For example Google and Meta both scan all material that is on their services, both of these providers actually go further by utilising AI to hunt imagery which may not be identified yet. So the idea of this compelling others to scan is false: Apple is last here.
Similar to Google and Meta, Apple's CSAM processes only come into play when the material is uploaded to Apple's servers. The minutiae of where the scanning occurs is irrelevant because Apple's implementation is merely a way of performing the function without giving up the user's privacy (by scanning locally rather than sending all of your photos to Apple's servers for review.)
The scanning process is merely the addition of verified hashes to a system which already uses this same technique to categorise ones photos into searchable images ("waterfalls" "dogs" "bicycles" etc.), because this is an existing system there is no impact to performance/device usage.