Three $3B+ Acquisitions in 30 Days Prove AI M&A Is a Data Grab, Not a Model Grab

Salesforce paid $3.6B for Fin (formerly Intercom) on June 15. Autodesk paid $3.6B for MaintainX on May 28. Schneider Electric paid $3.1B for Cognite on June 30. Three separate B2B acquirers, three separate categories — customer service, facility maintenance, industrial data — inside a single month, and SaaStr’s Jason Lemkin makes a specific case that all three bought the same asset: proprietary interaction data an AI model needs to actually work in that domain, not the model itself. The Fin numbers make the case concrete. After five straight quarters of declining net-new ARR, the company built a working AI support-agent prototype six weeks after ChatGPT launched, turned over roughly 40% of staff, and rebranded. The AI agent line alone crossed $100M ARR, growing 350% year over year and driving essentially all of Fin’s net-new growth against a flat legacy seat business — while net revenue retention moved from 112% to 146% as pricing shifted to outcome-based. Fin now resolves 2M+ conversations weekly across 8,000+ customers.

That 146% NRR is the number worth sitting with, because it’s evidence for a claim Forrester made the day before the SaaStr piece: principal analyst David Mooter argued that an AI agent without integration into a company’s existing systems is “merely an answer engine” — it can’t take action. Fin’s acquisition price and its retention curve both say the same thing from the buy side: the model was never the differentiator. The data connecting the model to a specific company’s live customer conversations was. Salesforce’s own positioning makes the acquisition read almost contradictory on its face — the company has been telling customers all year that “ownership is non-negotiable” and that its Data 360 platform won’t aggregate a customer’s records to benefit anyone else, while paying $3.6B specifically to own Fin’s aggregate support-conversation dataset outright.

The Ceiling Method read: when a vendor’s AI feature gets a price increase this quarter, the question worth asking isn’t whether the model improved — it’s whose proprietary data now sits behind it, and what that acquisition cost the company that’s about to pass the bill to you.