Snowflake bills by the second. No seat licenses, no flat monthly fee. Every query a customer runs moves through a metered credit system, and those credits accumulate into the product revenue figure that investors watch above everything else. That architecture is elegant when workload volumes grow. It becomes fragile when the underlying cost of running AI drops.
That tension is what makes Wednesday’s Q2 FY2027 report more than a routine earnings check. Snowflake will release results for the quarter ended July 31, 2026, after the close of U.S. markets on September 2. The Street is looking for roughly $0.45 in adjusted EPS and product revenue between $1.415 billion and $1.42 billion, which would represent 30% year-over-year growth. Those numbers matter. But the single figure that decides whether the AI data-cloud thesis is real is net revenue retention.
Why NRR Is the Only Number That Counts
NRR captures how much existing customers expand their spending versus the prior year. A 126% rate means the average customer is spending 26% more than they did 12 months ago, without counting any new logos. That is the critical mechanism in a consumption model: growth does not require selling to new accounts, it requires persuading existing accounts to run more.
NRR improved to 126% in Q1 FY2027 after multiple quarters at 125%, giving bulls their first clean signal that customer expansion may be stabilizing. The next test is durability: Q2 NRR must hold at 126% or rise again to prove this was a turn in customer expansion rather than a one-quarter bounce. Context matters here: Snowflake reported 158% NRR as recently as FY2023 and is now at 126%, a multi-year slide driven partly by customers getting better at optimizing their own workloads, which the consumption model translates directly into lower revenue per account.
Where Google Enters the Equation
The structural headwind is no longer just efficiency optimization. Google is actively cutting the price of AI, and the bigger point for Snowflake is that cheaper infrastructure compresses the perceived value of running inference and data processing inside a credit meter. At Google I/O 2026, the Ultra AI subscription price was reduced from $249.99 to $199.99.
For Snowflake, the relevant pressure is not consumer subscriptions. It is that cheaper compute on Google Cloud lowers the economic case for keeping AI inference and data processing inside Snowflake’s credit system. Google has also kept BigQuery’s on-demand query price at $6.25 per TiB scanned. When Google prices down to retain workloads, Snowflake’s consumption meter faces a harder comparison every renewal cycle.
The Bull Case Still Has Real Structure
Q1 was not a quarter to dismiss. Snowflake shares surged nearly 37% after the report after four things landed simultaneously: a record earnings beat, a full-year guidance raise, a $6 billion AWS infrastructure deal, and an announced acquisition. Product revenue rose 34% year-over-year to $1.33 billion, accelerating from 30% growth the prior quarter.
Cortex Code is now used across more than 7,100 accounts, while Snowflake Intelligence adoption more than doubled quarter over quarter, providing the AI thesis with measurable usage signals rather than management promises. Snowflake Cortex AI is built for the SQL user, exposing large language models, vector search, and document intelligence through SQL functions so analysts can run AI workflows without leaving a query editor. That ease-of-adoption is a real advantage over alternatives that require Spark expertise or GCP commitment.
What Investors Should Watch Wednesday
Three data points will decide the reaction. First, NRR: anything below 126% reopens the deceleration argument that plagued the stock through most of 2025. Second, Q3 product revenue guidance relative to the full-year $5.84 billion target. Snowflake raised its full-year product revenue guidance to $5.84 billion, representing 31% year-over-year growth, up from a prior forecast of $5.66 billion. Third, management commentary on Google’s pricing moves and whether enterprise customers are splitting AI workloads to arbitrage cloud costs.
The bear case is not that Snowflake loses customers. It is that customers stay but run less through the meter because compute got cheaper elsewhere. In a consumption model, a satisfied customer who optimizes their spend is indistinguishable from a churning one, right up until the NRR number lands.
Bottom Line
Wednesday’s report is less about whether Snowflake can beat a quarterly estimate and more about whether the consumption-billed AI data-cloud model holds its economics as Google systematically lowers the cost floor. One quarter of 126% NRR was enough to move the stock 37%. Two consecutive quarters would reframe the entire competitive discussion. One step back, and the Google threat becomes the dominant investor concern heading into calendar year-end.