Snowflake: AI consumption rebound is real, but the price now assumes years of it

Snowflake's stock jumped 22% in after-hours trading on Wednesday after its fiscal second-quarter report, to about $374, in a single session reversing the day's regular-session decline and pushing the shares above their old high. The move was the market's verdict on a quarter that did what the most bullish analysts had spent August predicting: product revenue accelerated to 37% growth — a third straight quarter of speeding up — and the company raised its guidance for the year. The surprise is not that the rebound is real. It is that the price now assumes it keeps going for years.

Before unpacking the price, it helps to see why this quarter mattered, because it goes to how Snowflake makes money. Unlike most SaaS companies, which bill flat subscription fees, Snowflake charges for actual usage — compute and storage consumed. That model cuts both ways. During a 2024-25 slowdown, customers burned through pre-purchased credits and optimized their bills, so revenue growth slowed sharply even though customers stayed. A consumption model conceals demand weakness until warnings show up in usage.

The bull case was that the slowdown was always temporary and that AI workloads would re-accelerate usage. In mid-August, Oppenheimer raised its target to $400 from $295 on exactly that thesis, citing stronger consumption, migration, and usage of CoCo, Snowflake's AI coding agent that generates data queries grounded in each customer's own data. It was a punchy call — among the highest targets on the Street — and it carried the risk that a target was running ahead of proof.

Wednesday's report supplied the proof. Product revenue of $1.49 billion grew 37% year over year, the third consecutive quarter of acceleration, after a 34% print last quarter. Total revenue reached $1.55 billion, up 35%. Net revenue retention held at 126% — meaning the average existing customer spent 26% more with Snowflake than a year earlier — and 828 customers now generate over $1 million in trailing twelve-month product revenue. The company lifted its full-year product revenue outlook to $6.07 billion, about 36% growth, and guided the current quarter to 37-38% growth.

That retention number is the heart of the consumption thesis. Because Snowflake bills on usage rather than subscriptions, revenue can only compound if the existing base keeps consuming more. A 126% net revenue retention ratio on a $1.5 billion quarterly base means the expansion is coming from customers actually running more workloads, not from one-time deals. Management's non-GAAP operating margin expanded more than 400 basis points to 15%, evidence that the growth is starting to convert into profit rather than just expense.

The stock's reaction — a 22% after-hours spike after the shares were already up roughly half their value this year — tells you how much of that was already priced in before the print. This is the point where the story splits into two questions the market often blurs: is Snowflake a good company, and is it a good stock at this price?

The company is not in dispute. The growth is real, the model retains revenue well, and the AI driver is showing up in actual consumption rather than a pitch. The dispute is the multiple. At about $374, the market capitalizes Snowflake at roughly $120 billion against about $6.1 billion in projected full-year product revenue — roughly 20 times forward product sales, somewhat less against total revenue. That is a premium multiple reserved for companies expected to sustain very high growth for many years.

The bears' rebuttal is worth taking seriously rather than waved off. Contracted-but-unrecognized revenue, called remaining performance obligations, grew just 30% — measurably slower than the 37% product revenue print. RPO is the future committed business, so a base that grows more slowly than current revenue is a hint that the acceleration could be front-loaded. The company still loses money on a GAAP basis, posting a $192 million net loss, because employee stock compensation is large enough that non-GAAP profit and GAAP loss coexist. And Snowflake faces relentless competition from Databricks, which is cheaper on some workloads and runs on more open standards. None of these kill the story, but they are the reasons a 36%-growing company does not always deserve a 20-times-sales sticker.

So what changes the view? The bull case now rests on a single falsifiable bet: that consumption keeps accelerating through the rest of fiscal 2027. The confirmation to watch is whether RPO growth catches up to product revenue growth — a sign the acceleration is durable rather than pulled forward — and whether that 126% retention ratio holds as CoCo incentives translate into billable usage. Hold the stock to that proof rather than to the momentum behind it.

Snowflake had a bad 2025 that reset expectations, and it has now more than doubled off those lows as the AI consumption narrative converted into disclosed numbers. That is a legitimate recovery. The practical question for a buyer today is narrower and unglamorous: whether paying roughly 20 times forward sales for a company that could stumble on execution, competition, or a slowing consumption base gives you a margin of safety or just a seat at a crowded AI trade. The company earned the rally. Whether the stock has earned this price is a different, and so far unanswered, question.