Snowflake财报拆解：AI定价、毛利取舍与RPO信号

Snowflake reported its fiscal Q2 2027 on September 2. And it crushed it:

• Product revenue was $1.49 billion, up 37% year over year, the third consecutive quarter of accelerating growth.
• The stock closed at $306 before the print and traded around $370 the next morning, within striking distance of the $401.89 all-time high it set back in November 2021.
• Fifty-two weeks ago it traded at $118.
• Snowflake also raised its full-year product revenue guide from $5.84 billion to $6.07 billion in a single quarter. A $230 million raise, and the implied full-year growth rate went from 31% to 36%.

Reaccelerating at a $6 billion run rate almost never happens. Snowflake’s product revenue growth went 26%, 32%, 29%, 30% across fiscal 2026, then 34%, and now 37%.

The five in brief:

• Sequential product revenue adds went from $68M to $108M to $158M in three quarters.
• About half the acceleration came from AI products, and the other half from those products pulling more core consumption behind them.
• Non-GAAP product gross margin fell from 76% to 75%, and the full-year guide went to 74%. Operating margin guidance went up on the same call.
• Total RPO grew 30% while current RPO grew 42%.
• 126% NRR, and only 49 net new customers crossed $1M in the quarter.

#1. Growth Went 30% to 34% to 37%, and the Sequential Dollar Add More Than Doubled in Two Quarters

The dollars are the better read at this scale:

• Q3 FY26: +$68 million sequential
• Q4 FY26: +$68 million sequential
• Q1 FY27: +$108 million sequential
• Q2 FY27: +$158 million sequential

Snowflake added more net new product revenue in the last three months than in any quarter in its history, and it has now done that two quarters running. Ramaswamy called it out on the call as a second consecutive record sequential dollar quarter. A percentage can accelerate on a weak year-ago comp. A record dollar add against your own best quarter is a harder thing to fake.

Two caveats. Roughly one point of the full-year growth guide comes from the Observe acquisition, so the organic guide is closer to 35%. And the guide implies deceleration from here: Q3 product revenue of $1.588 to $1.593 billion is a sequential add of about $99 million, and the implied Q4 is about $63 million, which puts Q4 growth back near 35%. Snowflake has beaten and raised every quarter this year, so treat the guide as conservative. But the company is not forecasting 37% to hold.

#2. About Half the Acceleration Came From AI Products, and the Other Half Came From AI Pulling Core Consumption With It

Ramaswamy was asked directly to separate the AI contribution from the core. His answer: roughly half the acceleration came from AI products (CoCo, CoWork, AI functions, the Cortex AI Gateway), and the other half came from those products dragging more core platform consumption behind them.

The adoption numbers:

• CoCo, the coding agent, surpassed 9,100 accounts, adding more than 2,000 in the quarter alone
• CoWork, the knowledge worker agent formerly called Snowflake Intelligence, reached 5,800 accounts
• Total customer count is 14,554

Roughly 60% of the account base is touching the coding agent. Snowflake counts these as the average of the last four weeks of the quarter across capacity and on-demand accounts, so the denominator is not perfectly clean, but the attach rate is real and it is fast.

The pricing decision underneath it: AI went on the same consumption meter customers already have, rather than into a separate product line with its own SKU, its own sales motion and its own quota. Every AI query runs on data already stored and governed in Snowflake, so the AI dollar and the core dollar arrive together. That is why the acceleration shows up in one revenue line instead of a new one.

Ramaswamy also said the cost-management skill inside CoCo ranks in the top ten skills customers use. Snowflake shipped an agent that helps its own customers spend less on Snowflake, and consumption accelerated 7 points anyway. He framed it as a lesson learned the hard way during the pandemic optimization cycle: pushing customers toward efficient spend builds enough trust that they bring you more projects.

And he was clear that the acceleration is broad-based, with AI-native companies still a small part of revenue. The buyers here are the Global 2000, not a handful of model labs running up a bill.

#3. They Gave Up Two Points of Gross Margin on Purpose and Raised Operating Margin Guidance Anyway

Non-GAAP product gross margin came in at 74.7%, down from 76.4% a year ago. Snowflake then guided the full year to 74.0%. CFO Brian Robins was direct about why: a higher revenue mix from fast-growing AI workloads that carry a lower contribution margin.

The list of companies reporting this is getting long. Figma’s gross margin came down roughly five points year over year on AI credits. Canva hit the same wall on its free tier. Box held its line at 81% because it sells the content and permissions layer and does not buy the inference. Snowflake buys the inference, so Snowflake pays for it.

Where Snowflake found the money is the interesting part. Non-GAAP operating margin came in at 15.3%, up more than 400 basis points year over year, and the full-year operating margin guide went from 13.5% to 14.5%:

Revenue grew 35% and total operating expenses grew 17%, at roughly 9,250 employees. Robins described it as offsetting growing cloud costs with slowing headcount expense.

Two points of gross margin for four points of operating margin and seven points of growth. The AI bill came out of headcount, and Snowflake told investors the gross margin line was going down before anyone asked.

#4. Total RPO Grew 30% While Current RPO Grew 42%

Remaining performance obligations were $9.00 billion, up 30%. That growth rate has gone the wrong direction all year: 42% at the end of fiscal 2026, 38% in Q1, 30% now. In absolute dollars RPO is down from $9.77 billion at year end.

Meanwhile 54% of that RPO is expected to be recognized in the next twelve months, and that current portion grew roughly 42% year over year. Deferred revenue fell from $3.35 billion to $2.57 billion over six months.

So the forward book decelerated while revenue accelerated, and the near-term book accelerated while the long-term book shrank as a share. In a consumption business both can be true at once. RPO measures when a customer signed a capacity contract. Revenue measures when they burned it. A 126% net revenue retention rate means customers are burning through committed capacity faster than the contract assumed, which pulls revenue forward and pushes the recommitment into a future quarter.

Salesforce reported the same shape in August: current RPO up 14% while noncurrent RPO grew 7.5%. Two very different companies, both with the acceleration living inside twelve months.

If you sell consumption, RPO will make you look worse in your best quarters and better in your worst ones. Snowflake handles that by reporting both numbers and walking through the 54% split on every call.

#5. 126% NRR, but Only 49 Net New Customers Crossed $1M

Net revenue retention was 126%, up from 125% at the end of the fiscal year. Most public B2B companies are reporting NRR in the low 100s right now. Box printed 106% last week and treated it as a win, correctly.

The customer metrics underneath it:

• 828 customers with trailing twelve-month product revenue over $1 million, up 27% year over year, and up 49 from 779 in Q1
• 692 net new customers in the quarter, up 32%
• 829 Forbes Global 2000 customers, 14 more than last quarter
• 14,554 total customers
• 43% of customers now share data with at least one stable edge

Product revenue grew 37%. The count of customers over $1 million grew 27%. Snowflake is growing revenue faster than it is minting seven-figure accounts, so the money is coming from accounts that were already large getting larger.