A technical-diligence practitioner’s article argues that AI acquisitions should assess actual code and operational architecture rather than infer proprietary technology value from revenue metrics alone. It identifies external model API pricing and shutdown risks, training-data rights, dependence on key developers and post-acquisition integration costs as priorities. Customer examples cannot be generalized into typical acquisition outcomes. A RAND report reviewed alongside it also emphasizes data, infrastructure and problem definition, but the often-quoted 80% AI failure rate was not directly measured in its 65 interviews.