 release and the December 2025 release are the company’s own words, and companies pick the metrics they publish. Net retention, for example, appeared in December 2025 and not in August 2026¹². I would not read much into that omission, but a careful reader should notice it. The run-rate also leaps from $5.4B (a figure reported for Q4 2025) to above $7B in about two quarters⁷¹. Sacra’s estimate, $5.4B in January and $6.9B in June 2026, is consistent with that but is not independent, since it likely draws on the same announcements¹³.

One checkable cross-check on ’positive free cash flow’: Snowflake, with a 37% growth rate, reported an adjusted free cash flow margin of 6.0% in its latest quarter and guides to 23% for the year⁵. Databricks has given no margin, so I cannot say whether it is close to Snowflake, well above it or barely positive.

Valuation history (company-announced unless marked)

VIWho runs it and who owns it

Ali Ghodsi is CEO and one of seven co-founders¹. The others named are Andy Konwinski, Arsalan Tavakoli-Shiraji, Ion Stoica, Matei Zaharia, Patrick Wendell and Reynold Xin¹⁴. The company’s roots are academic: Databricks says the seven were researchers at the Berkeley lab where Spark began¹⁴. I cannot judge the board or how much day-to-day control the founders keep, because I could not open a source on either.

Ownership is spread across founders, employees and a long list of investors. The August 2026 release names Coatue (lead), Blackstone, MGX, accounts advised by T. Rowe Price and Sixth Street Growth, plus BOND, Clearlake, Point72, Premji Invest and TPG as new investors, and Andreessen Horowitz, Fidelity, GIC, Insight, NEA, Temasek, Thrive and others as existing ones¹. Ownership percentages are not published. Many of the recent buyers are large mutual-fund and crossover investors, which usually means they expect a public listing or a later sale.

What an owner can take out is limited. Private shares cannot be sold freely. Employees get liquidity through company-run sales, and the company has said part of its raising is for employee liquidity². The CEO has said an IPO would create what he called a market transaction mechanism for staff¹⁰. Late-stage preferred shares may carry rights that common shares do not, such as priority in a sale. I could not see the share terms, so I cannot say how much of the $190B sits above the common stock.

VIIValuation: is the last price supported?

Equity value in US$ billions. Not a share price: a private company has none. Latest mark: primary round, $5B strategic round, described as $190B valuation (pre- or post-money not stated in the release; I treat it as post-money).

What the last valuation assumes

At $190B and a run-rate of $7B, the price is about 27 times the current run-rate (my calculation from¹). To justify that, I need to believe run-rate grows to something like $35B or more by 2030 and that a mature Databricks earns margins like the best software companies. That sits at or beyond my bull case. The price also assumes that a public market would pay a premium for growth near 80%, which today only Palantir among the comps I looked at is matching³.

Bear, base and bull

• Bear. Run-rate of $7B in Q2 2026 reaches about $8.0B at end-2026 and $16B at end-2030 as growth fades to 12%. Terminal EV/revenue of 5x, below where mature software with slowing growth trades. Discount rate 13%, equity-style risk for an illiquid private share. Net cash $3B, dilution 12% from stock pay and further raises. Result: about $45B.
• Base. Run-rate reaches about $8.2B at end-2026, then grows 45%, 35%, 28% and 22% a year to about $25B at end-2030. Terminal EV/revenue of 9x on that run-rate, roughly half Snowflake’s current multiple of about 19x on guided revenue, because growth should be lower by then⁵⁹. Discount rate 11%, a high-single-digit cost of equity plus a premium for illiquidity. Net cash $4B, dilution 8% over four years. Result: about $139B.
• Bull. Run-rate reaches about $8.4B at end-2026 and $34B at end-2030 with growth of 60%, 45%, 35% and 28%. Terminal EV/revenue of 12x, at a discount to Snowflake today. Discount rate 10%. Net cash $5B, dilution 6%. Result: about $262B.

Comparables used

Where I come down

The latest valuation of $190B is above my base case of about $139B, by roughly a third. It sits between my base and my bull case of $262B, and far above my bear of $45B. In plain terms the price looks high against what public data can support. To be right, it needs growth to stay near or above 40% for several years and margins to rise. The market might still be right. Growth is accelerating, not fading, which is rare at this size. Sophisticated investors with access to non-public detail priced the round, and it is based on a company-reported run-rate that I cannot audit. Also, my terminal multiples are my own judgement: if public software with this kind of growth keeps trading near Snowflake’s 19x, my base would be much higher. My margin of safety price is about $104B, 25% below my base, and the round is nearly double that. This is research, not investment advice.

What I cannot judge

I cannot judge true gross margin, operating margin, cash, debt, share terms, the cap table, or whether run-rate matches billed revenue, because Databricks has not published audited accounts. I also cannot judge how long AI-driven usage keeps compounding, or how the cloud providers behave. The valuation range is wide on purpose: $45B to $262B. Public data is too thin to narrow it. Because the figures come from company statements, they may be selective, and I have no way to test that. A confidential IPO filing would not be visible to me.

VIIIThe IPO

No IPO plan reported. Databricks is not public and I found no public S-1 or F-1 on EDGAR in a company search on 2 November 2026⁶. A confidential submission would not appear there, and I found no report of one. CEO Ali Ghodsi told Bloomberg Television on 4 June 2026 that the company would go public eventually, and also called 2026 a terrible year to go public¹⁰. Coverage points to 2027 or later as a possible window, which is commentary, not a company statement⁹¹². He has given employee liquidity as the main reason to list, not a need for cash¹². No exchange, ticker, banks or size have been reported in the sources I read. Some aggregator pages claim an S-1 was filed in 2026; I found no primary evidence for that and treat those claims as unreliable.

What to watch

• A public S-1 on SEC EDGAR, or news of a confidential filing
• Whether the company publishes audited revenue, margin and cash numbers
• Growth in the next run-rate update compared with the above 80% reported in August¹
• Snowflake’s and Datadog’s next results, since they set the public multiple for this group⁵⁴
• Condition of the IPO market after the large 2026 listings the CEO mentioned¹⁰
• Any further private round, tender offer or debt raise, and the valuation attached

IXWhat could go wrong

The most likely way this business loses value is a growth stall. A company valued on 80% growth is repriced fast if growth falls to 30%. Usage revenue is more exposed to this than subscription revenue, because customers can cut spending without cancelling.

Several things have to go right together. As I see it: growth stays above about 40% for two more years, margins reach a Snowflake-like or better level, the cloud providers keep Databricks as a partner, AI workloads keep landing on the platform and not in model labs’ own tools, and the public market is open and generous when the company lists. If I give each of those a rough 70% chance and treat them as independent, the combined chance is about 17%. They are not independent, since a strong AI market helps most of them at once, so the real number is higher, but not by a lot. That is a rough estimate of mine, not a sourced figure.

Disclosure risk is the other issue. We have no audited accounts, and the headline numbers are run-rates. Run-rate can include contracted but not yet billed revenue, and the company has not said how it counts. Private investors also see more detail than the public does. The cash on hand is unknown. Debt of up to $2B was reported⁷, which is small against the equity value, but the terms are not public.

Concentration and talent matter as well. The company’s revenue depends on a few hundred very large customers, and its engineers are in demand from AI labs and large technology companies. Pay in stock is part of the cost of keeping them, and an illiquid stock that stops rising makes that harder. The company has said part of its funding goes to employee liquidity, which is a sign it takes the issue seriously².

Market risk matters too. The CEO said 2026 was a bad year to list because of other large offerings¹⁰. If the IPO window shuts for longer, early investors and staff cannot sell and the company has to keep raising private money, which means more dilution.

The next best use of the same dollar is a listed rival. Snowflake trades at about 19 times its guided fiscal-year product revenue (my calculation: about $114B market cap per TechTarget⁹ over $6.07B⁵) and Datadog and Palantir offer public, audited alternatives, though each has its own risks.