Industrial AI in the articles context regards the use of industrial products not the manufacturing of them, where the challenges around data are different again.
Indeed, the challenges around data when designing the products are of different nature, making sure data are collected (choice of sensors) and possibly also processed locally (Edge AI), like the case of CERN I mentioned where Blue Yonder's detectors were used to filter out 100k events out of the 600M received by the detectors' CMOS arrays.