The Apple Vision Pro definitely uses dynamic foveated rendering. It only fully renders what’s at the center of your vision, but it should be adjusting where this is in real-time based on the gaze tracking data. (It’s easy to observe this if you’re giving a demo and watching an external display.)

Especially given that gaze tracking is the primary input method in their UI - they spent a lot of time to get this right.

I feel like something might have been wrong with op’s demo unit?