Snowflake shares jumped roughly 23% after hours on Wednesday, to about $374 and through their 52-week high, after the company reported a third consecutive quarter of accelerating product revenue and raised its full-year outlook. Jim Cramer called it a thing of beauty. The acceleration is real.
So is the sentence buried in the same guidance that almost no coverage put next to it: Snowflake cut its non-GAAP product gross margin target to 74.0%, specifically because the AI workloads driving the acceleration carry lower contribution margins than the core data platform. Management said the reason out loud, which is that Snowflake has to pay for the models and the compute underneath its AI products. The market repriced the growth rate on Wednesday night. It does not appear to have repriced the mix.
The Quarter, Honestly
The beat was not marginal. Product revenue reached $1.49 billion, up 37% year over year, with total revenue of $1.55 billion against roughly $1.48 billion expected, and adjusted earnings of $0.62 a share against $0.45. CNBC put total revenue growth at 35% for the quarter ended July 31. Full-year product revenue guidance rose to $6.07 billion, about 36% growth, and the non-GAAP operating margin target went to 14.5% from 13.5% after the quarter itself came in at 15%. Free cash flow margin guidance held at 23%.
Three straight quarters of acceleration in a business this size is genuinely rare, and management attributed roughly half of it to AI products, naming CoCo and CoWork, with the balance from faster customer migrations and wider adoption of notebooks and applications. Chief executive Sridhar Ramaswamy has been arguing for a while that as AI makes software cheaper to build, the durable asset is a company’s data and the context around it. This quarter is the best evidence he has offered for that thesis.
The Two Numbers Underneath
Set the celebration aside and two disclosures deserve more weight than they got.
The first is the margin guide. Product gross margin coming down to 74% is a small number in isolation and a structural signal in aggregate, because it tells you the incremental dollar Snowflake is winning is worth less than the average dollar it already had. Snowflake is a consumption business, which has always been its strength: customers who use more pay more, with no seat negotiation. But AI consumption runs on GPUs and third-party models that Snowflake buys, so the same mechanism that converts AI enthusiasm into revenue also converts it into cost of goods sold. Seeking Alpha flagged the margin and revenue guidance together, which is the correct way to read them.
The second is remaining performance obligations. RPO closed at $9.0 billion, up 30% year over year, with 54% expected to convert to revenue in the next twelve months, per the company’s second-quarter results release. Product revenue grew 37%. The contracted book is growing about seven points slower than the revenue being recognised out of it. That is not alarming at these levels and consumption models routinely recognise ahead of bookings, but it is the opposite of what a 23% move implies. A stock that adds nearly a quarter of its value in a night is pricing an acceleration that extends; RPO is the line that tells you how much of the extension is already signed, and it is decelerating relative to revenue.
Consumption Cuts Both Ways
The bull case and the bear case share a mechanism, which is what makes this quarter interesting rather than simply good.
Consumption pricing means Snowflake captures AI adoption immediately, without waiting for a renewal cycle, and that is why the acceleration showed up three quarters in a row while seat-based software vendors were still explaining their AI roadmaps. It also means Snowflake carries the volatility. Experimental AI workloads spin up fast and spin down just as fast, and nothing in a consumption model distinguishes a workload that has entered production from one a customer is still evaluating. Snowflake is not disclosing that split, and until it does, the durability of half the acceleration is an assumption rather than a disclosure.
The comparison worth making is with how the market has treated AI revenue elsewhere in enterprise software this year. When Salesforce reported, attention landed quickly on what was operating performance and what was not. Snowflake’s AI revenue is unambiguously operating revenue, which is a real point in its favour, and it is simply lower-margin operating revenue. Those are different criticisms and only the second one applies here.
Where We Land
This was a strong quarter and the stock was too cheap for the growth it is now delivering, so a large move was justified. Our disagreement is with the size of it.
A 23% repricing treats a mix shift as free. It is not: Snowflake told investors in the same breath that the growth is coming in at a lower gross margin, and cutting a margin target while raising a revenue target is the company being straight with the market about a genuine tradeoff. Investors took the raise and skipped the cut. Snowflake has earned the benefit of the doubt on execution, having now beaten and raised repeatedly through this cycle, but the operating margin guide of 14.5% is what turns this into a durable story, and that number depends on the 74% holding.
Watch two things next quarter. Whether product gross margin comes in at or below 74%, because a second cut would mean AI mix is scaling faster than Snowflake can negotiate its model and compute costs down. And whether RPO growth closes the gap with revenue growth, because a consumption business recognising ahead of its contracted book for several quarters running is eventually recognising ahead of its demand. Neither is a problem today. Both are the difference between a company that grew 37% once and one that keeps doing it.