Snowflake Beat the Plan, Not the Prior Year: A 500bp Margin Beat on a Guide Set Low, and an FY26 That Only Gets Back to FY24 Margins

Results vs. Consensus

Q4 FY2025 Scorecard

Consensus figures are the pre-print sell-side averages; providers disagreed within a narrow band on revenue ($952.7M to $957.5M) and by a penny on non-GAAP EPS ($0.17 to $0.18). We use the most widely cited of each. Guidance figures are Snowflake's own Q3 FY25 outlook. The implied non-GAAP operating income at the 4% guide is our calculation applied to actual revenue and is presented as an approximation, since Snowflake guides margin rather than dollars.

Year-Over-Year Comparison — Fourth Quarter

Sequential Comparison

Q3 FY25 non-GAAP margin figures are as Snowflake reported them, rounded to whole percentage points, so the sequential margin deltas above are approximate to the nearest 100bp.

Full-Year FY2025 — Where the Story Actually Sits

The four reported FY25 quarters sum to the full-year figures shown. Product revenue by quarter was $789.6M, $829.3M, $900.3M and $943.3M; the $0.1M residual against the reported $3,462.4M is rounding in the first three quarters' disclosed figures. The stock-based compensation line is the cash-flow-statement figure net of capitalized amounts, which is the basis management uses for the "41%" reference; the income-statement reconciliation figure of $1,564.3M additionally includes employer payroll-tax items.

Quality of Beat

Revenue. This is a clean beat with no help. There is no acquisition contribution to strip out, no FX distortion in a business that reports and largely bills in dollars, and no accounting change. Product revenue is recognized on consumption, so the $32.3M above the guidance ceiling represents platform usage that actually happened rather than bookings that were signed. The one item that deserves scrutiny is the disclosure that several large customers exhausted their capacity commitments mid-contract and elected to buy on demand rather than re-up early. That behavior pulls revenue into the current period without a corresponding bookings event, which is precisely why remaining performance obligations decelerated to 33% growth from 55% while revenue accelerated past the guide. The consumption was real. The contractual coverage behind it thinned.

Margins. The 9.4% non-GAAP operating margin decomposes with unusual clarity, and the decomposition is the story. Non-GAAP gross margin fell 211bp year on year, to 72.6% from 74.7%. Non-GAAP operating expense fell 236bp as a share of revenue, to 63.2% from 65.5%, essentially all of it from sales and marketing, which came down 240bp. Research and development actually rose 19bp as a share of revenue. Net those two and you land within a few basis points of the +25bp the operating line delivered. Snowflake's efficiency program is working exactly as advertised and its entire output is being absorbed by the cost of serving AI workloads. That is a very different quarter from one in which margin expanded.

EPS. Non-GAAP diluted EPS of $0.30 beat the $0.18 consensus by 67% and fell 14.3% against the $0.35 printed a year ago. The apparent contradiction resolves below the operating line. Other income (expense), net was $2.4M this quarter against $47.5M in Q4 FY24, a $45.1M swing driven almost entirely by unrealized gains on strategic equity investments in the year-ago period. Those gains are not adjusted out of Snowflake's non-GAAP net income, so the prior-year comparative was flattered by a mark-to-market item that has nothing to do with the data platform. Adjust for it and the underlying earnings trajectory is up sharply rather than down. The offsetting drag is share count: non-GAAP diluted shares rose to 370.7M from 363.8M, including 14.4M shares from the if-converted treatment of the zero-coupon notes issued in September 2024, against none a year ago.

Segment Performance

Snowflake operates as a single reportable segment and does not disclose revenue by product, vertical, or geography. What it does disclose is a two-line revenue split, a full operating-expense breakdown on both a GAAP and non-GAAP basis, and a handful of consumption metrics. The analysis below works with what exists rather than inventing a segmentation the filings do not support.

Revenue Composition

Product Revenue

Product revenue is the whole company and it is recognized on consumption, not on contract duration. That construction is the reason Snowflake is analytically harder than a seat-based SaaS business: revenue reflects what customers actually ran last quarter, and a customer who optimizes a query pattern reduces Snowflake's revenue without churning. The FY25 sequence of year-on-year product growth was 34%, 30%, 29% and 28%. The deceleration is steady and shallow, which is the shape you want at this scale, but it is a deceleration in every quarter of the year, and the Q1 FY26 guide steps it down again to roughly 21.3% at the midpoint.

Management pushed back on reading Q1 too literally, and the pushback is legitimate arithmetic. Q1 FY25 ended April 30, 2024 and contained the February 29 leap day; Q1 FY26 does not.

"And remember, our guidance for Q1 seems low, but that is because of the -- that we miss one business day because of leap year and that's a fair chunk of revenue that comes out of that when you're looking at that growth rate just in Q1."
— Mike Scarpelli, CFO

One day out of ninety is roughly 110bp, which lifts the underlying Q1 growth rate to something near 22.4%. That does not close the gap to 28%.

Assessment: The consumption engine is healthy and slowing at a manageable rate. What has changed is that the slowdown is now visible in the guide rather than only in the reported line, and management has chosen to bridge it with second-half product contribution rather than with core-business acceleration. That is a harder promise to verify and a harder one to keep.

Professional Services and Other

The services line grew 18.7% to $43.5M and continues to lose money on a GAAP basis, at a $(16.5)M gross loss for the quarter and $(58.6)M for the year. On a non-GAAP basis it cleared $0.9M of gross profit in the quarter, down from $2.6M a year ago, and 3% gross margin for the year against 5%. Snowflake runs services as a migration accelerant rather than a profit center, and management explicitly framed it that way, citing improved migration execution from both internal services and partners as a driver of the FY26 workload pipeline.

Assessment: Immaterial to earnings, material to the FY26 story. If the second-half acceleration is going to come from migrations landing in production, this line is the leading indicator worth watching, and 18.7% growth against 27.8% in product is not yet the signature of a migration wave building.

Cost Structure — Where the Margin Went

Assessment: This table is the quarter. Sales and marketing produced 240bp of genuine leverage, which is the largest single-quarter improvement in that line the company has shown, and it is exactly what the efficiency program was designed to deliver. Cost of revenue took 211bp back. Research and development did not lever at all, which is a deliberate choice given the product cadence but leaves the operating line with almost nothing to show. For FY26 management guides product gross margin down a further point while guiding operating margin up 160bp, which means the cost-of-revenue drag continues and the opex line must produce roughly 300bp of leverage on its own.

Key KPIs

The two customer-count rows show no Q4 FY24 figure because Snowflake restates prior-period counts for acquisitions, consolidations and spin-offs, and re-bases the Forbes Global 2000 cohort to the current-year list, so the restated base is not disclosed. The year-on-year growth rates shown are the company's own. For reference, the as-originally-reported Q4 FY24 counts were 461 and 691; those figures will not reconcile arithmetically to the growth rates above, which is why we do not present them as comparatives.

The two KPIs that disagree. The >$1M customer cohort added 38 accounts sequentially, the strongest quarter of FY25, while net revenue retention fell another point and RPO growth halved. Those move in opposite directions because they measure different things: the cohort count is a land-and-expand success metric, retention is a same-customer consumption metric, and RPO is a contractual-coverage metric that management explicitly told us was distorted this quarter by large accounts choosing to consume on demand rather than re-contract early. Taken together the picture is a business acquiring and growing large customers well while the average dollar of installed base expands more slowly each year. That is the normal maturation path for a consumption platform. It is also exactly the path that makes a 24% full-year guide with a back-half step-up difficult to underwrite.

The KPI nobody asked about. Forbes Global 2000 customers fell from 754 to 745 sequentially. The release attributes the cohort definition to the 2024 Forbes list and notes that counts are adjusted for market activity, so the decline is plausibly a re-basing artifact rather than churn among the world's largest enterprises. Management did not mention it, and no analyst raised it. We flag it not because we believe it signals attrition but because a nine-account sequential decline in the flagship enterprise-penetration metric passed without a sentence of explanation.

Key Topics & Management Commentary

Overall Management Tone: Confident and well rehearsed, with a noticeably tighter script than the transition-year calls that preceded it. The prepared remarks led with product velocity and customer proof points rather than with the numbers, and the numbers were delivered as confirmation rather than as argument. Where management was least convincing was on the second-half acceleration embedded in the FY26 guide, which was defended with qualitative pipeline language and an explicit refusal to quantify new-product contribution, and on the gross-margin trajectory, which was acknowledged in the guide and addressed only with a forward hope about GPU access. The retirement announcement was placed at the end of the prepared remarks and immediately reframed as continuity rather than change.

1. The Efficiency Program Landed, and It Landed in One Line

Snowflake spent FY25 restructuring how it operates, and Q4 is the first quarter where the effect is unambiguous in the reported numbers. Non-GAAP operating margin of 9.4% against a 4% guide is a 500bp-plus beat on plan, and the whole of it traces to sales and marketing coming down 240bp as a share of revenue.

"Last quarter, we discussed our plans to improve efficiency. These efforts include centralizing teams, targeted early career hiring, removing redundant management layers, and continuous performance management. In Q4, these efforts yielded meaningful margin gains."
— Mike Scarpelli, CFO

Management also framed this as permanent rather than episodic, describing the operational rigor instituted in FY25 as the company's way of life going forward, with headcount growth concentrated in engineering and sales.

Assessment: Real and durable, and the most important positive in the print. The caution is that a program producing 240bp of leverage in the same year gross margin gave back 211bp is a treadmill, not a staircase. The FY26 guide asks this program to produce leverage again while gross margin declines again.

2. Product Gross Margin Is Guided Down, and the Explanation Is a Hope

Non-GAAP product gross margin was 75.8% in Q4 against 78.0% a year ago and 76.3% for FY25 against 77.8% in FY24. FY26 is guided to approximately 75%. This is the second consecutive year of compression and the first year in which management has guided it down in advance.

"Turning to margins. We expect non-GAAP product gross margin of approximately 75%. Longer-term, we expect easier GPU access and growing AI revenue to benefit product gross margins associated with new product features."
— Mike Scarpelli, CFO

The forward argument has two legs and both are conditional. Easier GPU access depends on a supply market Snowflake does not control. Growing AI revenue improving margin depends on AI features eventually being sold at accretive prices, which is the opposite of what is happening now, since the compression is itself evidence that AI workloads currently carry a lower incremental margin than the warehouse.

Assessment: This is the single most underweighted item in the quarter. A 76% gross margin business guiding to 75% while promising 160bp of operating margin expansion is asking a lot of the opex line, and it removes the pricing-power cushion that has historically justified a premium multiple on this name.

3. RPO Growth Halved and Management Pre-Empted the Question

Remaining performance obligations reached $6.9B, up 33%, against $5.7B and 55% growth in Q3. Rather than wait for the question, Scarpelli addressed it in the prepared remarks.

"During the quarter, several large customers ran out of capacity before their contract [end date] (ph) as the revenue outpaced their contracted bookings. Instead of pulling forward the renewal cycle, these accounts are now purchasing as they consume. This is common for large customers and we do not view this choice as indicative of their future consumption patterns."
— Mike Scarpelli, CFO

The explanation is internally coherent and, on its face, benign. Customers burning through commitments early is a consumption signal, not a demand signal in reverse. But part of the deceleration is also comparison arithmetic: Q4 is Snowflake's bookings-heavy quarter and Q4 FY24 delivered a very large RPO step-up, so the 55% Q3 print was struck against a soft base and the 33% Q4 print against a hard one. Both effects are real and management surfaced only one of them.

The corroborating datapoint sits on the balance sheet. Total deferred revenue grew 17.3% year on year to $2,595.5M while RPO grew 33%. Contracted-but-uninvoiced backlog is growing faster than invoiced backlog, which is what you would expect if large customers are lengthening the gap between commitment and billing rather than shrinking the commitment.

Assessment: We accept the explanation and do not accept the framing. The disclosure was volunteered, specific, and consistent with the deferred-revenue evidence, which is more than most managements offer on a decelerating backlog metric. But an account consuming on demand at list terms until contract expiry is an account with less contractual coverage behind next year's revenue than it had, and that is the definition of lower visibility regardless of intent.

4. Net Revenue Retention Gets a Floor, Verbally

Net revenue retention of 126% is down from 131% a year ago, the fourth decline across five prints. Asked directly where it settles, management gave the most specific forward commitment of the call.

"First of all, I think NRR is pretty stable. I'm not saying it's not going to go to 125% or 124%, but I don't really see it. It's going to be in the mid 120s is what we are predicting."
— Mike Scarpelli, CFO

Assessment: This is the commitment to grade next quarter and the one most likely to be tested. A mid-120s retention rate is a genuinely strong number for a platform at this scale, and if it holds it removes the most mechanical bear argument against the FY26 guide. If it prints 124% or below, the FY26 revenue bridge loses its foundation and the back-half acceleration becomes arithmetically dependent on new products alone.

5. The FY26 Guide Is Back-Half Weighted by Construction

FY26 product revenue is guided to approximately $4.28B, up 23.6% against FY25's $3,462.4M. Q1 is guided to $955M to $960M, roughly 21.3% growth at the midpoint. That combination requires the remaining three quarters to grow about 24.3% in aggregate, and management placed the step-up specifically in the second half.

"We are forecasting stable-growth within our core business. We expect new product features to contribute to the step-up in year-over-year growth rates in the second-half of the year."
— Mike Scarpelli, CFO

Pressed on what underwrites the acceleration, the answer named three sources: the core business, new products entering production, and new-logo cohorts acquired during FY25 maturing into consumption.

"It's both. The core of our business is very strong, coupled with the new products that we see that will be going into production. Also on top of, I'm pleased with the muscle we've been building on new customer acquisitions and that's going to start to kick-in, in the second half of the year."
— Mike Scarpelli, CFO

Assessment: Every component of the acceleration is real as a mechanism and unquantified as a forecast. Snowflake has historically guided conservatively and beaten, which argues the $4.28B is a floor. But a guide whose growth rate improves after the first quarter, on contributions management has explicitly declined to size, carries more forecast risk than the same number front-loaded. This is the item that determines the rating.

6. Management Withdrew a Disclosure

For several quarters Snowflake had quantified the revenue headwind from platform performance improvements, the efficiency gains that let customers do the same work for fewer credits. That practice ended on this call.

"As always, our forecast includes headwinds associated with performance improvements. As I said last quarter, product improvements are an ongoing part of our business and we will not be breaking out assumptions for specific features moving forward."
— Mike Scarpelli, CFO

Assessment: The rationale is defensible: performance work is continuous, and quantifying a specific feature's drag invites the market to model a rolling series of one-offs. But this removes the only disclosure that let outsiders separate volume growth from price-equivalent deflation in a consumption model. The withdrawal happened in the same call as a guide that requires believing an unquantified second-half contribution, and the two together move the name toward requiring more trust and offering less evidence.

7. Iceberg Flips From Threat to Tailwind

The bear case against Snowflake for two years has been that open table formats let customers keep data in cheap object storage and query it with anything, dissolving the storage-and-compute bundle. Management now argues the format is expanding their addressable data rather than draining it, and brought a number to the argument.

"In terms of Iceberg, we're really now seeing Iceberg as more of a tailwind than a headwind as we -- it's opening up so much more data for us. We've yet to see massive amounts of data move out of Snowflake. Yes, there has been some, but storage still remains at roughly 11% of our revenue and we are seeing workloads that otherwise wouldn't have been accessible to Snowflake."
— Mike Scarpelli, CFO

The CEO framed it as a strategic claim rather than a defensive one, arguing that taking Snowflake to where data already lives is a larger opportunity than the risk of data leaving, and the product organization added the only quantitative-adjacent claim on the topic.

"And I'll add that I think on Iceberg, we've shifted from theoretical to actual consumption. Actual revenue coming already materializing to us."
— Christian Kleinerman, EVP of Product

Assessment: The 11% storage disclosure is the strongest single rebuttal available and it does real work: if storage is only a ninth of revenue, storage leakage cannot break the model. The compute question is unresolved, because a customer who keeps data in Iceberg can in principle bring a competing engine to it, and no one on the call asked whether Iceberg-resident workloads carry the same compute margin as native tables. Given that product gross margin is compressing, that is a question worth asking next quarter.

8. Microsoft, OpenAI, and Optionality Priced at Zero

Snowflake announced an expanded Microsoft partnership placing OpenAI models inside Cortex and making Cortex Agents available in Microsoft 365 Copilot and Teams. The CEO's framing was that the models run inside Snowflake's security perimeter, which is the commercially relevant part: enterprise customers get frontier models without data leaving the platform. Asked to size it, the CFO declined.

"Obviously, we wouldn't do it if we didn't think it could be impactful to our business, but it's too early to say. And I would say that would be upside in our revenue once we see traction with that, but it's going to take some time before we get this rolled out to customers."
— Mike Scarpelli, CFO

Assessment: Explicitly excluded from guidance, which is the correct treatment and also means the FY26 number does not depend on it. Being the one data platform hosting both Anthropic and OpenAI models natively is a genuine differentiator in enterprise procurement. It is also the third consecutive quarter in which the most exciting product news carries no revenue attached, and at some point the AI narrative has to appear in the disclosure table rather than the customer-count table.

9. Dilution: $1.9B of Buybacks and a Larger Share Count

Snowflake spent $1.9B in FY25 repurchasing 14.8M shares at a weighted-average price of $130.87, and made no repurchases in Q4. Weighted-average basic shares still rose, to 332.7M from 328.0M. Stock-based compensation was $1,479.3M on a cash-flow basis, 40.8% of revenue. Management guided the ratio down.

"Our strong revenue growth, combined with a more thoughtful approach to hiring, and increased leverage from AI, will benefit stock-based compensation as a percent of revenue. In FY ‘26, we expect SBC as a percent of revenue to decrease to approximately 37% from 41% and will continue to decrease year-on-year."
— Mike Scarpelli, CFO

Non-GAAP diluted share count is guided to 374M for FY26, including roughly 13M shares from the if-converted treatment of the zero-coupon notes.

Assessment: $1.9B of shareholder capital bought the absence of dilution rather than accretion. The Q4 pause is defensible on its own terms, since the shares traded well above the $130.87 average paid earlier in the year, and declining to buy at a higher price is discipline rather than neglect. The problem is what the full-year arithmetic shows: the largest buyback in the company's history, executed at good prices, still left the weighted-average share count higher than it started. A 380bp reduction in the SBC ratio is meaningful if delivered, but 37% of revenue in equity compensation still means the GAAP and non-GAAP versions of this company are different businesses.

10. GAAP Economics Went the Wrong Way All Year

FY25 GAAP operating loss widened to $(1,456.0)M from $(1,094.8)M, GAAP net loss attributable widened to $(1,285.6)M from $(836.1)M, and GAAP loss per share went to $(3.86) from $(2.55). Stockholders' equity fell 42.1% to $2,999.9M from $5,180.3M, a function of the buyback and the accumulated loss. Against that, the company ended the year with $5.29B of cash and investments and generated $941.5M of adjusted free cash flow.

Assessment: The cash generation is genuine and the GAAP loss is dominated by non-cash equity compensation, so the two coexist without contradiction. What deserves attention is direction: on the metrics that include the full cost of compensating employees, FY25 was worse than FY24 in absolute dollars and as a share of revenue. An investor paying 13.9x forward product revenue is buying the non-GAAP company, and needs a view on when the two converge.

11. CFO Retirement, Announced Without a Successor or a Date

The last item in the prepared remarks was the CFO's retirement, disclosed in a separate Item 5.02 filed the same day as the print.

"Before opening up the line for questions, you've likely seen our filing sharing that I plan to retire once the successor is in place and up to speed."
— Mike Scarpelli, CFO

The filing sets no effective date, names no successor, and attaches a 12-month post-departure consulting agreement under which outstanding equity awards continue to vest. On the call the tone was managed carefully, with the CFO twice noting he expects to be there a while longer and the CEO delivering a tribute before the first question.

Assessment: The structure signals an orderly, self-initiated exit rather than a rupture, and the open-ended timing removes the cliff risk. The concern is not integrity, it is continuity of the guidance regime. Snowflake's forecasting credibility is substantially personal to this CFO, who set the conservative-guide-and-beat convention the Street now prices in, and it is being handed over in the year the company is asking for the most benefit of the doubt it has requested in some time. A successor with a different calibration is a real, unmodellable variable in the FY26 setup.

Guidance & Outlook

Q4 FY25 Delivered Against Guidance

New Guidance

Snowflake does not guide EPS or total revenue. The FY25 SBC comparative is the cash-flow-statement figure net of capitalized amounts, which is the basis management referenced. Diluted share guidance includes approximately 13 million shares from the if-converted treatment of the zero-coupon notes.

Implied quarter-over-quarter ramp. The FY26 product revenue guide of $4,280M less the Q1 midpoint of $957.5M leaves $3,322.5M across Q2 through Q4. The same three quarters in FY25 produced $2,672.9M. The implied growth rate for the back three quarters is therefore about 24.3%, against 21.3% guided for Q1. Netting out the leap-day effect of roughly 110bp, the guide still embeds about 200bp of underlying acceleration after Q1, concentrated by management's own description in the second half. Q1 also carries roughly $15M of sales-kickoff expense, which explains most of the gap between the 5% Q1 operating margin guide and the 8% full-year figure.

Street at. The FY26 product revenue guide came in above the sell-side average of approximately $4.21B, and the Q1 guide midpoint sat above the roughly $949M consensus. Both are modest beats in percentage terms, about 170bp and 90bp respectively, which is consistent with Snowflake's practice of setting an initial full-year number it expects to raise rather than one it expects to defend.

Guidance style. Conservative by design and explicitly so. Asked how conservative the assumptions were, management declined to characterize them and instead reset what a beat should look like at this revenue scale.

"And given the scale we're at right now, we think a 3% beat is a — should be considered a big beat. And I feel good about the guidance that we've set for the quarter and the year."
— Mike Scarpelli, CFO

That is a useful anchor and worth holding management to. Q4's product revenue came in 3.5% above the guidance ceiling, which by the company's own definition is a big beat. It also means the FY26 initial guide should be read as a number designed to be exceeded by low single digits, not as a central estimate.

Analyst Q&A Highlights

Whether Large Customers Consuming Off-Contract Signals a Problem

The first question of the call went straight to the RPO disclosure management had volunteered minutes earlier, asking how unusual the behavior was and whether those accounts would re-contract. The answer laid out the mechanics of Snowflake's capacity contracts in more detail than the prepared remarks had, and drew a clear line between a timing effect and a demand effect.

Q: "On your comment on these large customers in Q4 exhausting their commitments, but not necessarily renewing going on-demand, how usual or unusual is that? And do you expect these customers just to sign a commitment contract in due course?"
— Sanjit Singh (for Keith Weiss), Morgan Stanley

A: "Yeah. I fully expect they're going to sign a new commitment. But as you know, we typically sign a three-year contract with our customers once they're up and running. Many new customers start at one year, the first year, then go to three years. And under their contract, they make a capacity commitment to us based on what they expect they're going to go through. But if they burn through that commitment before the end of the contract term, the customer has two options, they can do an early renewal and make a new capacity purchase. Typically, they only do that if they're going to get some economics for that, that makes sense to them or they continue to purchase under the same terms as they go to the end of their contract period, at which time they must do a new capacity purchase of equal to or greater than to get the same economics that they have in their contract. So this has always happened. We just had a few big customers that that happened to in this quarter, and I fully expect sometime over the next month to six months that those people will sign new contracts."
— Mike Scarpelli, CFO

Assessment: A complete and specific answer that puts a testable timeline on the claim, which is unusual and useful. Management said one to six months, so this is gradeable by the Q2 FY26 call. The response also volunteered the incentive structure, namely that customers only re-up early when they get better economics, which quietly explains why the behavior clusters in a quarter when growth outran plan. The residual risk is that customers running to contract expiry at existing terms give Snowflake less pricing leverage at the eventual renewal than an early re-up would have.

What Underwrites the Second-Half Acceleration

The most consequential question of the call named the anomaly directly: the guide implies the first growth acceleration in three and a half years, and it arrives in the second half of a year that has not started. The answer distributed the confidence across the core business, new products entering production, and new-logo cohorts maturing, and added visibility into customer migration plans as the underlying evidence.

Q: "But I did want to ask you a question around product growth. The guide implies an acceleration for the first time into the second half in 3.5 years. What's driving the optimism on acceleration? Is it all new products or are you feeling better about consumption activity across the base?"
— Brent Bracelin, Piper Sandler

A: "It's both. The core of our business is very strong, coupled with the new products that we see that will be going into production. Also on top of, I'm pleased with the muscle we've been building on new customer acquisitions and that's going to start to kick-in, in the second half of the year. We do get a lot of visibility into our existing customers with planned migrations that they have. And I think that is looking very good for the second half of the year."
— Mike Scarpelli, CFO

Assessment: The answer is honest and thin at the same time. "Visibility into planned migrations" is the strongest claim in it and the only one grounded in something management can actually see, but it was offered without a pipeline number, a workload count, or a conversion assumption. Combined with the same call's decision to stop breaking out feature-level assumptions, the market is being asked to take the shape of the year on management's word. Snowflake's track record earns some of that. Not all of it.

How Conservative the FY26 Assumptions Are

A direct attempt to calibrate the guide against the prior year's initial guide, which had been set during a CEO transition and proved beatable, drew a deflection followed by the most quotable framing of the call.

Q: "And then, Mike, for you on guidance. How conservative are the assumptions this year in the guidance versus kind of what you, for example, did last year when you had the CEO change? Obviously, how did the economy also play into the guidance for this year?"
— Raimo Lenschow, Barclays

A: "And then, Raimo, on your question of guidance, I'm just going to say the guidance I feel is appropriate. We've always tried to give meaningful guidance. And given the scale we're at right now, we think a 3% beat is a — should be considered a big beat. And I feel good about the guidance that we've set for the quarter and the year."
— Mike Scarpelli, CFO

Assessment: The macro half of the question went entirely unanswered, which is notable in a February 2025 call. On the conservatism half, resetting expectations so that a 3% beat counts as large is a deliberate move to compress the beat distribution the market prices in, and it is being made in the same quarter the company delivered a 3.5% beat. Read as a signal, this is management telling the Street to stop modeling 5% upside to the guide.

Iceberg's Direction of Travel and Where Net Revenue Retention Settles

A two-part question tried to pin down both of the structural bear arguments at once, asking whether the open-format shift is now a tailwind and where the retention rate stabilizes. The answer put a verbal floor under retention and produced the storage disclosure that is the most effective rebuttal available on the format question, with the CEO and the product head each adding a layer.

Q: "How do we think about Iceberg as a tailwind or a headwind in that guide? How do we think about NRR stabilization at 126% in the context of that guide? Any finer points you'd put on it?"
— Alex Zukin, Wolfe Research

A: "Yeah. First of all, I think NRR is pretty stable. I'm not saying it's not going to go to 125% or 124%, but I don't really see it. It's going to be in the mid 120s is what we are predicting. In terms of Iceberg, we're really now seeing Iceberg as more of a tailwind than a headwind as we -- it's opening up so much more data for us. We've yet to see massive amounts of data move out of Snowflake. Yes, there has been some, but storage still remains at roughly 11% of our revenue and we are seeing workloads that otherwise wouldn't have been accessible to Snowflake."
— Mike Scarpelli, CFO

Assessment: The most informative exchange of the call and the one that most improves the bull case. Two of the three standing bear arguments got a specific, falsifiable answer in a single response. What did not get addressed, and what nobody followed up on, is whether compute run against externally-held Iceberg tables earns the same gross margin as compute on native tables. Given that product gross margin is compressing and guided lower, the unasked question is the one that matters most for the next four quarters.

Whether the Microsoft and OpenAI Arrangement Can Be Sized

The final question of the call asked management to put a revenue figure on the expanded Microsoft partnership. The refusal was explicit and the framing placed the opportunity entirely outside guidance.

Q: "You had an expansion with your partnership with Microsoft and relates to OpenAI. I'd love to just get a little more granular about what that means for you guys at Snowflake. And, Mike, if you can potentially quantify what that could mean from a revenue standpoint in the future?"
— Joel Fishbein, Truist Securities

A: "And I'll just say, Joel, from a -- what is that impact going to be on our business. Obviously, we wouldn't do it if we didn't think it could be impactful to our business, but it's too early to say. And I would say that would be upside in our revenue once we see traction with that, but it's going to take some time before we get this rolled out to customers."
— Mike Scarpelli, CFO

Assessment: The correct answer, and it does useful work for the guide by establishing that the FY26 number does not lean on the partnership. It also completes a pattern: the three most prominent product announcements of the quarter, Cortex Agents, the OpenAI integration, and the Datavolo-derived connectors, are all discussed as capability and none as revenue. That is defensible at this stage of an AI product cycle and it is also why we treat the AI narrative as optionality rather than as an earnings driver.

Where the Center of Gravity for Enterprise Data Sits

A question about the wave of partnerships between data platforms and application vendors, and specifically a competitor's arrangement with a major ERP vendor, drew a long strategic answer built around customer choice and bidirectional integration, followed by a notably careful non-announcement about the vendor in question.

Q: "Over the last six to 12 months, we've seen a lot of new announcements from data companies sort of partnering up with enterprise software vendors like ServiceNow or Salesforce and you guys have partnered with them. One of your competitors just partnered with SAP. I guess what's the best way to interpret these announcements in terms of where the center of gravity from a data perspective lives?"
— Kirk Materne, Evercore ISI

A: "So, as Sridhar said, we've always done bidirectional type of sharing and data movement with a number of partners. While we don't have a specific announcement to share at SAP today, I can tell you that we are working with SAP. We like what they're doing with the business data cloud product and we have a common commitment to foster an open data environment and hope to share more with all of you soon."
— Christian Kleinerman, EVP of Product

Assessment: A pre-announcement dressed as a non-answer, and the closest thing to news in the Q&A. The question underneath it is real: if application vendors succeed in keeping their own operational data inside their own analytic layers, the horizontal data platform's share of enterprise analytic spend narrows. Management's answer is that customers want choice and Snowflake sits in the middle of it, which is the right answer competitively and does not address the share question at all.

Changes to the Sales Compensation Model

A brief exchange on sales incentives produced a structural change that got no follow-up and deserved one: reps will now carry a bookings quota alongside the consumption-revenue quota that has driven their pay.

Q: "And then, Mike, really quickly for you. Just any adjustments to the sales comp model that we should be aware of this year?"
— Kirk Materne, Evercore ISI

A: "Yeah. On the comp model, probably the biggest change is taking a little bit of a step back in terms of allocation. We're still principally paying people on their variable comp on revenue, but we do think it's important that reps also have a total contract bookings number as well too. And so there's going to be a quota as well on bookings, but the bulk of their earnings will come from consumption revenue."
— Mike Scarpelli, CFO

Assessment: This is the most concrete forward-looking disclosure in the entire Q&A and it went unremarked. Adding a bookings quota in the same quarter that RPO growth halved is not a coincidence; it is management installing an incentive to fix the contractual-coverage problem they had explained away thirty minutes earlier. It should mechanically support RPO growth in FY26, and it introduces a second-order risk that reps chase commitments customers do not consume, which would eventually show up in the retention rate rather than in bookings.

What They're NOT Saying

1. Any revenue figure for AI. Snowpark was quantified at 3% of FY25 product revenue. Cortex, Cortex Search, Cortex Analyst, Cortex Agents and Snowflake Intelligence were quantified at nothing. The AI disclosure is a customer count, more than 4,000 using AI or ML weekly, with no revenue, no consumption, and no attach rate behind it. A company whose gross margin is compressing because of AI compute costs is not telling us what AI revenue is.
2. Who the next CFO is, or when. No successor, no effective date, no interim arrangement, and no stated criteria. The search was announced in the same breath as the retirement.
3. The size of the performance-improvement headwind. Management explicitly retired this disclosure. In a consumption model, efficiency improvements are functionally a price cut, and this was the only line item that let outsiders separate volume growth from credit deflation.
4. Why Forbes Global 2000 customers fell sequentially. The count went from 754 to 745. It is almost certainly a list re-basing, and it was not mentioned by management or raised by a single analyst.
5. Why no shares were repurchased in Q4, and what the price discipline implies. The pause was disclosed in one clause with no rationale, and no analyst asked. The likely answer is that the shares traded above the $130.87 average paid earlier in the year, which would be a rational reason to wait. If so, it is worth management saying out loud, because it tells the market what price the company itself considers value with $2B of authorization outstanding.
6. Any path or timeline to GAAP profitability. Not mentioned in the prepared remarks, not asked in Q&A. The FY25 GAAP operating loss widened by $361.2M.
7. What proportion of the second-half acceleration comes from each named source. Three drivers were named. None was sized, and the refusal to size feature contribution was made policy on the same call.
8. Whether Iceberg-resident compute carries the same margin as native compute. The storage-leakage question was answered thoroughly. The compute-economics question, which is the one that connects Iceberg to the gross margin guide, was never put and never volunteered.
9. Anything about the macro environment. A direct question on how the economy factored into the FY26 guide was answered entirely on the conservatism half and not at all on the macro half.

Market Reaction

• Pre-print setup: SNOW closed at $166.19 on February 26, up 7.6% year to date but down 29.0% over the trailing twelve months and down 5.3% over the trailing thirty days, against a 52-week closing range of $108.56 to $234.03. The stock entered the print 29% below its 52-week closing high with the S&P 500 up 1.3% year to date, so the name had already de-rated substantially into the release.
• After-hours move: shares gained approximately 11% in the extended session on February 26, trading around $181, with the move beginning on the product-revenue beat and extending on the above-consensus FY26 guide. The CFO retirement, disclosed in the same release, did not interrupt the rally.
• Next-day session: the stock opened at $187.01 on February 27, a 12.5% upside gap, traded a $173.59 to $187.95 range, and closed at $173.69, up 4.5% or $7.50 on the day. The close sat $0.10 above the session low.
• Volume: 20.6M shares against a 30-day average of 5.4M, roughly 3.8 times normal.
• Index context: the S&P 500 fell 1.6% on the same session, so SNOW outperformed the index by about 6 points while surrendering two thirds of its opening gain.

The direction was positive and the shape was not. A stock that gaps 12.5% higher and closes on its low has met a seller for every buyer the guide attracted, and the fade was too orderly and too complete to attribute to the 1.6% index decline alone. Roughly two thirds of the opening move was given back in a single session on nearly four times normal volume, which is distribution rather than digestion.

What the tape was pricing. The overnight move priced the two headline facts: a 3.5% beat against the product-revenue ceiling and an FY26 guide about 170bp above the sell-side average. Both are real and both are small. The daylight session appears to have priced the second layer, where the full-year table shows non-GAAP operating margin falling 180bp, gross margin guided down again, adjusted free cash flow margin guided down, and a growth rate that decelerates in Q1 before it is asked to accelerate in the back half. Those facts require reading past the press release's summary tables into the reconciliation, which is roughly the amount of work a market does between the after-hours print and the following afternoon.

The setup mattered. Down 29% over twelve months entering the print, this was not a crowded long defending a rich multiple, and a beat-and-raise into that positioning would normally hold its gains. That it did not is the more informative half of the reaction. The buyers who wanted the quarter got filled at the open, and the marginal seller took the print as an opportunity to exit at a better level rather than as a reason to re-rate.

Street Perspective

Debate: Is the Second-Half Acceleration Real?

Bull view: Snowflake has guided conservatively and beaten every quarter under this CFO, the initial full-year guide is designed to be raised, and the drivers behind the back-half step-up are identifiable rather than hypothetical: new products entering production, FY25 new-logo cohorts maturing into consumption, and visibility into contracted customer migration plans. The bookings quota added to sales compensation reinforces it.

Bear view: An acceleration that arrives after the guided quarter, sized by nobody, disclosed by nothing, in a year when the company simultaneously stopped quantifying feature-level assumptions, is the structure of a guide that gets trimmed rather than raised. Net revenue retention has fallen in four of the last five prints and RPO growth just halved.

Our take: The bear framing is right about the evidentiary standard and wrong about the base rate. This management team has not missed, and the +24.3% implied for the back three quarters against +21.3% guided for Q1 is a 300bp step, of which 110bp is a leap-day artifact. The genuine ask is about 200bp. That is a modest number dressed up by the calendar as a dramatic one, and it is more likely to be met than not. The problem is that it will not be verifiable until the Q3 FY26 print in November.

Debate: Does the Gross Margin Guide Break the Operating Leverage Story?

Bull view: The compression is the cost of building an AI franchise on scarce GPUs, it is temporary, and management has said as much. Cost per unit of AI compute is falling industry-wide, and as AI features convert from adoption to monetization the mix turns accretive. Meanwhile the efficiency program is delivering 240bp of sales and marketing leverage per quarter and the SBC ratio is coming down 380bp.

Bear view: Two consecutive years of product gross margin compression, now guided for a third, is not a temporary phenomenon. It is the market telling Snowflake that AI workloads are worth less per dollar of cost than warehouse workloads. Every basis point of hard-won opex leverage is being consumed at the cost line, and FY26's promised 160bp of operating margin expansion only returns the company to where it was in FY24.

Our take: The bears have the better of this. The bull argument requires believing a cost curve Snowflake does not control and a pricing outcome it has not yet demonstrated, and the FY26 guide itself concedes the near-term point by taking product gross margin down another point. Until product gross margin stabilizes, this is a company converting operational excellence into a flat margin rather than a rising one, and that caps the multiple.

Debate: Is Iceberg Settled?

Bull view: The storage-leakage thesis is dead. Storage is 11% of revenue, so even severe leakage cannot move earnings, and the open format is a net expander of Snowflake's addressable data because it lets the engine reach data that never would have been loaded. Consumption on Iceberg tables is already materializing as revenue.

Bear view: Storage was never the risk. The risk is that once data sits in an open format outside Snowflake's control, the compute layer becomes contestable, and the customer can bring a cheaper engine to the same tables at renewal. The 11% figure answers the wrong question.

Our take: The bears are asking the right question and it went unasked on the call. That said, Snowflake's advantage was never storage lock-in, it was that the platform is easier and cheaper to operate than the alternatives, and open formats do not change the operational comparison. We treat Iceberg as neutralized on storage, unresolved on compute margin, and worth revisiting the moment product gross margin either stabilizes or takes another leg down.

Debate: Is 13.9x Forward Product Revenue Cheap?

Bull view: A 24%-growing platform with 76% gross margins, $941.5M of adjusted free cash flow, $5.29B of cash and investments, and a business inflecting into operating leverage is trading at a fraction of the multiple it commanded eighteen months ago. The de-rating has already happened; the fundamentals have not deteriorated.

Bear view: The multiple is on product revenue because there are no earnings to put a multiple on. GAAP operating loss widened to $1.46B, equity compensation runs at 40.8% of revenue, and $1.9B of buybacks left the share count higher than it started. On enterprise value to guided adjusted free cash flow the stock trades near 53 times, which is not cheap on any measure that includes the cost of the workforce.

Our take: Both are describing the same company at different levels of adjustment, and the gap between them is the investment question. We think the honest number is the free cash flow one, because adjusted free cash flow is the only figure here that is simultaneously positive, cash, and net of the buyback needed to hold dilution flat. At roughly 53 times, the stock is priced for the second-half acceleration to arrive and for margins to expand beyond FY24 levels. It is not priced for both to slip.

Model Update Needed

Valuation framework. At the $173.69 close and the guided 374M non-GAAP diluted share count, the equity is worth approximately $65.0B. Cash, cash equivalents and short- and long-term investments were $5,294.1M at year end; under the if-converted treatment already embedded in the share count, the zero-coupon notes are equity rather than debt, so enterprise value is approximately $59.7B. That is roughly 13.9x our FY26E product revenue of $4.33B and approximately 53x our FY26E adjusted free cash flow, the latter derived by applying the guided 25% margin to a total-revenue base we estimate near $4.5B on an unchanged product mix.

Fair value range: $160 to $195. The low end corresponds to about 12.7x EV to FY26E product revenue, which is where we think the stock finds support if the second-half acceleration slips a quarter or product gross margin takes another leg below 75%. The high end corresponds to about 15.8x, achievable if net revenue retention holds at 126%, product gross margin stabilizes at 75%, and the FY26 guide is raised at the Q2 print. The midpoint of $177.50 sits about 2% above the current price, which is the arithmetic behind the rating.

What would change the number. Upside: a Q2 FY26 guide raise of more than 2%, product gross margin printing at or above 75.5%, net revenue retention holding at 126% or better, a named CFO with a comparable guidance philosophy, or the first quantified disclosure of AI revenue. Downside: net revenue retention at 124% or below, product gross margin below 74.5%, RPO growth decelerating again in Q1, or an FY26 guide trimmed at any point in the year.

Thesis Scorecard Post-Earnings

This is our initiating view on Snowflake, so the table below establishes the pillars rather than grading a standing thesis. Each is written to be falsifiable against a disclosed metric so that subsequent quarters can be scored against the same list rather than a fresh one.

Overall: The operating business is performing and the two bear points that are already materializing are both about the quality of the growth rather than its existence. Snowflake grew product revenue 29.8% in FY25 and finished the year less profitable on every non-GAAP margin line than it started. That is the fact the initiation rests on. It is not a broken company and it is not yet a compounding one.

Action: Initiate at Hold. Build a position on weakness toward the low end of our $160 to $195 range, where the second-half risk is priced. Upgrade on two conditions met together: product gross margin stabilizing at or above 75% for two consecutive quarters, and a Q2 or Q3 FY26 guide raise that confirms the back-half acceleration rather than deferring it. Downgrade if net revenue retention breaks below 124% or the FY26 guide is trimmed.