Databricks vs Snowflake (3/7): The Money Race — the Biggest Software IPO Ever vs the Private Mega-Round Machine

TL;DR — Snowflake raised ~$3.4B in the biggest software IPO on record on Sept 16, 2020, popped 112% on day one, and has since compounded to $4.47B in FY2026 product revenue — all of it audited, and none of it yet GAAP-profitable. Databricks never listed: it stacked mega-rounds from $13.9M in 2013 to a reported $188B valuation in July 2026, and says it is now at a $6.9B annualized run rate. One banked the cash under public scrutiny. The other owns the momentum — on numbers nobody has audited.

After two rounds that leave Databricks a point up, the ledger turns to the one dimension that can't be argued in a blog post: money. Who raised it, who banked it, who has to open the books every ninety days.

Here the two companies stop rhyming. Snowflake's capital story is a closed arc — eight years private, one enormous liquidity event, then six years of quarterly accountability. Databricks' is open: thirteen years private, roughly $20.2B raised across ~15 rounds (per Tracxn, pre-July-2026), and a CEO who said in June 2026 that "2026 is a terrible year to go public."

The concepts you need

Six ideas do the work in this story. Without them the tables below are just big numbers.

Consumption pricing vs seat licences. A seat licence bills per user per year: fixed, forecastable, renewed on a date. Both companies bill for what you run instead — Snowflake meters per-second compute credits plus storage and egress, no flat fee; Databricks meters DBUs (a normalized compute unit, also per-second), with cloud infrastructure billed separately on top. Nobody commits to an amount of querying, so a paused pipeline shrinks vendor revenue without anyone cancelling. That elasticity is why both firms' growth tracks usage discipline rather than renewal dates.

Net revenue retention (NRR). Take last year's customers, ignore every new logo since, and ask what that same cohort spends now. 125% means the base grew a quarter on its own; 178% means it nearly doubled. Above 100% you grow even if sales closes nothing. Snowflake's slide from 178% to 125% is therefore not churn — its $1M+ accounts went 184 to 733 — it is a larger, maturer base in which each new dollar moves the ratio less.

ARR / run-rate vs audited GAAP revenue. $4,472.3M is money Snowflake earned across a defined twelve months, recognized under accounting rules and signed by an auditor. "$6.9B annualized" is a recent period multiplied up: forward-leaning by construction, self-defined, unaudited, published when it flatters. Not fake — a different instrument. Don't subtract one from the other.

What an S-1 discloses. The IPO registration statement forces a private story into standard shape: audited multi-year financials, customer concentration, share counts, executive pay, risk factors written by lawyers paid to make them ugly. Snowflake filed one in 2020 — that is why we can quote FY2020 revenue of $264.7M. Databricks has filed none, public or confidential, as of mid-2026.

Market cap vs private valuation. Market cap is shares outstanding times the last price a stranger actually paid, repriced continuously by people free to sell. A private valuation is what a small group agreed to pay for a new slice, often with preferences attached to that slice alone; multiplying it across every share gives a headline, not a market. Even the public number wobbles — Snowflake reads $114.55B on Aug 7, 2026 and $101.65B–$110.22B on Aug 8.

Free cash flow vs profit. FCF asks whether more cash entered the building than left. GAAP profit also charges non-cash costs, chiefly stock-based compensation. Snowflake generated $1,120.3M of FCF in FY2026 and still booked a $1,329.0M GAAP loss. Databricks says it has been FCF-positive on a trailing-twelve-month basis since around Q4 2025 — a real claim, and not the same word as "profitable."

Snowflake: eight years private, one record day

Sutter Hill seeded Snowflake in 2012 — reported as a $5M Series A, a check worth roughly $12B by the IPO. A $26M Series B followed in October 2014. The last private round was the one that mattered: $479M at a $12.4B valuation in February 2020, led by Salesforce Ventures and Dragoneer, tripling the prior mark seven months before listing.

Then Sept 15–16, 2020. The IPO priced at $120/share — guided at $75–85, revised to $100–110, and priced above even that. 28M shares, ~$3.4B raised, ~$33.3B market cap at the offer. Berkshire Hathaway and Salesforce each took $250M in a concurrent private placement; Berkshire added ~4.04M shares in a secondary, putting its pre-trading stake near $730M.

The stock opened at $245 — 104% above the offer — hit roughly $319 intraday, and closed near $253.93, up ~112%. Per a CNBC/Jay Ritter analysis, the pop left about $3.8B "on the table," the most in twelve years.

Databricks: the mega-round machine

Databricks took a16z's $13.9M Series A in September 2013 and never stopped — a two-year pause after 2021, then re-acceleration into the AI boom at a cadence no public company could match.

Two details matter. The Series J close on Jan 22, 2025 brought Meta in as a first-time strategic investor alongside Temasek, wrapped around a JPMorgan-led $5.25B credit facility ($2.5B revolver + $2.75B term loan). And the Series L grew between announcement and close — equity from >$4B to >$5B after JPMorgan Asset Management raised its contribution, with Microsoft, Morgan Stanley, QIA and Goldman Sachs Alternatives joining.

The July 2026 $188B mark — led by existing investor Coatue, a ~40% jump in five months — needs a caveat: it is a signed term sheet "expected to close later this summer," with no close announcement found. Treat it as pending.

The timeline, side by side

Note the reversal: in late 2021 Snowflake's market cap was roughly three times Databricks' private mark; by August 2026 Databricks' reported valuation is 1.6–1.9x Snowflake's.

Both market-cap figures deserve an asterisk. Snowflake's 2021 peak is reported at $122.9B by one point-in-time source and $103.76B by a year-end series — different dates, different share counts. Its all-time-high close was $401.89 on Nov 16, 2021. Today's readings spread too: $114.55B (Aug 7) against $101.65B–$110.22B (Aug 8). Use ranges, not decimals.

Revenue: audited vs announced

The most important sentence here: Snowflake's revenue is SEC-filed and audited; every Databricks figure is a company disclosure of ARR or run-rate. Not the same kind of number. Snowflake's fiscal year ends Jan 31, so FY2026 = the year ended Jan 31, 2026.

Read the growth columns, not the levels. Snowflake decelerated from triple digits to 29% — a normal, gravity-obeying curve. Databricks re-accelerated: ~50% in mid-2025, 65% in February 2026, >80% by the June 2026 Data + AI Summit, with AI products alone at a $1.4B run rate (~26% of total in February) and data warehousing crossing $1.5B ARR. Follow-on reporting notes margins compressing as that growth accelerated — a trade you can make when no public shareholders are grading you.

Quality of revenue, and who actually makes money

One correction to the folklore: the widely repeated "169% NRR at IPO" is shakier than it looks. Sources disagree — 169% for a half-year in the S-1, 158% trailing at listing. "High-150s to 160s%" is honest; the trend (178% → 125%) is what's verified.

Wall Street's reaction to that trend has a date: Feb 29, 2024. Snowflake guided FY2025 growth to ~22% against a Street expecting ~30%, and Slootman retired the same day, with Sridhar Ramaswamy taking over. The stock fell ~20% — roughly $17B of market cap in one session. That is the price of the scrutiny Databricks has declined to buy.

Two business-model stacks, weighed

Neither model is the correct one. They are different machines with different failure modes, and this is the dimension where the difference bites hardest.

Snowflake — the public-company stack

Advantages. Every figure above is auditor-signed and SEC-filed, so a buyer doing vendor diligence can read the losses, the $9.77B RPO and the retention curve without asking permission. Employees and early backers got liquidity on a known date. And the ninety-day cycle is a discipline: non-GAAP operating margin went from (3)% in FY2022 to 10% in FY2026 while growth halved.

Disadvantages. That grading punishes candour about deceleration — a ~22% guide against a ~30% expectation, plus a same-day CEO change, cost ~$17B in one session. Long-horizon bets get priced as margin risk before they get priced as strategy. And the margin optics lean on stock comp: GAAP losses widened from $679.9M to $1,329.0M while non-GAAP margin improved. The shareholder pays that, in dilution.

Databricks — the private mega-round stack

Advantages. No quarterly jury. It can let margins compress to buy growth — reportedly what happened as it accelerated past 80% YoY in June 2026 — and nobody downgrades it on Monday. Capital arrives in shapes a public company uses less easily: $10B of Series J equity beside a $5.25B JPMorgan-led credit facility in one financing. And the valuation resets on its own schedule, $62B to a reported $188B in nineteen months.

Disadvantages. Nothing is audited. NRR ">140%", $6.9B annualized, FCF-positive TTM — all company-stated, none falsifiable from outside, and the $6.9B came in conference remarks rather than a filing. Valuations are marks, not markets: the $188B is a signed term sheet with no confirmed close, and the $165–175B talks reported a month earlier show how far a mark travels before a dollar is banked. Thirteen years private also means employees hold paper whose exit their own CEO just pushed to 2027 at the earliest.

What to Learn From This

1. Match the instrument before you compare. A deck that sets "annualized run rate" beside a rival's audited fiscal-year revenue is committing a category error, not keeping score. Ask three things of every vendor number: what period, whose definition, who audited it.
2. Read growth and retention, not headline size. $4.47B growing 29% and a $6.9B run rate growing >80% are different points on the same curve. The level says who is bigger today; the rate says who is bigger in three years.
3. Under consumption pricing, your tuning is their revenue. There is no renewal cliff to negotiate from, so leverage lives in committed-use contracts (Snowflake capacity commitments, Databricks DBCUs) and in workloads you can genuinely move. Budget the meter, not the licence.
4. Price the whole bill. Databricks bills DBUs; the cloud compute, storage and networking underneath arrive on a separate invoice, reportedly adding 50–100%. Model both or your TCO comparison is fiction before it starts.
5. Scale vendor-risk work to what you can read. From a public vendor, 10-K risk factors and RPO are free intelligence. From a private one you get only what it chooses to publish — so compensate in the contract: exit terms, data portability, price protection, open formats.
6. Deceleration is normal; a sudden guidance cut is the signal. 174% to 29% over six years is gravity. A one-session repricing tells you more about near-term roadmap and pricing pressure than the multi-year curve does.

Scorecard

Round 3: Snowflake. It banked ~$3.4B in cash and set the software IPO record — a number settled forever, not a mark that resets each round. It has published six years of audited results including the ugly ones: a GAAP loss widening to $1,329.0M, NRR falling from 178% to 125%, and a single session that erased ~$17B. Databricks looks better on every growth line, but those lines are company-claimed, unaudited, and untested by a market that can vote against them daily. A $188B term sheet is a promise; $3.4B wired in September 2020 is a fact. The caveat is real: momentum belongs entirely to Databricks — >80% growth at a $6.9B run rate, FCF-positive by its own account — and if it lists in 2027 anywhere near its private mark, this verdict flips.

Running tally: Databricks 1 — Snowflake 1 (1 draw)

Tomorrow

Part 4: the nastiest public fight in data — the 2021 benchmark war, when TPC-DS results became a shouting match between two vendors and neither side ever conceded.

Sources

• Snowflake announces pricing of initial public offering
• Snowflake opens at $245 on NYSE debut (CNBC)
• Berkshire Hathaway and Salesforce agree to buy into Snowflake IPO (CNBC)
• Snowflake FY2026 fourth quarter and full year results
• Snowflake shares drop on CEO retirement and weak guidance, Feb 29 2024 (CNBC)
• Databricks closes $15.3B financing at $62B valuation, Meta joins (TechCrunch)
• Databricks grows 65% YoY, surpasses $5.4B revenue run rate (PR Newswire)
• Databricks hits $188B valuation (TechCrunch)