Presently, to do the AR processing to extract features from a space uses either cameras or some kind of LIDAR or time of flight scheme to get a point cloud of l

Presently, to do the AR processing to extract features from a space uses either cameras or some kind of LIDAR or time of flight scheme to get a point cloud of locations of objects around you. The data from a single camera that is say ~1-2Mpixels (say 1280x800 RGB camera) at say 30fps is only processable on an SOC or an asic. If you choose to process the raw image locally then you need to have feature detection, extraction, etc algorithms that can run locally on that processor. For AR, a more realistic approach to a true all-in-one solution that sits on your face is to create an ASIC that simply gobbles up raw camera data, H265 compresses it, then sends it over a low latency link to your phone sitting in your pocket. Then when you consider that the chip also needs to drive a display (or 2 displays) of some sort, then you realize you need to also must have a way to receive that video data (potentially H265 video decoder now also needed), a way to display that video stream and do some potential fix-up on that video stream (at least some kind of GPU needed) -- now you realize you need a relatively capable SOC that runs on your AR device on your head. So you need a cable / battery pack etc -- even if your phone in your pocket is still the main thing running the show. So it just means that the combination of needing to use cameras, drive displays, and receive data wirelessly means you have some fixed costs, power needs, and limited set of SOCs to choose from at present. So its do-able, just its early days for the hardware to support it.