If you're hoping to commercialize and are a 1-man company, I hope you have a good USP/feature that's beyond "it's cheaper than the competition". If you do have that, I'd say just get an MVP out while keeping systems as platform agnostic as you can. Doesn't matter if you need an extra Jetson, if your product is good enough the extra cost shouldn't really deter customers.
Once you've started, hire/contract folks and streamline it out.
BTW Dev kits do exist for the snapdragon 8 gen 2/3 I think, but they're about $700 last I checked. If you're handy you could even try to reuse a rooted phone, if that's feasible for you.
The CPU included in this is based on the same Cortex-A78 cores as the latest Jetson Orin, but it should have much better performance per dollar, because the cheaper Jetson variants are intentionally crippled, with many disabled cores and very low maximum clock frequency.
The GPU is weaker than on Jetson Orin, but it should be enough for many applications. There are also a couple of AI accelerators and dual simultaneous video camera capture at high resolution.
In any case, Genio 1200 is much faster than any Pi and also than RK3588. It has a similar performance to a flagship smartphone of 2021.
In the beginning, the SOMs and SBCs with Genio 1200 were expensive, in the $400 to $500 range, so while they were a better deal than Jetsons they still had worse performance per dollar than a SOM or SBC made with an Intel N100 CPU (or its Atom equivalent).
However not long ago Radxa NIO 12L has been announced, which uses Genio 1200 in a SBC of smartphone size at a price of not much over $100, similar to the older and slower RK3588.
Therefore it appears that it has become possible to obtain Genio 1200 at a decent price.
Perhaps if you contact MediaTek sales they might be more willing to provide design information than Qualcomm and they certainly have better prices.
Moreover the Radxa NIO 12L schematics and PCB assembly map are published on their site (a practice that was standard many decades ago, including for the leading US companies, but which now sadly survives mostly in Chinese companies), you can download them and study them to evaluate the effort of designing your custom board and possibly avoid any pitfalls in your design.
https://www.modalai.com/products/voxl-2-mini?variant=4424638...
Maybe this is something you could develop on instead of building your own.
When we're talking about new development, what are the expectations that I should have for direct procurement of higher end components (this goes for Sony/Samsung camera sensors too)? Are there typically very large minimum purchase quantities? Do I have to already have a large production output before I can even talk to them about it? Is it possible to get datasheets even if I'm not at that stage yet?
I get the feeling I'm stuck with the Jetson + Off-the-shelf camera approach until I can demonstrate the ability to mass produce and sell my existing designs, but it would be nice if I could find out more about how that is supposed to play out.