Snowflake Cortex Lifts Product Revenue and Cuts Margin Guide

$1.33 billion in Snowflake Cortex AI-linked product revenue landed for the quarter ended April 30, up 34% from a year earlier. CEO Sridhar Ramaswamy called Cortex Code and Snowflake Intelligence a turning point for the AI Data Cloud. On Sept. 2 the company put product revenue for the quarter ended July 31 at $1.49 billion, up 37%, and Ramaswamy said first-party AI products contributed about half of that extra speed.

The second print is the bill coming due on data that was already in Snowflake. Full-year product gross-margin guidance is now 74.0%, and remaining performance obligations, the contracted revenue not yet recognized, stood at $9.00 billion on July 31 after $9.21 billion on April 30.

Product Revenue Jumped to $1.33 Billion, Then Accelerated

Snowflake’s May 27 release listed first-quarter product revenue of $1.33 billion, with total revenue of $1.39 billion, up 33%. Net revenue retention was 126%. The customer count was 13,912 after 616 net new accounts, up 38% year over year, including 13 Forbes Global 2000 names that took that roster to 813.

Large spenders moved faster than the logo count. Customers with more than $1 million in trailing 12-month product revenue reached 779, up 29%, after 46 accounts crossed that line in the quarter against 26 a year earlier. Remaining performance obligations were $9.21 billion, up 38%. CFO Brian Robins said AI was speeding the core data platform as those million-dollar accounts piled up, and the company raised full-year product-revenue guidance to $5.84 billion, or 31% growth, from $5.66 billion, or 27%.

That May outlook still assumed a cooler summer. Product revenue for the second quarter was guided at $1,415 million to $1,420 million, about 30% growth. The quarter ended July 31 did not cool.

Q1 VERSUS Q2 ON THE METER

The Sept. 2 results put second-quarter product revenue of $1.49 billion at 37% growth, with total revenue of $1.55 billion, up 35%. Ramaswamy said it was the second straight quarter of record sequential-dollar growth and a seven-point acceleration from the 30% rate at the end of fiscal 2026. Robins called it the third consecutive quarter of faster product-revenue growth, with the core platform and AI both contributing.

New logos did not vanish. Snowflake added 692 net new customers in the second quarter, up 32% year over year, taking the base to 14,554, with 14 new Global 2000 accounts and 829 in that group. Forty-eight more customers cleared $1 million in trailing product revenue, taking that cohort to 828, up 27%, and 65 accounts were above $10 million. The installed base still did most of the work. Retention held at 126%, and use cases on the platform rose 89% year over year.

CoCo Added More Than 2,000 Accounts in a Quarter

The May release said more than 13,600 accounts were using Snowflake AI features, against 13,912 customers, with Cortex Code already on over 7,100 accounts and Snowflake Intelligence more than doubling quarter over quarter. Those tools were later branded CoCo and CoWork. Snowflake said Cortex Code had the fastest adoption of any new product in its history, and Q2 showed why the meter kept spinning.

THE AI ACCOUNTS ON THE PLATFORM

• All AI features, Q1: More than 13,600 accounts used at least one Snowflake AI capability in the last four weeks of the April quarter.
• CoCo in Q2: The coding agent surpassed 9,100 accounts after adding more than 2,000 in the quarter.
• CoWork in Q2: The business agent reached 5,800 accounts.
• The split Ramaswamy gave: AI products contributed about half of the recent growth acceleration, with core products and faster migrations supplying the rest.

CoCo is the engineer-facing agent. It reads Snowflake schemas, metadata, account context and docs, then writes transformations, debugs pipelines and stands up agents against that warehouse. CoWork is the business-facing agent. A manager can ask which campaigns drove pipeline in EMEA and get an answer without waiting on a dashboard, because the query still runs inside the customer’s warehouse under existing roles.

Those are different meters. A coding agent will keep spreading among data teams, which is a smaller group than the people who live in chat. CoWork, and the Teams bot that reuses the same agents, is the wider surface. Counting 9,100 CoCo accounts and 5,800 CoWork accounts as one “AI win” hides that split. Overlap is likely, and Snowflake does not add the two together.

Ramaswamy told analysts the extra speed was broad, not concentrated in a handful of AI-native firms, which he said remain a small share of revenue. He also said CoCo makes it easier to kill waste, because a team can point the agent at a slow query or the most idle warehouses. Consumption can rise while customers get better at trimming it. That is the tension inside the 126% retention rate.

They contributed approximately half of the acceleration that we are seeing.

Sridhar Ramaswamy, CEO, Snowflake Q2 FY2027 earnings call

He put CoCo and CoWork in that half, along with AI Functions and Cortex AI Gateway. Notebooks, Streamlit and React apps, and faster migrations made up the rest. The June story that “Cortex demand” lifted Q1 was directionally right. The mechanism is older than the branding: queries, warehouses and now agents, billed as product revenue.

AI Mix Cut the Gross-Margin Guide to 74%

Q1 GAAP product gross margin was 71.0% on $947.5 million of product gross profit. The May outlook still aimed at 75.0% non-GAAP product gross margin for the year and lifted non-GAAP operating-margin guidance to 13.5% from 12.5%. Q2 non-GAAP product gross margin came in at 74.7%, and GAAP product gross margin was 70.9%.

The full-year non-GAAP product gross-margin guide is now 74.0%. Management tied the cut to a higher mix of fast-growing AI workloads that carry a lower contribution margin today. Operating leverage still improved. Ramaswamy said non-GAAP operating margin expanded by 400 basis points year over year to about 15%, and the full-year operating-margin guide moved to 14.5% from 13.5%.

GAAP results remain a loss. The April quarter showed a $326.2 million operating loss. The July quarter showed a $263 million operating loss and a $191.7 million net loss, with $4.3 billion in cash, equivalents and investments. Stock-based pay is still large relative to sales, even as the non-GAAP margin widens. The 37% product-revenue rate and the 74.0% margin guide are the same mix story, read from two lines.

Robins has been explicit that AI revenue is now material, not a feature footnote. He also said the company is trying to raise growth without losing cost control. The margin guide is where that trade becomes visible. If CoCo and CoWork keep adding accounts at the Q2 pace, the mix that juiced growth will keep leaning on the lower-margin AI slice unless Snowflake raises prices, shifts models, or gets more efficient inference.

Remaining Obligations Slipped While Revenue Ran Hotter

Booked-but-unrecognized revenue did not keep up with the consumption spike. Remaining performance obligations were $9.21 billion at April 30, up 38% year over year. They were $9.00 billion at July 31, up 30%. Snowflake said about 54% of the July backlog should be recognized over the following 12 months.

A sequential dip from $9.21 billion to $9.00 billion, while product revenue ran at $1.49 billion, is the objection the June 34% story never had to answer. Consumption can outrun new contracts for a quarter, especially if bookings bunch in Q4, which management has flagged. It can also mean customers are burning existing commits on CoCo, CoWork and warehouse jobs faster than they are signing larger follow-ons.

THREE QUARTERS OF FASTER PRODUCT GROWTH

1. January 31, 2026: Fiscal 2026 closes with fourth-quarter product revenue growth at 30%.
2. April 30, 2026: Q1 product revenue reaches $1.33 billion, up 34%, after more than 13,600 accounts touch AI features.
3. May 27, 2026: Snowflake raises full-year product guidance to $5.84 billion, or 31%, and guides Q2 product revenue to $1,415 million to $1,420 million.
4. July 31, 2026: Q2 product revenue reaches $1.49 billion, up 37%, as CoCo crosses 9,100 accounts.
5. September 2, 2026: Full-year product guidance rises again to $6.07 billion, or 36%, including about 1 percentage point from Observe, and Q3 product revenue is set at $1.588 billion to $1.593 billion, or 37% to 38%.

The $6.07 billion outlook is a $230 million raise from the May figure. Observe, the observability deal, is called out as about a point of that growth, so the rest is the core platform plus AI. Snowflake also said it took Observe and Natoma Labs through close, folding Natoma’s model-context work into Cortex AI Gateway. Gateway is the control layer for first-party agents and for outside agents that need governed access to Snowflake data.

Q2 marks our third consecutive quarter of product revenue growth acceleration, driven by strength in both our core data platform and a meaningful step-up in AI revenue.

Brian Robins, CFO, Snowflake Sept. 2 results

Ramaswamy described three loops: new AI workloads landing on Snowflake, CoCo and CoWork adoption, and higher platform consumption among customers that turn those agents on. Holiday Inn Club Vacations and Houzz showed up as Q1 wins. 1Password and Indeed showed up in Q2. Snowflake said Sayari is using CoCo to migrate 12 billion records and cut costs by more than half, and it cited an Australian bank moving a financial-crime platform that processes 17 billion transactions with queries running 10 times faster.

Cortex Agents Run Inside Teams and Copilot

The Microsoft path is how a lot of those business questions will be asked. Snowflake’s documentation for Cortex Agents for Microsoft Teams and Microsoft 365 Copilot puts the same agents a customer builds for CoWork into a Teams bot and into Copilot. Prompts go to the Cortex Agents API. The SQL still runs in the customer’s virtual warehouse. Underlying tables do not leave Snowflake’s boundary, and role-based access still applies.

Deployment is the listing on Microsoft AppSource. An Azure Global Administrator has to grant tenant-wide Entra ID consent for two Snowflake-hosted apps, a resource principal and a client principal. A Snowflake ACCOUNTADMIN or SECURITYADMIN then creates a security integration so Entra is an external OAuth provider, and only then can accounts be linked to the bot. End users need their own Microsoft and Snowflake identities. A Copilot license is required if the Copilot agent is in scope.

WHAT THE TEAMS BOT STILL WILL NOT DO

• Region consent: Accounts outside Azure US East 2 must accept that prompts and bot replies are processed, not stored, in Snowflake’s Azure East US 2 region.
• Private Link: Private Link configurations are unsupported, so that feature has to be off.
• Sovereign clouds: The integration is unavailable for Snowflake accounts in sovereign regions.
• Admin load: Setup needs Global Administrator rights on the Microsoft tenant and ACCOUNTADMIN or SECURITYADMIN on each Snowflake account that will be linked.

Those limits matter more than the AppSource badge. A bank that insisted on Private Link cannot put Cortex in Teams without changing its network design. A public-sector tenant on a sovereign cloud is out. Everyone else is sending prompts through Azure East US 2 even when the warehouse lives on AWS or Google Cloud. Microsoft is in the path of the conversational product, including for customers who never bought Azure as their Snowflake host.

Those agents can also be reused through the Cortex path into Microsoft 365 in Copilot and Teams, so a CoWork agent does not have to be rebuilt for chat. That is the idiotproof surface next to CoCo. Engineers will keep driving warehouse consumption with a coding agent. Everyone else will ask questions where they already work, and those questions still bill as Snowflake product revenue.

OpenAI Models and a New Preview Land on Cortex

Model choice is how Snowflake keeps that meter from depending on one vendor. A June 4 product post said OpenAI models inside Cortex AI are generally available, after a $200 million partnership announced earlier in the year, for customers on Amazon Web Services, Google Cloud or Microsoft Azure. Snowflake said it now manages the endpoint inside its own control plane rather than routing prompts through a separate cloud AI service, so the same role-based access, encryption and audit trail cover the data and the model call.

Q1 also brought a $6 billion multi-year AWS infrastructure agreement, the largest Snowflake has signed with that cloud, plus general availability of joint SAP work. Anthropic’s Claude line is on the same Cortex menu. On Sept. 1 Snowflake put Claude Fable 5.1 into private preview for CoCo, CoWork, Cortex Agents, Cortex AI Functions and Cortex Inference, with same-day availability as a launch partner.

On Sept. 9 the company added another preview on that same list.

https://x.com/Snowflake/status/2097724533432033779

GPT-6 Astra is in private preview across CoCo, CoWork, Cortex Agents, Cortex AI Functions and Cortex Inference, with OpenAI’s latest reasoning and computer-use features inside the Snowflake perimeter. Ramaswamy had already argued, when Cortex AI Gateway picked up dynamic model routing, that more AI use is not the same as more impact, and that routing should send each task to a cheaper or better model under customer policy. Gateway is also the place Natoma’s connector fabric shows up, so outside agents can be allowed, logged and billed without a second data copy.

In the first half Snowflake took more than 330 product features to general availability, up 35% year over year, and added Cortex Sense for business context beside Gateway. The product machine is the other half of the 37% rate. Models change by the week. The warehouse, the roles and the bill stay put.

The Q3 product-revenue guide is $1.588 billion to $1.593 billion, or 37% to 38% growth. Full-year product revenue is now set at $6.07 billion, or 36%. That guide only holds if CoCo, CoWork and the Teams path keep turning stored data into jobs that customers are willing to run at a 74.0% product-margin mix.

Disclaimer: This article is news reporting and analysis of Snowflake’s public financial results, filings and product announcements, and it is for information only. It is not investment advice, a recommendation to buy or sell Snowflake or any other security, or a prediction of future revenue, margins or share price. Readers should consult a licensed financial adviser or securities professional before making any investment decision. Revenue figures, customer counts, remaining performance obligations, margin guides and product statuses reflect company statements as of the dates cited and can change in later quarters.