From Madhur Jain | Product & Market Analysis

Snowflake Q3 FY27 Preview: A $1.59B Guide and a Stale AI Run-Rate

On this page

Snowflake has not reported Q3 FY27 yet. Results land on 2 December 2026, and the only hard number today is a guide of $1,588 million to $1,593 million in product revenue, up 37% to 38%. The Snowflake Q3 FY27 earnings question is narrower than the headline: its last dollar AI run-rate, $100 million, dates from December 2025.

Key takeaways

  • Q3 FY27 is a guide, not a result. Snowflake guided product revenue to $1,588 million to $1,593 million for the quarter ended 31 October 2026 and reports on 2 December.
  • The last dollar AI figure is 11 months old. Snowflake's CEO cited a $100 million AI revenue run rate in December 2025, and the Q2 FY27 release and call give no updated dollar figure.
  • Consumption moved revenue before it moved retention. Product revenue growth rose 8 points across four reported quarters, from 29% to 37%, while net revenue retention rose 1 point, from 125% to 126%.
  • AI demand is arriving at a lower margin. Non-GAAP product gross margin fell from 76% to 74.7% in those quarters, and the 74.0% full-year guide implies about 73.2% for the second half.

The short answer

Yes, on the evidence so far, consumption pricing is passing AI demand into Snowflake's revenue faster than seat pricing passes it into Salesforce's or Workday's. The proof is growth acceleration, not a disclosed AI number. Snowflake has published no dollar AI run-rate since December 2025, so the direct share remains an inference.

$1.59BMidpoint of the Q3 FY27 product revenue guide, up 37% to 38%. Source: Snowflake Q2 FY27 release, September 2026.
126%Net revenue retention at 31 July 2026, up 1 point in four quarters while growth rose 8. Source: Snowflake earnings releases, 2026.
$100MLast dollar AI revenue run rate we could find, stated 3 December 2025. Source: Q3 FY26 call transcript.

What is known before Snowflake's Q3 FY27 earnings

Snowflake's third quarter of fiscal 2027 closed on 31 October 2026. The company will report after the US market close on 2 December. Any article quoting actual Q3 numbers today is wrong. This one uses the guide, the 10-Q for the prior quarter, and four reported quarters of disclosure.

The reader this is written for runs or funds a software business and is deciding how much of its pricing should meter. Snowflake is the cleanest large public test of that decision, because customers buy capacity and pay as they consume it.

The midpoint of the Q3 guide is $1,590.5 million. Against Q3 FY26 product revenue of $1,158.4 million, that is 37.3% growth. Sequentially, the guide adds about $98.6 million. Q2 added $157.6 million and Q1 added $107.7 million.

The same arithmetic applied to the full year is more revealing. Snowflake guides $6,070 million of FY27 product revenue. First-half product revenue was $2,826.2 million. Subtracting that and the Q3 midpoint leaves about $1,653.3 million for Q4, or roughly 34.8% growth on $1,226.6 million. Management is guiding deceleration again, as it has done before each of the last two beats.

Guides at Snowflake have been floors. Q1 FY27 product revenue beat the guide midpoint by $69.8 million and Q2 beat it by $74.4 million. In both quarters, a guide stated at 27% and 30% growth became 34% and 37%. CFO Brian Robins said on the Q2 call that the forecast rests on "observed consumption patterns", and that the guidance philosophy has not changed.

Snowflake quarterly ledger, Q3 FY26 to the Q3 FY27 guide
QuarterProduct revenueGrowthNRRNon-GAAP product marginRPO (growth)
Q3 FY26 (Oct 2025)$1,158.4M29%125%76%$7.88B (+37%)
Q4 FY26 (Jan 2026)$1,226.6M30%125%75%$9.77B (+42%)
Q1 FY27 (Apr 2026)$1,334.3M34%126%75.1%$9.21B (+38%)
Q2 FY27 (Jul 2026)$1,491.9M37%126%74.7%$9.00B (+30%)
Q3 FY27 guide (Oct 2026)$1,588M to $1,593M37% to 38%Not guidedNot guided for the quarter. FY27 guide is 74.0%Not guided

All figures come from Snowflake's quarterly earnings releases. NRR is net revenue retention. RPO is remaining performance obligations. The last row is guidance and contains no reported result.

The Snowflake AI run-rate: one dollar figure, 11 months old

On 3 December 2025, CEO Sridhar Ramaswamy told investors Snowflake had reached "$100 million in AI revenue run rate, achieved one quarter earlier than anticipated". On the same call he said AI was influencing 50% of bookings signed that quarter. "Influencing" is a soft word, and it should be read as one.

We then checked the three earnings releases since, the Q2 FY27 call in full, the readable part of the Q4 FY26 call, and the Q2 FY27 10-Q. None gives a newer dollar AI run-rate. Snowflake now reports AI by accounts instead. CoCo, the coding agent also called Cortex Code, surpassed 9,100 accounts, and CoWork reached 5,800.

I think that switch is the most important disclosure choice in the story. Accounts rise almost automatically once a feature ships to a base of more than 14,000 customers. Dollars do not.

Ramaswamy gave one quantitative clue on the Q2 FY27 call on 2 September. Growth had added 7 points in 2 quarters, from 30% to 37%, and of AI products he said: "They contributed approximately half of the acceleration that we are seeing." He named Cortex Code, CoWork, AI functions and the AI gateway. Asked about consumption uplift among AI adopters, he said the company was "not ready to share the exact" numbers.

That sentence supports one original calculation. Seven points on the year-earlier quarter's $1,090.5 million is $76.3 million. Half is about $38.2 million a quarter, or about $153 million annualised. That is the slice of AI growth that appears in the acceleration since January.

Adding it to the $100 million disclosed earlier gives about $250 million annualised, or about 4% of Q2's annualised product revenue of $5,967 million. Treat that as an order of magnitude and nothing more. It mixes bases and depends on one rounded remark, but it does show the scale: AI is half of the acceleration and only a small slice of the revenue.

Growth moved 8 points. Retention moved 1. Dark bars: product revenue growth, %. Light bars: net revenue retention less 100, points. Source: Snowflake earnings releases. 29% 25 30% 25 34% 26 37% 26 Q3 FY26Q4 FY26Q1 FY27Q2 FY27 Gap 4Gap 5Gap 8Gap 11
Read the bottom row. The gap widens every quarter. It is a Zan Digital subtraction, not a Snowflake metric, and the two measures use different time windows.

Snowflake net revenue retention moved 1 point while growth moved 8

Net revenue retention, or NRR, measures how much more an existing customer group spends over time. If AI demand were simply expanding existing customers in real time, NRR should have climbed with growth. It went from 125% to 126% while growth went from 29% to 37%.

That is not evidence against AI demand. It is mostly evidence about the definition, and it is why I would not use NRR to judge this quarter.

Why this NRR is slow by construction

Snowflake's measure compares a cohort's product revenue in year 2 with year 1 of a trailing two-year window. A full year of revenue goes into each side. A surge that began two quarters ago reaches only part of the numerator.

The cohort is also limited to customers already under capacity contracts at the start of the window. New logos never enter, and AI workloads that began this year sit diluted inside a 12-month sum. A measure this smooth is designed to be hard to move, which is the reason it is trusted.

What fills the gap between 26 and 37

Three things fill it, and Snowflake gives no split. New customers: the company added 692 net new customers in Q2, up 32%. Acquired revenue: Robins said the full-year guide includes approximately 1 percentage point from Observe. And recent expansion that NRR has not absorbed yet.

Comparing NRR across companies is covered in the 12-company net revenue retention comparison. The narrower point here is that one company's growth has run far ahead of its own retention metric, and that gap will close from one side or the other over the next few quarters.

Consumption pricing versus seat pricing: where AI demand shows up first

Consumption pricing charges for work done, so a new AI workload bills from the first query. Snowflake states it plainly: "we recognize revenue based on platform consumption, which is inherently variable at our customers' discretion". A seat vendor has to sell a new SKU, wait for a renewal, or reprice.

Seat vendors disclose more about AI and show less of it in growth

The irony is in the disclosure. Salesforce publishes Agentforce ARR above $1.5 billion, about 3.5% of annualised subscription and support revenue. That is close to our rough 4% estimate for Snowflake. Yet Salesforce grew 12%, and about 7% without Informatica on our arithmetic, while Snowflake grew 37%.

Similar AI share, very different revenue effect. That is the strongest evidence I can find that the billing model is doing real work. Salesforce's AI revenue largely substitutes for or sits beside seat revenue. Snowflake's AI revenue adds to a consumption line and appears to pull core consumption with it, as Ramaswamy's remark about the other half implies. The full basket test sits in the metered versus seat-priced growth comparison, and the Agentforce numbers are tracked in the Agentforce ARR disclosure ledger.

How AI demand reaches revenue under each pricing model, latest reported quarter
CompanyPricingLatest growthWhat it discloses on AI
SnowflakeConsumption, capacity contractsProduct revenue +37%Accounts: CoCo 9,100, CoWork 5,800. No dollar run-rate since December 2025
SalesforceSeats plus creditsSubscription and support +12%, includes $440M InformaticaAgentforce ARR above $1.5B, definition widened in Q2 FY27
WorkdaySeats, per employeeSubscription +13.9%AI above 25% of new ACV, more than 5,500 customers on an agent, no dollar figure
DatadogUsage plus commitmentsRevenue +35%, Q4 guide about 25%Largest customer reduced usage in Q3
Where seats winVisibilityWorkday 12-month backlog +14.2%, tracking revenueRevenue does not swing on one customer's workload

The Datadog warning

Metering passes demand through in both directions. Datadog grew 35% in Q3 2026 and guided Q4 revenue to $1.19 billion to $1.20 billion, about 25% growth. Its CFO said on the call that "our largest customer reduced their usage" and that the Q4 guide assumes that customer at its minimum commitment. A seat contract does not do that between renewals.

The Snowflake AI mix is already costing product gross margin

Consumption also passes through the bill. Non-GAAP product gross margin has fallen from 76% to 74.7% across four quarters. Robins attributed the 74.0% full-year guide to "a higher revenue mix from fast growing AI workloads which carry a lower contribution margin today". "Today" is doing a lot of work in that sentence, and the filings do not show when it stops being true.

The 10-Q states that third-party cloud infrastructure, including AI inference and GPUs, was about 75% of cost of product revenue, against 70% a year earlier. Applied to cost of product revenue of $434.4 million and $302.3 million, that is about $325.8 million against $211.6 million. The bill rose about 54% while product revenue rose 36.8%.

Put another way, each dollar of product revenue carried about 21.8 cents of cloud infrastructure, up from about 19.4 cents. The percentages are approximate in the filing, so the cents are approximate too. The direction is not.

A full-year margin of 74.0% on $6,070 million implies about $4,491.8 million of non-GAAP product gross profit. The first half delivered $2,116.8 million. That leaves about $2,375 million on about $3,243.8 million of second-half revenue, or about 73.2%. This is the number I would watch in the Q3 release, since Snowflake gives no margin guide for the quarter itself. For a method to apply to your own product, see how AI cost of goods sold changes real gross margin.

Non-GAAP product gross margin, and what the full-year guide implies Percent. Axis starts at 72%. Dashed segment is Zan Digital arithmetic on the 74.0% guide, not a Snowflake figure. 76%75%75.1%74.7%About 73.2% Q3 FY26Q4 FY26Q1 FY27Q2 FY27H2 FY27 implied
The last step is the steepest. It is inferred from a guide Snowflake has so far been cautious about. A beat on revenue mix could move it either way.

Snowflake backlog is shrinking while revenue accelerates

Remaining performance obligations, or RPO, is contracted revenue not yet recognised. Snowflake's fell from $9.77 billion in January to $9.21 billion in April and $9.00 billion in July, and its growth slowed from 42% to 30%. In a seat business, that pattern would be alarming. Here it shows customers consuming contracted capacity faster than they sign new commitments, at least in the first half of the fiscal year.

About 54% of RPO is expected to be recognised in the next 12 months, about $4.86 billion. That covers about 81% of Q2's annualised product revenue, which leaves roughly a fifth of current run-rate to be bought through renewals, expansions or on-demand use. Consumption customers can burn down a contract and sign the next one quickly.

Robins said "we expect bookings to be increasingly weighted towards the fourth quarter", so a soft Q3 RPO would not by itself be a bad sign. Ramaswamy also said new customer cohorts are reaching 80% of purchased capacity faster. For the same disclosure across vendors, see the cloud backlog comparison.

Backlog peaked in January and fell for two quarters Remaining performance obligations, $ billions, with year-over-year growth inside each bar. Source: Snowflake earnings releases. $7.88B+37% $9.77B+42% $9.21B+38% $9.00B+30% Oct 2025Jan 2026Apr 2026Jul 2026
Bookings at Snowflake have historically been heavy in the fiscal fourth quarter. That makes the January peak and later decline partly seasonal. The slowing growth rate is the part to watch.

Where this argument is weakest

Most of what follows should temper the conclusion. All of it is a reason to re-read this after 2 December.

There is no Q3 result in this post

Every Q3 figure here is guidance. Snowflake could report a very different mix, or disclose a new AI number that makes the "stale" framing obsolete. It could also release the figure without changing the conclusion. We would expect the post to be updated either way.

Our evidence on the silence is incomplete

We could not open the Q1 FY27 call transcript, and only part of the Q4 FY26 call was readable. A dollar AI figure could have been stated there. It is not in any of the three earnings releases since December, the Q2 FY27 call, or the Q2 FY27 10-Q text we could read.

One remark carries the AI estimate

The 4% figure depends on a rounded answer given on an earnings call, and on a run-rate that management has not updated. It also lumps in Observe's acquired growth, which Snowflake has not broken out for Q2. If you need a precise AI share, this post does not supply one. Nobody outside Snowflake can.

Snowflake is infrastructure, and that flatters the comparison

Comparing an infrastructure vendor with application software confounds pricing with what is being sold. Data platforms sit directly in the path of AI workloads, and migrations are also feeding the acceleration. A seat vendor with a better product could still lag. So the claim that the billing model is doing the work is my judgement, not a proven result.

How to read the Snowflake Q3 FY27 earnings release

Four lines will settle most of this. I would read them in the order below and ignore the adjusted free cash flow headline until last.

Four tests for the 2 December release
LineReference pointStrong readingWeak reading
Product revenue growthGuide 37% to 38%. Beats of $70M to $74M in Q1 and Q2 would imply about 43.5%Growth above 40%, showing the beat pattern intactInside the guide, a first break in that pattern
AI dollar disclosure$100M run-rate, December 2025A new dollar figure with a definitionOnly account counts again
Non-GAAP product gross margin74.7% in Q2, about 73.2% implied for H2At or above 74%, with mix absorbing better than guidedBelow 73%, with AI growth costing more than guided
RPO$9.00B, up 30%A return to sequential growth, up about 30% or betterFurther decline with growth below 25%

The 43.5% in the first row is a scenario, not a prediction. It adds the average of the last two dollar beats to the guide midpoint. Past beats do not guarantee the next one, and Datadog shows how fast a usage line can slow. For the competitor angle on these economics, read the Cortex and Databricks LLM cost comparison.

Frequently asked questions

When does Snowflake report Q3 FY27 earnings?

Snowflake will report results for its third quarter of fiscal 2027, the quarter ended 31 October 2026, after the US market close on 2 December 2026. The company announced the date on 2 November. Until then, no actual Q3 revenue, retention or margin figures exist. The most recent reported quarter is Q2 FY27, ended 31 July 2026.

What is Snowflake's Q3 FY27 product revenue guidance?

Snowflake guided Q3 FY27 product revenue to between $1,588 million and $1,593 million, representing 37% to 38% year-over-year growth, with a non-GAAP operating margin of 15.5%. Against Q3 FY26 product revenue of $1,158.4 million, the midpoint implies about 37.3% growth. In the prior two quarters, reported growth exceeded the guided rate by about 7 points.

What is Snowflake's AI revenue run rate?

The last dollar figure we could find is $100 million, stated by CEO Sridhar Ramaswamy on 3 December 2025. Later releases report AI by accounts, including 9,100 for CoCo and 5,800 for CoWork. A rough inference from later remarks suggests about $250 million annualised, or about 4% of product revenue, but that is an estimate and not a company figure.

Is Snowflake's net revenue retention rising?

Slightly. Net revenue retention was 125% in Q3 and Q4 of fiscal 2026 and 126% in Q1 and Q2 of fiscal 2027. Product revenue growth rose from 29% to 37% over the same quarters. The gap reflects a slow definition that compares a customer cohort across a trailing two-year window, plus revenue from new customers that the metric excludes.

Does consumption pricing capture AI demand faster than seat pricing?

At Snowflake, the evidence says yes. Growth accelerated 7 points in two quarters, and its CEO attributed about half to AI products. Salesforce and Workday disclose meaningful AI metrics but grew 12% and 13.9%. The comparison is imperfect because Snowflake sells infrastructure, and usage billing also lets customers cut spending quickly, as Datadog's largest customer did in Q3 2026.

Why is Snowflake's product gross margin falling?

Snowflake's CFO said the lower full-year guide of 74.0% reflects a higher revenue mix from AI workloads that carry a lower contribution margin today. Its 10-Q shows third-party cloud infrastructure, including AI inference and GPUs, rising to about 75% of cost of product revenue from about 70%. Non-GAAP product gross margin fell to 74.7% in Q2 FY27 from 76% in Q3 FY26.

Where to start before 2 December

If you sell software, take your largest AI feature and write down which billing event it creates. If the answer is "none until renewal", you have Salesforce's lag and not Snowflake's acceleration. That is worth knowing before your next pricing review.

Then split your AI gross margin from your core gross margin for one quarter, even roughly. Snowflake's 74.0% guide shows what a blended number hides. A metered AI line can raise your revenue growth and lower your margin in the same quarter, and your board will ask about both.

The one thing to check

When the release lands on 2 December, search it for "run rate" before you read anything else. Whether that phrase comes with a dollar figure will tell you how much Snowflake wants AI judged on revenue rather than accounts.

References

  1. Snowflake, Snowflake to announce financial results for the third quarter of fiscal 2027, 2 November 2026. Used for the reporting date. Retrieved 9 November 2026.
  2. Snowflake, Second quarter fiscal 2027 results, 2 September 2026, read in a full Business Wire republication. Used for Q2 actuals, Q3 and FY27 guidance, AI accounts and the NRR definition. Retrieved 9 November 2026.
  3. Snowflake, Form 10-Q for the quarter ended 31 July 2026. Used for cost of product revenue and the third-party infrastructure share. Retrieved 9 November 2026.
  4. The Motley Fool, Snowflake Q2 2027 earnings call transcript, call held 2 September 2026. Used for all Q2 call quotes. Retrieved 9 November 2026.
  5. StockAnalysis, Snowflake Q3 2026 earnings call transcript, call held 3 December 2025. Used for the $100 million run rate and the 50% bookings remark. Retrieved 9 November 2026.
  6. Snowflake earnings releases: Q3 FY26, Q4 FY26 on Form 8-K and Q1 FY27. Used for the quarterly ledger. Retrieved 9 November 2026.
  7. Seat vendors: Salesforce Q2 FY27 results, 26 August 2026, and Workday Q2 FY27 results, 27 August 2026. Retrieved 9 November 2026.
  8. Datadog, Q3 2026 earnings release and earnings call transcript, 4 November 2026. Used for Q4 guidance and the largest-customer remark. Retrieved 9 November 2026.

Last verified 9 November 2026. The weakest thing about this source base is that the Q3 FY27 result does not exist yet. The AI estimate also rests on one rounded spoken remark from a transcript service, not a filed figure. Update this post after Snowflake reports on 2 December 2026.

MJ
Madhur Jain
Founding Member, Zan Digital. Writes about AI product economics, B2B software markets and what the numbers behind vendor claims actually say.

Related reading