Snowflake books $370m less contracted revenue than analysts modelled

Remaining performance obligations are a paperwork number before they are a demand number, being the sum of what customers have signed and not yet burned through. Snowflake's own release says the measure "is not necessarily indicative of future product revenue growth because it does not account for the timing of customers' consumption", and names renewals, contract length and seasonality among the things that move it ^[16]. A customer that holds annual spend flat and signs for two years instead of three shrinks that figure without spending a dollar less.

An AI adoption slide typically assumes that buyers try the feature and then commit more money for longer. Buyers on a consumption meter instead renew and put an engineer on query cost. SiliconANGLE reports that some analysts see customers using large language models to trim their spend on platforms like Snowflake, and that Ramaswamy's answer was that efficient use is "a key ingredient for our future success" ^[5]. He may be right about the value, and the effect still lands in contracted backlog before it lands in revenue.

The gap is small in proportion. Subtract $9.00bn from $9.37bn and you get $370m, which is 3.9% below where the models sat ^[23]. Set against the raised full-year product guide of $6.07bn, the backlog still covers about 1.48 years of it ^[22]. Finance chief Brian Robins called this the third consecutive quarter in which product revenue growth accelerated, with margin expanding at the same time ^[6]. The pressure is comparative rather than absolute: product revenue, which excludes professional services, grew 37% ^[2], while Databricks, private and under no obligation to publish audited results, says it passed $7bn of revenue run-rate growing more than 80% ^[17].

CoCo at more than 9,100 accounts and CoWork at 5,800 are counted by averaging the last four weeks of the quarter ^[7], which measures how many accounts have the feature switched on. Neither number says credits consumed, or whether an account that enabled CoCo in the spring still runs it. The depth signals in the release are net revenue retention of 126% ^[8] and 828 customers spending more than $1m a year ^[9]. The cost signal is non-GAAP product gross margin of 74.7%, down from about 76% a year earlier ^[18].

One more note on the profit being quoted. Stock compensation-related charges of $456.4m are 1.93 times the non-GAAP operating income the beat rests on ^[19][24], and Snowflake lists reducing stock compensation as a share of revenue, and reaching GAAP profitability, among its own risk factors ^[20].

The takeaway applies to anyone with an AI feature and a renewal calendar. The two numbers that matter are accounts with the feature enabled and consumption per account among the accounts that already had it a quarter ago. High breadth with flat depth means pilot sprawl that renews at the old number. Flat breadth with rising depth means a working product with a distribution problem that renews higher inside a narrow set of accounts. Both rising is expansion, which reaches the contracted-backlog line only when buyers agree to longer terms. Snowflake published its breadth axis this week; depth is available only as a portfolio average.