While public cloud-based AI processing has traditionally been the standard, McKinsey’s findings suggest that specialized AI assistants designed for specific domains are becoming increasingly relevant. This shift marks a potential departure from general-purpose large models, which often struggle to address niche needs and possess billions of parameters. The increasing demand for tailored solutions like Lilli suggests smaller models will play a crucial role in enterprise scenarios. Furthermore, balancing the cost of self-building AI infrastructure against API usage presents a key challenge. The combination of edge computing with small models necessitates significant initial investment, but promises substantial long-term operational savings for large and medium enterprises.