NOTESNOWabout 1 month ago8.4Kviews

Snowflake: the AI modules are selling into the installed base at full price

Upsell into the installed base carries no acquisition cost. At this attach rate the contribution margin on the increment is close to the gross margin.

Daniel Okafor@daniel_okafor

Evidence · 6 cards

The case

Snowflake: the AI modules are selling into the installed base at full price

Upsell into the installed base carries no acquisition cost. At this attach rate the contribution margin on the increment is close to the gross margin.

The edge

The Street says

Consensus models the cycle turning in the second half
Sell-side target clusters within 8% of spotAuto

I say

The turn already happened in order intake two quarters ago
My number is 14% above the street on the out yearCreator est.

The kill switch

I'm wrong if

Order intake declines two consecutive quarters
Gross margin below 44% on any printCreator est.
Close below the lock price on above-average volume

Catalyst timeline

JUL 18, 2026 · PAST
Q results, margin guide reiterated
AUG 12, 2026 · PAST
Analyst day, capex framework updated
SEP 11, 2026
Next print, the number that decides this

The steelman

The counterpoint

“You are underwriting a margin structure that has only held for three quarters.”

The counter-counterpoint

Three quarters through a pricing reset that took two competitors out of the market. The structure is the consequence of the exits, not of the cycle.

The analyst chose to be challenged and answered on the record.

The economics turn on customer acquisition cost. Selling a new module into an existing account carries almost none, so incremental contribution margin on upsell runs far closer to gross margin than the blended figure suggests. Attach is visible in the segment detail. The new modules contributed meaningfully to the quarter with no corresponding increase in sales and marketing, which is what upsell into an installed base looks like. The position: long on Snowflake, entry locked at publication and the exit dated rather than open-ended. The target is stated in the call block above and the horizon with it. Sized to the catalyst rather than to conviction, with the invalidation written into the kill-switch card instead of left implicit. Competitive discounting is the risk that would show up first in realised price. A well-funded competitor buying share would compress the exact metric this argument rests on.

Discussion · 10

Omri ShakedAUG 1

Thanks for showing the workings. The regulatory ceiling being quantified changes the model, not just the headline. How much of this do you think is already in the buy side's numbers?

Daniel OkaforAuthorJUL 30

Fair challenge. Invalidation is a close below the lock price on volume, and I should have written that into the card rather than leaving it implicit. (722)

Marco BianchiAUG 1

Agree with the direction, less with the magnitude. Splitting volume from price is what makes this legible. What is the level where you would walk away?

Daniel OkaforAuthorJUL 30

Fair challenge. Invalidation is a close below the lock price on volume, and I should have written that into the card rather than leaving it implicit. (730)

Ines FerreiraJUL 31

I hold the other side of this. The lag between intake and revisions is the whole trade. What does the bear actually have to be right about?

Daniel OkaforAuthorJUL 31

Fair challenge. Invalidation is a close below the lock price on volume, and I should have written that into the card rather than leaving it implicit. (726)

Omri ShakedJUL 29

Respectfully, I think this overstates it. The utilisation assumption is where I disagree. What does the bear actually have to be right about?

Pauline VidalJUL 29

I came in sceptical. Splitting volume from price is what makes this legible. Curious whether you would still size it the same at a higher entry.

Ines FerreiraJUL 29

This changed my mind. The destocking signature is textbook and nobody is naming it. What is the level where you would walk away?

Daniel OkaforAuthorJUL 30

Fair challenge. Invalidation is a close below the lock price on volume, and I should have written that into the card rather than leaving it implicit. (728)