Databricks vs Snowflake: The 2x Valuation Gap Halved, and Snowflake Moved

In February 2026, Databricks was priced at $134 billion, more than twice Snowflake's market value. By early September the private mark was $190 billion and Snowflake was worth about $124 billion, a gap of roughly 1.5x. The Databricks vs Snowflake premium did not shrink because Databricks slowed. It shrank because Snowflake's public price recovered faster than any private round could move.

Short answer: Databricks is valued above Snowflake mainly because it grows more than twice as fast. At about 27 times its run rate it costs more per dollar of revenue, but less per point of growth. AI revenue claims and private-market pricing explain the rest, and neither can be checked against audited numbers yet.

The Databricks vs Snowflake valuation gap, measured twice

The comparison everyone repeats comes from one week in February. On 9 February 2026, Databricks announced it had crossed a $5.4 billion revenue run rate growing more than 65%. It also completed more than $7 billion of financing, including about $5 billion of equity at a $134 billion valuation.

Coverage the next day described that price as over twice the value of Snowflake, its closest listed competitor. That week sat inside a broad selloff in software stocks, which the same report tied to fears raised by Anthropic's Claude Cowork launch. The mechanics of that selloff are covered in the breakdown of the SaaSpocalypse selloff.

Six months later the numbers had moved on both sides. Databricks closed a $5 billion round at $190 billion on 13 August, up about 42% from February. Snowflake was worth about $124 billion in early September, according to PitchBook. If Snowflake was below $67 billion in February, as "over twice" implies, its value rose by at least 85% over the same stretch.

That is the first finding, and it reframes the whole question. The February gap was a snapshot of a public company during a drawdown, set against a private price negotiated over weeks. The gap narrowed mostly because the public side recovered.

Databricks vs Snowflake side by side, with every figure labelled

The two companies report in different ways, so a side-by-side table needs a column the usual comparisons skip. Each row below says whether the figure is filed with a regulator, announced by the company, reported by an analyst, or derived by us from those numbers.

"Filed" here means Snowflake's earnings release furnished on Form 8-K. Those figures are unaudited until the annual report, but they carry securities-law liability. Databricks' figures come from its own press releases, which carry no equivalent check.

Read the last column before any other. Seven of nine Databricks cells are company statements that no regulator has reviewed. The two multiples are also measured on different bases, run rate against expected fiscal-year revenue.

Growth explains most of the Databricks valuation premium

Start with the cleanest explanation. A company growing at 80% compounds into a much larger business than one growing at 36%, even from a similar base. Investors pay for the future revenue, not the current figure.

Databricks' run rate rose from $5.4 billion to above $7 billion between February and August, about 30% in six months. Its stated growth accelerated from more than 65% to more than 80%. Snowflake also accelerated: Q2 was its third consecutive quarter of faster product revenue growth, and it raised its fiscal 2027 product revenue guide from $5.84 billion to $6.07 billion.

Both businesses are speeding up at the same time. That is unusual for two direct competitors and it says the data platform market itself is expanding. It is not a zero-sum share fight, at least not yet.

The multiple per point of growth

PitchBook puts Databricks at about 27 times its run rate and Snowflake at about 20 times the revenue it expects this fiscal year. Divide each multiple by its growth rate and the ranking flips. Databricks costs about 0.34 turns of revenue per point of growth. Snowflake costs about 0.56.

This is crude arithmetic, and it is our own, not PitchBook's. It ignores margins, durability and the fact that growth rates fall. It still makes one point hard to dispute. My read is that on growth-adjusted terms, Databricks is not obviously the expensive one.

Run rate against reported revenue

The two multiples do not share a denominator. A run rate annualises a recent period, and Databricks does not say which period or how it treats one-off consumption. Snowflake's figure is a fiscal-year guide built from reported quarters.

The bias runs in a known direction. For a company still growing fast, a current run rate understates next year's revenue, so the 27x would fall on a forward basis. The problem of reading valuations off run rates is treated at length in the analysis of OpenAI's run-rate multiple, and the same caution applies here.

Snowflake AI revenue against Databricks AI revenue: a definitions problem

The second explanation for the premium is AI exposure. On the disclosed numbers it looks overwhelming. In February, Databricks said its AI products had crossed a $1.4 billion revenue run rate, about 26% of its $5.4 billion total.

Snowflake's most specific AI figure is older and far smaller. On its December 2025 earnings call, management said it had reached a $100 million AI revenue run rate, one quarter ahead of its own plan. Against quarterly product revenue of $1.16 billion at the time, that is roughly 2%.

Two boundaries, drawn by the companies themselves

Databricks' release names "AI products" without listing them. Snowflake's figure covers revenue it attributes to AI features, and the same call said AI influenced 50% of bookings that quarter. Those two statements alone show how wide the boundary can be drawn. The same company could report 2% or half its new business as "AI", depending on the sentence.

I would not put weight on the 13x gap between the two AI figures. A company that wants an AI premium has every reason to define AI broadly, and a company defending a core business has some reason to define it narrowly. Without a definition, the number tells you about the communications strategy. Which disclosure formats actually hold up is ranked in the ranking of AI revenue disclosure metrics.

The figure that matters more: warehousing

The more telling disclosure in Databricks' August release is not about AI at all. Its Lakehouse data warehousing product passed a $1.5 billion run rate, growing over 100%. Warehousing is Snowflake's core market. That line says Databricks is winning share in Snowflake's home territory, which is a more durable claim than any AI label.

Snowflake's answer is product usage rather than dollars. Its Q2 release counted more than 9,100 accounts on its CoCo product and 5,800 on CoWork. Account counts are not revenue, and the release gives no dollar figure for either.

Free cash flow: where Snowflake's filings say more

The angle that started this post assumed Snowflake has stronger cash flow. The honest answer is that Snowflake has published cash flow and Databricks has not. Those are different claims.

Snowflake generated $1.12 billion of free cash flow in fiscal 2026, a 24% margin, and guides a 23% adjusted margin for fiscal 2027. Cash flow is seasonal: Q2 fiscal 2027 produced only $83.8 million, 5.4% of revenue, because collections cluster in the fourth quarter. Read the full year, never a single quarter.

Databricks said in February it delivered "positive free cash flow over the last 12 months." By August the wording had become "positive adjusted free cash flow." The word "adjusted" arrived without a definition, and neither statement gives a number.

Stock compensation is larger than the cash

Snowflake's filing carries a less flattering line too. Stock-based compensation was $1.60 billion in fiscal 2026, more than the $1.12 billion of free cash flow. The GAAP operating loss was $1.44 billion. Free cash flow is real cash, but shareholders pay for part of it through dilution.

Databricks almost certainly pays heavily in stock as well. Nobody outside the company knows how heavily, and that is the point. A public investor can subtract Snowflake's dilution from its cash flow. A private investor in Databricks is trusting a single adjective.

What a private mark is, and what it is not

The third explanation is the one least discussed: the two numbers are not the same kind of price. Snowflake's $124 billion is what the last trade implied for every share, and it changes every second the market is open. Databricks' $190 billion is the price one group of investors agreed for newly issued preferred shares in one round.

Late-stage preferred shares usually carry protections that common shares do not, such as liquidation preferences that pay out first in a sale. A headline valuation applies that protected price to every share, including common stock that lacks the protection. How those terms shape real payouts is set out in the explainer on liquidation preferences and IPO ratchets.

No exit date, and no audited figures behind it

Databricks' chief executive Ali Ghodsi has ruled out a 2026 listing, calling it a terrible year to go public, and said 2027 is the earliest. PitchBook reported that the company had not filed an S-1, even confidentially. Until it does, there are no audited numbers to test the private price against.

PitchBook's own model is blunt. Its analyst Harrison Rolfes estimated an operating value of $68.7 billion, a 64% discount to the round, and an optimistic case of $182.1 billion that still falls short of $190 billion. He called Databricks a high-quality business but an unattractive investment at the current price. That is one analyst's model, and a model with a 10-year horizon is mostly assumption.

Where this argument is weakest

Every number here deserves a second look, and three of them deserve it more than the rest.

The February Snowflake mark is an inference

"Under $67 billion" comes from a news report saying Databricks was worth over twice Snowflake. I could not open a dated market-value figure for Snowflake that week, and published market caps for 2026 conflict badly by source and date. If Snowflake was closer to $67 billion than below it, the 85% rise is the floor, not an estimate. The direction of the finding holds either way. Its size is softer than the chart suggests.

The growth-adjusted case may flip back

Dividing a multiple by a growth rate assumes growth persists in proportion. It does not. PitchBook assumes Databricks growth slows to 48% next year and into single digits by the mid-2030s. If Databricks decelerates faster than Snowflake does, the 0.34 against 0.56 comparison flips. The case for Snowflake is that its slower growth comes with filed margins, filed cash flow and a stock you can sell on any trading day.

There is also a selection problem. Databricks chooses which metrics to announce, and it changes them between releases. February named an AI run rate and net retention. August named neither. A company publishing its best numbers each time will always look stronger than one that must file every number every quarter.

How data leaders should read the data platform market

If you buy a data platform rather than shares, the valuation gap matters for different reasons. Both vendors are well funded, growing faster than a year ago, and not going anywhere. Vendor viability is not the question.

The useful signal is pricing pressure. Databricks growing its warehouse business at over 100% means Snowflake has a credible alternative in the room on every renewal, and the reverse is true for Databricks' AI products. Your renewal is the one place where this competition shows up as money you keep.

Both price on consumption, which shifts risk to you. Snowflake's net revenue retention of 126% means the average customer spends 26% more each year, and some of that is new workloads rather than new value. The pattern across vendors is laid out in the review of net revenue retention in Q2 filings.

My position: run any new AI or agent workload as a priced bake-off between the two, with your own data and a fixed budget. Neither vendor's AI revenue claim tells you which one will be cheaper for your workload. Databricks' new Lakebase database, now past a $100 million run rate, adds a third option for agent state worth testing against the alternatives in the comparison of AI agent database pricing.

Frequently asked questions

Where to start

If you hold or price either company, rebuild the gap yourself after each event rather than quoting the February ratio. Snowflake reports its fiscal third quarter around early December. Write down its market value that day, divide Databricks' latest round by it, and note whether the gap moved on news from Snowflake or from Databricks.

If you run a data team, take the next renewal on either platform as the moment to run a priced trial against the other. Ask the incumbent for its AI pricing in writing, with the unit of consumption defined, before the trial starts.

Related reading