To realize the future of what applied AI looks like, you must have control of your entire experience and pipeline top-to-bottom. People ask us why we're building our own hardware, this is the reason.
I can see the benefit of an AI having access to more of your data (owning the whole experience). But can you explain why the hardware is so important? Why couldn't it be implemented as an app?
Dedicated hardware would be much, much faster than an app, and would give much more guarantees about real time performance. There are a growing number of FPGAs and ASICs being used in AI research because they can be intimately tuned for the task at hand.