Snowflake buys $230m of guided revenue with two points of gross margin

The trade reads better at the gross profit line than at the margin line. Guiding $5.84bn at 76 per cent non-GAAP product gross margin is $4.44bn of gross profit; guiding $6.07bn at 74 per cent is $4.49bn, so the full exercise nets about $53m more gross profit ^[1]. Charge the entire two-point decline against the new money and each incremental dollar of guided revenue arrives carrying roughly 23 cents of gross profit ^[2]. That attribution is the harsh end of a range nobody has disclosed: because AI queries run on the same consumption meter the existing base already pays on ^[12], part of the margin decline sits on revenue that was coming anyway, and the real incremental margin on AI consumption is somewhere above 23 per cent and below 74 per cent.

What the two points bought is attach rate. CoCo, the coding agent, is past 9,100 accounts against 14,554 total customers, about 63 per cent of the base ^[11][8], though Snowflake computes those account figures as an average of the last four weeks of the quarter across capacity and on-demand accounts, so the denominator is not clean ^[16]. CoWork, the knowledge worker agent, is at 5,800 ^[11]. Ramaswamy told the call that the cost-management skill inside CoCo ranks in the top ten skills customers use ^[13]. Snowflake shipped an agent that helps customers spend less on Snowflake, and consumption growth went up 7 points anyway ^[13].

Ramaswamy's split of the acceleration, as reported by SaaStr, is roughly half from AI products and half from those products pulling more core consumption behind them ^[5]. Take him at his word, and the back half of the year still slows on his own arithmetic: the Q3 guide implies about $99m of sequential add and the implied Q4 about $63m ^[10], which is $162m of second-half adds against $266m in the first half, 61 per cent of the pace ^[4]. About one point of the full-year guide comes from the Observe acquisition, putting organic growth nearer 35 per cent ^[9]. Snowflake has beaten and raised every quarter this year ^[17], which is the case for reading that guide as a floor rather than a forecast.

The price question and the value question separate here. At a little over three times where the equity traded 52 weeks ago and within 8 per cent of the $401.89 high set in November 2021 ^[15][6][7], buyers are paying for $158m sequential adds as a run rate rather than a peak. My read, with the counter in the same breath: given 126 per cent NRR, which implies about 26 of the 37 growth points came from the installed base ^[10], and current RPO growing 42 per cent against total RPO's 30 per cent ^[7], giving up two points of margin to keep the AI budget on a meter you already own is the correct use of the money. The counter-thesis is that the half-from-AI figure is management attribution rather than a reported segment, and a consumption meter cannot tell an AI query from a Global 2000 migration query that would have arrived regardless ^[14]. What settles it is gross margin printing under the 74 per cent guide while the sequential add lands at the guided $99m, which is the shape of paying more and getting less.