Chess Is Where AI Got Cheap. Salesforce Is Betting the Money Moves to the Agent Layer.

Salesforce spent the summer in the AI showrooms. It signed with Visa Cash App Racing Bulls, the Formula 1 team, to run "fan engagement" on Agentforce 360, Tableau, and Slack. And the chess federation is being cast as the sport whose future the agent platform will power — the latest turn in a tradition, going back to a Google-hosted chess challenge with FIDE, of using the board as AI's proving ground. It is easy to read gestures like these as confirmation that the agentic-AI bet is working. They are confirmation of almost nothing. They are brand lighthouses, and the economics have to be read from the income statement, not the paddock or the board.

Chess is worth keeping in mind anyway, because chess is where AI commoditization already finished. Free, open-source engines have beaten every human on the planet since 2017, and nobody pays for a world-class chess program anymore. The value did not stay in the engine; it migrated to the network, the content, and the experience wrapped around it. That is precisely the argument Salesforce is making about enterprise AI: the model itself becomes cheap and interchangeable, and the durable money moves to the layer that holds the data and the workflow.

The test of that argument is concrete, and it just printed. In the quarter ended July 31, Salesforce reported Agentforce annual recurring revenue over $1.5 billion, up 240% year over year, a combined Agentforce-and-Data 360 run rate of nearly $3.9 billion, and 3.2 billion agentic work units delivered in the quarter alone — up 97% from the prior one. Current remaining performance obligations grew 14%, an acceleration. All of it landed on an 11% revenue increase to $11.3 billion and a 34% non-GAAP operating margin.

The number to read twice

The $1.5 billion Agentforce figure deserves a second look before it carries a thesis, because the measurement got wider this quarter: it now folds in Slackbot and Headless 360 revenue. And the combined $3.9 billion run rate includes roughly $1.1 billion of Informatica Cloud annual recurring revenue sitting inside Data 360's $2.4 billion. None of this is deception — it reflects a genuinely unified agentic platform — but it means the growth number is not apples to apples with the quarters that came before it. The honest reading is a fast-growing, deliberately re-defined slice.

Scale, and what's bought

Put that slice next to the whole. Roughly $3.9 billion of combined agent-and-data ARR against the roughly $46 billion of revenue Salesforce expects this year is under 9%; Agentforce on its own is about three cents of every dollar. Fast, real, and still small.

The same quarter that accelerates the top of the funnel keeps the headline at 11%, and a meaningful part of that 11% is bought rather than built. The $8 billion Informatica acquisition sits inside the numbers, and full-year guidance was pushed up about $200 million by the pending Contentful and Fin deals while the currency move cost the company about $100 million. The underlying organic engine is the open question, not the spectacle.

The return curve, not the thesis

The tension for a holder is not whether Agentforce is real. The work-unit growth and the 34% margin say it is. The tension is that the market has been shown this story for a year and has not paid up: the shares round-tripped from near $269 down to the $146s and back to about $245, leaving the trailing twelve-month return roughly flat while the agent business compounded. At a market cap of roughly $200 billion and about 29 times forward earnings, you are paying for the data-and-agent layer to graduate from growth showcase to core growth driver — for the ~9% slice to start pulling the whole top line rather than the narrative.

A chess federation, or a paddock full of them, changes the perception and not the subscription math. What would change the math is the combined-ARR line compounding against an organic, not purchased, core. Until that happens, the thesis can be fully intact and the stock can still be a worse place for your capital than the turn it has already made. The direction was never really the question. The near-term return curve was.