Salesforce, Inc. CRM

Key data

The business

Salesforce rents software that companies use to track every customer they have: who a rep is calling, what a support agent is fixing, which email a marketer sent. About 95% of the money is subscription, $10.8B of the $11.3B booked last quarter, billed year after year. The newer pieces are Data Cloud, which pools a customer's data, and Agentforce, AI agents sold on top to answer tickets and chase leads. The moat is switching cost of the dullest, strongest kind: the sales floor lives inside the Salesforce tab all day, and pulling it out means retraining every rep and rebuilding every workflow that touches it. That is why the base does not leave, and why the whole question is price and pace, not survival. The weakness is that the same base is now big, mature, and growing at roughly the speed of the software market itself.

The numbers

The quarters read as steady, until the last one jumps.

Operating income in that last quarter was $2.33B, flat against the $2.35B a quarter earlier. Profit still leapt to $3.5B because $2.2B of gains landed below the operating line, and that, not the business, is what made $4.29 a record. Non-GAAP EPS printed $5.90 against a $3.27 estimate, the fourth straight quarter above the Street, but that beat carries the same one-timer. Read the operating line and this was an ordinary quarter dressed as a triumph.

Revenue compounded about 12% a year over the four years to FY2026, but the rate is fading: +18% in FY2023 down to about +10% now. The 1H figure shows +12%, a small reacceleration, and part of that is Informatica, freshly bought and consolidated, not customers spending more. Strip the deal and organic growth is closer to 9%. What has genuinely improved is cash: operating cash flow went from $6.0B in FY2022 to $15.0B in FY2026, and free cash flow margin now runs about 35%.

What I believe that the tape does not: the AI attach is a real, unpriced option that a decade-low sales multiple hands you for free. What would settle it is one number pair over the next two quarters, contracted future revenue growing faster than reported revenue, or not.

Management

Capital allocation is the whole story here, and the leverage-up flagged a week ago went through exactly as expected rather than breaking: long-term debt jumped from $14.4B last January to $39.3B by April, funding a single $27.2B accelerated buyback that cut the diluted share count about 15% year over year, alongside the closed Informatica deal. That is aggressive. It shrank the float at a low multiple, which rewards holders, but it also put net debt at 2.1 times cash earnings on a business whose growth is decelerating, and it leaned the balance sheet on gains and cash flow holding up. Interest is covered about 11 times, so there is room, not comfort. Insider activity is nil on the buy side: two small director sales totaling $1.2M in the last year, plan status not disclosed, and no purchases. Nobody inside is backing up the truck at these prices.

How it fails or surprises you

Seat erosion. The +12% first half leans on Informatica. Organic subscription growth is nearer 9%, and if AI agents let customers do more with fewer paid users, the base drifts toward high single digits and the decade-low multiple is earned, not a gift. First print: organic subscription growth and contracted future revenue over the next two quarters.

The flattering gain. Reported profit of $3.5B and $4.29 sit far above the roughly $2.3B the business actually operated at, on a $2.2B below-the-line gain. Anyone buying "record earnings" bought a one-timer. First print: Q3 GAAP operating income and net income stripped of investment gains.

AI attach reprices the base (right tail). If Agentforce and Data Cloud add spend per customer faster than seats fade, growth reaccelerates to the mid-teens with 35% cash margins intact, on a base the market has written off at its cheapest sales multiple in twelve years. First print: contracted future revenue growth pulling clearly above reported revenue growth.

Closing thoughts

The market prices this as a mature compounder that has stopped surprising anyone, which is why the stock sits at the low end of, or below, a range it has held for twelve years. So the edge is narrow and specific. You are betting the AI attach reaccelerates a base the Street has capped at 10%, against holders who think 10% is the ceiling and that agents threaten the seat count that pays the bills. The left tail is priced, mostly: at 24 times earnings and 35% cash margins, a stall does real but bounded damage. The right tail, an actual growth reacceleration, is priced at nothing, which is what makes the setup interesting rather than cheap-for-a-reason. The fatter tail is the flat one, slow grind, but the payoff is skewed the other way.

The bet is still that companies keep paying Salesforce every year to run their salespeople, their service desks and their marketing, and that AI agents add to the bill instead of shrinking it, that when the software does more of the work, customers buy more features rather than fewer logins. What breaks it is seat erosion, customers buying fewer logins as the software does more of the work. The one pair of numbers that tells you first, before the revenue line ever moves, is organic subscription growth set against contracted future revenue growth. When the second stops leading the first, the reacceleration case is dead, whatever the reported EPS says.