AI data acquisition is how model builders get training data they cannot scrape. The public web is used up; what labs acquire now is licensed operating records from real companies, the mail, documents, tickets and code of people doing work.
What gets acquired
Email stores, shared drives, document systems, CRMs, ERPs, ticketing systems and code repositories from companies with real headcount and years of history. The value is in how decisions were made, not in any single file.
How it is acquired
By licence. The company keeps ownership and the originals. A buyer gets a defined right to train on a copy, for a defined scope and term.
Who runs it
Data acquisition teams inside the labs, the intermediaries holding their contracts, and brokers working for the seller. A broker acting for the seller runs several buyers against each other.
How a company sells into it
Three questions at /worth give the range. The form at /start takes the systems and years. Samples go out under NDA, a buyer prices, and the licence is signed.