Wall Street pays 23% for one percentage point of Salesforce growth

A stock that ends the week about 5% lower for the year after a 23% session was down roughly 23% before that session opened ^[2][1][1], so one Thursday retraced the entire year's drawdown on a quarter that added a single percentage point to the midpoint of a growth rate Salesforce itself describes as roughly where it has sat for four years ^[4][4]. The repricing therefore happened in the multiple rather than in the P&L, which is the honest way to read it: what changed was the probability the market had been assigning to an installed base of more than 150,000 corporate customers ^[11] rebuilding its own CRM, and a probability like that lives nowhere except the multiple.

Then there is the gain that flattered the bottom line. The mark on a three-year-old Anthropic stake is about 1.7 times the entire annualized revenue of Agentforce ^[2], the product the AI case actually rests on, and Salesforce says the figure could rise as Anthropic moves toward an IPO ^[7]. The mark is a receipt whose size is set by the private valuation of the same company whose model Salesforce is now wiring more deeply into Slack, a property it bought for almost $28bn ^[10]. Earnings quality and supplier valuation have been welded together, and they will move together on the way down as readily as up.

Attaching the house suffix to somebody else's product for the first time ^[9] is a capital allocation statement as much as a marketing one. Salesforce is renting frontier capability and paying in brand and distribution, with 37 pre-built sales skills as the visible shape of the rent ^[8], and Dario Amodei's claim that this is how all enterprise systems will run in future ^[13] is a description of a channel he did not have to build. Benioff's line that frontier models depend on CRM and do not replace it ^[12] is the same trade seen from the other side of the table.

The labour flow gets less attention than the headline numbers, but it may say just as much. Two senior people have come back from OpenAI after about five months each ^[15], and The Information reported on Wednesday, citing a person familiar, that another 22 former Salesforce employees there are in talks to return ^[16]; Salesforce called boomerang talent a celebrated part of its culture and declined to comment on the OpenAI arrivals, while OpenAI did not respond ^[17]. Enterprise go-to-market turning out to be harder than model quality is exactly what Benioff has been arguing since the 21% slide of last year ^[14].

This is probably wrong in one direction or the other, but the narrowing of the disruption discount looks deserved and the trigger for it looks thin. From here the bull and bear cases split. Agentforce compounds off a base of roughly $440m a year ago ^[3], attach rates carry the guide past its midpoint, and Thursday was early rather than generous; or growth off that base decelerates from 240% while the Anthropic mark stops rising, and a non-cash gain is revealed as the load-bearing item in a quarter that was sold as an operating one. The falsification is clean: two more quarters guiding low double digits with Agentforce growth halving would make this a price event with no value behind it, whatever William Blair's buy rating ^[18] says about the interim.