They resisted (most of the) LLM boosterism and kept decent focus on SLMs that can run on-device.
I think the decision is first a self-serving one that's in line with how they want their devices and services to operate, but it also happens to be (in my opinion) the future-proof way of integrating consumer AI.
Maybe not on the investment/new hotness side, but as a user I'm not clamoring for greater LLM integration. Maybe I'm a Luddite, though.