Don't know your design - I have a system here with 48no 1080p IP cameras - in theory I could do that all from a single 48-port switch.
In practice I'm using 4no 24-port switches for resilience, each camera is covered by another camera (so I can see if someone is attempting to tamper with a camera) and no two adjacent cameras are cabled to the same switch (so if a miscreant kills one switch, the neighbouring camera is unaffected). Every camera has the cabling present to cable back to two different locations.
Those 4no switches are not stacked and are in 4no separate locations, each with utility power and a UPS - they are each dual 10G connected to two distribution layer devices, with the facility to add to the aggregation if needed. I could have gone straight for 40G uplinks but 10G is sufficient for 12no cameras (and the additional link is for resilience) and would continue to work with up to 24no on a single 10G link.
Think you're right to be concerned at performance of your network - I've kept my cameras on a separate physical LAN, mostly for reasons of security with the side benefit that someone hammering the "regular" LAN cannot impinge on the performance of the security LAN and vice versa.
As for choice of model - my start point was the inherent reliability of equipment - as someone with a background in reliability theory, I know that if a device is designed to supply 370W and I ask it to supply close to 370W full-time, it will inevitably fail sooner than if I use a device capable of 740W or more ("derating for longevity" - see https://www.weibull.com/hotwire/issue92/relbasics92.htm for the very basic overview, it concentrates on single component reliability, but remember that your switches are made up of thousands of individual components).
Given that this is a security-based installation, as far as possible everything is at least duplicated (2no power supplies in each of the 4no switches, served from separate sources, housed in separate locations) - reliability, performance and resilience doesn't arrive for free, so it depends on what your security is protecting... mine is a layer in a system that is used to gather evidence of (attempted) abduction and (threats of) violence against staff, so cost is typically a minor factor in decision making.
Short story: Start by drawing out the most basic possible system and then ask yourself what if each part (and sub-part) of that fails? One switch fails? One power supply fails? One source of utility power? One Link? Then think like a criminal: what if someone tries to disable a camera? etc
Put your full paranoia hat on, plan for the worst and hope for the best.
