Snowflake Is Priced as an AI Winner While Databricks Wins the AI Budget

---

Begin with the genuine strength, because Snowflake is a real and dominant business and the critique is about relative position and price, not quality. Snowflake pioneered the cloud data warehouse — a place for enterprises to store and analyze their structured data — and built it into a franchise with enormous "data gravity": once a company's data lives in Snowflake and its analysts and applications are built on top of it, moving away is painful and expensive, which makes the customer relationships sticky and durable. In its most recent quarter, product revenue grew 34% to roughly $1.33 billion, the company added enterprise customers at a healthy clip, and its AI features have seen broad adoption — nearly 70% of customers using some form of AI, and a growing share using its newer intelligence products. The market rewarded all of this with a roughly 30% surge in the stock.

So this essay does not argue that Snowflake is failing. It is a genuinely good company with a real moat in its core business. It argues that the stock has been priced and celebrated as a clear winner of the AI-data era at the exact moment three uncomfortable truths are visible in the numbers: that the business remains deeply unprofitable, that its growth engine is decelerating in a way the consumption model partly disguises, and that the company genuinely winning the AI-specific battle is its private rival, Databricks — which is more profitable, faster-growing in AI, and valued by private investors at more than twice Snowflake's public price.

A decade public, and still $1.3 billion in the red

Start with profitability, because it is the simplest and most overlooked fact. Snowflake generated about $4.7 billion of revenue in its last fiscal year and lost approximately $1.3 billion doing it. This is not a young startup still finding its model; Snowflake went public years ago, is one of the largest and most established software companies in its category, and is still posting nine-figure quarterly losses on a GAAP basis. The bulls point, fairly, to the company's adjusted profitability and free cash flow, and there is real improvement on those non-GAAP measures. But the GAAP loss is large and persistent, driven substantially by enormous stock-based compensation and the heavy investment required to keep growing — and a company that has been public for years and still cannot turn a real accounting profit on nearly $5 billion of revenue is a company whose path to durable profitability is more a matter of faith than of demonstration.

The contrast with Databricks sharpens the point uncomfortably. Databricks — Snowflake's primary rival, still private — reportedly turned free-cash-flow positive while growing faster, a detail that separates it from the many AI-era companies still bleeding cash. So the public company that the market just bid up 30% as an AI winner is the one losing over a billion dollars a year, while the private rival winning the AI battle is generating cash. When the less-profitable company is celebrated and the more-profitable one is its competitor, the celebration is about narrative momentum, not financial reality. There is a particular irony in an "AI era" that was supposed to reward efficiency and automation handing its richest public-market enthusiasm to the less-efficient of two rivals, simply because that one happens to be the one investors can buy today — the accident of being public, rather than the merit of the business, doing the work the valuation pretends the fundamentals are doing.

The consumption model is decelerating behind the headline

Now the growth, because the 34% headline conceals a more important trend. Snowflake's revenue runs on a consumption model: customers pay for the compute credits they actually use, rather than a fixed subscription. This is a genuinely powerful model in a growth phase — as customers run more workloads, they consume more credits, and revenue compounds without Snowflake having to sign proportionally more new customers. The bulls love it because AI workloads are compute-hungry, so the AI era should, in theory, drive ever-rising consumption.

But the consumption model has a darker side that the single best metric reveals. Snowflake's net revenue retention — how much more an existing cohort of customers spends this year versus last — has fallen from over 170% in its early post-IPO years to around 125% now. That is still a healthy number in absolute terms, but the trajectory is the tell: existing customers are expanding their spending far less aggressively than they used to. And the consumption model is precisely what makes this dangerous, because consumption revenue is the opposite of locked-in subscription revenue — customers can, and increasingly do, optimize their usage to spend less, tuning their queries and workloads to consume fewer credits, and in any downturn they can cut compute spending immediately rather than waiting for a contract to expire. The model that compounds beautifully on the way up gives customers a dial they can turn down at will, and the falling net revenue retention suggests they have been turning it. A 34% growth rate built on a decelerating consumption base, where the customers control the spending dial, is a less durable thing than a 34% growth rate built on contracted, expanding subscriptions — and the market, dazzled by the AI-consumption story, has priced the optimistic interpretation.

The AI battle is being won by the other guy

Here is the most important and least appreciated fact, the one that most undercuts the "Snowflake is an AI winner" narrative: in the actual competition for AI workloads, Snowflake is losing to Databricks. The two companies grew up serving adjacent but different needs — Snowflake the data warehouse for structured analytics and business intelligence, Databricks the data lakehouse for machine learning, data engineering, and AI model work. And the AI era is flowing toward Databricks' turf, because building and running AI is exactly the ML-and-data-engineering work Databricks was built for.

The numbers are stark. Databricks reportedly generates more than $1 billion in AI revenue run-rate and is winning something like 70% of incremental AI and machine-learning budgets in head-to-head enterprise evaluations. Snowflake's competing AI products — its Cortex AI and intelligence offerings — are earlier-stage, with an AI revenue run-rate on the order of $100 million, roughly a tenth of Databricks'. So while Snowflake can truthfully say that 70% of its customers are "using AI features," the actual dollars of AI workload — the budget that represents the future growth the valuation is paying for — are flowing predominantly to its rival. Snowflake is participating in AI; Databricks is winning it. The distinction matters enormously, because the entire bull case for Snowflake's premium rests on it capturing the AI-data opportunity, and the evidence is that the company capturing it is the one Snowflake is losing to.

This does not mean Snowflake collapses — the two platforms genuinely coexist, with many large enterprises running Snowflake for their structured analytics and Databricks for their AI and ML work, and Snowflake's core BI franchise is sticky and durable. But "durable core business plus losing the AI battle" is a very different investment than "the picks and shovels of the AI era," and the stock has been priced as the latter.

What the bulls genuinely get right

In fairness, the bull case is real and Snowflake's quality is not in question — the debate is the price and the AI-competitive position. Several points genuinely favor Snowflake. The data-gravity moat is real and powerful: enterprise data is enormously sticky once it lives in a platform, and Snowflake's core analytics business is defensible and growing. The re-acceleration to 34% product growth is genuine and reflects real demand, not financial engineering. The consumption model, for all its volatility, does mean that as AI and analytics workloads grow, Snowflake's revenue compounds efficiently. The AI-feature adoption is real — getting 70% of customers to touch AI features is a genuine distribution advantage, and Snowflake's bet is that proximity to the customer's own data is the single ultimate moat in all of enterprise AI, because the AI ultimately has to run where the data already lives. If that bet is right, Snowflake's early-stage AI revenue could yet inflect sharply as its customers operationalize AI on the data already sitting in Snowflake. And the company is improving on adjusted profitability and free cash flow even as the GAAP loss persists.

The honest synthesis is that Snowflake is a strong company with a durable core and a contested AI future, priced and celebrated as if the AI future were already won. The bull is right that the data-gravity moat is real and the AI-on-your-own-data thesis is plausible. The skeptic notes that the actual AI revenue and the actual AI budget wins are going to Databricks, that the consumption model masks a real deceleration, and that a decade in, Snowflake still loses a billion dollars a year. A stock that just surged 30% on AI enthusiasm, trading at a rich multiple of sales, has priced the optimistic resolution of a competition it is currently losing — which is a demanding place to start, and a more demanding one still when the company beating you is about to become a publicly traded benchmark the whole market can measure you against.

The profitability that depends on which number you use

It is worth dwelling on the gap between Snowflake's GAAP loss and its adjusted profitability, because the difference between them is largely one enormous expense that the bull case waves away: stock-based compensation. Snowflake, like many software companies, pays its employees heavily in stock, and under GAAP that compensation is a real expense that drives much of the billion-dollar loss. The company prefers to direct investors to its "non-GAAP" results, which add that stock compensation back and show healthy adjusted operating margins and free cash flow — and the bulls follow, treating the adjusted numbers as the true picture and the GAAP loss as an accounting artifact.

But stock-based compensation is not an artifact; it is a real cost borne by shareholders in the form of dilution. Every year Snowflake hands employees billions of dollars of stock, the existing shareholders' slice of the company shrinks, and the company frequently buys back stock simply to offset that dilution — spending real cash to mop up the shares it prints as "non-cash" compensation. So the free cash flow the bulls celebrate is, in part, being consumed to neutralize the dilution from the compensation the adjusted earnings excluded. The honest way to see Snowflake's economics is somewhere between the rosy non-GAAP picture and the stark GAAP loss — a company that is genuinely improving its underlying efficiency but is still, when you count the cost of the stock it pays out, a long way from the durable, GAAP-real profitability that its premium valuation implies it has already achieved. Profitability that exists only after you stop counting one of your largest expenses is a softer kind of profitability than the headline adjusted margin suggests.

A picks-and-shovels trade priced like the gold

Step back and Snowflake fits a pattern documented elsewhere in this series: the "picks and shovels" of a boom priced at the premium normally reserved for the gold itself. The appeal of owning Snowflake, like owning the data-center suppliers and the chip-design partners, is that it supposedly lets you profit from AI without betting on any single AI application — you own the platform every AI workload must run on. That is a genuinely attractive idea. But the whole point of a picks-and-shovels investment is that it is supposed to be safer, and therefore priced more modestly, than the speculative end of the boom — and Snowflake, trading at a rich multiple of sales after a 30% surge, is being priced with the optimism of a frontier bet rather than the modesty of a safe supplier.

And the safety is more questionable than the picks-and-shovels framing implies, precisely because the AI workloads are flowing to Databricks. A shovel-seller is safe only if all the miners buy his shovels; a shovel-seller whose biggest competitor is winning most of the new mining business is not the safe, diversified supplier the metaphor promises — he is a contender in a contest, priced as if he had already won the franchise. Snowflake's core analytics business is the genuinely safe, durable part, and on that part a reasonable multiple is defensible. The premium on top of it is paying for the AI-data future, and that future is precisely the contested ground where Snowflake is, for now, the runner-up. Paying a gold-rush multiple for the shovel-maker who is losing the richest seam to a rival is the specific mispricing the 30% pop embodies.

The Databricks IPO is the reckoning

There is a specific event on the horizon that will test all of this, and it is worth naming because it reframes the whole comparison: Databricks is expected to go public, and when it does, the market will be able to directly compare the two companies in a way it cannot today. Right now, Snowflake's premium is judged in isolation and against an abstract sense of the AI-data opportunity. The moment Databricks trades publicly — reportedly being valued privately at $165 to $175 billion, more than double Snowflake, on a rapidly growing, AI-winning, cash-generative business — the market will have a direct, side-by-side benchmark: the profitable AI-winner versus the unprofitable AI-also-ran.

That comparison could cut either way for sentiment, but it introduces a clarifying risk for Snowflake. Private markets have valued Databricks like a model lab, at a premium that public markets may or may not extend when the IPO arrives — but if public investors do validate Databricks' premium, they will be doing so for a company that is beating Snowflake in AI while being more profitable, which raises the obvious question of why Snowflake deserves its own premium for losing that battle. And if public investors discount Databricks to something closer to Snowflake's multiple of sales, that repricing of the category leader would drag on Snowflake too. Either way, the arrival of a direct, public, better-performing comparison removes the ambiguity that currently lets Snowflake be priced as the AI-data winner. The reckoning is not whether Snowflake survives — it will — but whether it can sustain a winner's valuation once the market can see, in public quotes, who is actually winning.

The kicker

Snowflake is a good company that just had a great quarter and got celebrated as an AI champion, and the celebration was not baseless — the data-gravity moat is real, the growth re-accelerated, and the AI-feature adoption is genuine. But the celebration skipped over the parts that matter most for the price the stock now carries: that the company still loses over a billion dollars a year a decade after going public, that its consumption engine is decelerating beneath the headline as customers learn to spend less, and that the actual AI budget — the prize the entire premium is paying for — is being won by the private rival valued at twice Snowflake's price. Being in the AI-data business is not the same as winning it, and Snowflake has been priced for winning a battle that, on the evidence of the revenue and the budgets, it is currently losing. The data may have gravity. The AI dollars, for now, have somewhere else to go.

Snowflake's customers are nearly all using its AI features, and almost none of them are spending real AI money there yet — that is flowing to the rival down the road, the private one that makes a profit and wins the bake-offs. The market threw a thirty-percent party for the AI winner. It may have invited the wrong company — and the moment the right one rings the opening bell on its own IPO, the guest list gets a great deal harder to fudge, because the market will finally be able to see, in two live prices side by side, which data platform the AI era actually chose.