Kirkwood/ Arm Mass Production Runs

Does anyone know the process for how an Arm-based product is developed and then mass-produced?

I've been researching the PogoPlug ethernet+usb+arm device, and it's really cool that they're something like $20 a pop nowadays. What would you suspect the cost per unit to have been when they first did their production run? I saw a price sticker on one indicating they retailed at $100 originally so maybe there isn't some Kirkland magic going on that puts the board + case at below the cost of most SoC dev boards.

If you were going to go to market with an arm-based product of your own, what would your starting point be that allowed you to develop from day one, but also be easy to go into production with?

I guess raspberry pi and its "replacement"?

Yeah I had a hunch RPI had sort of solved the low cost linux-based device problem and there wasn't much else in the market (unless you consider those ESP32 type devices, which are cool, but not quite linux so considerably more effort to develop for).

I ran across this article the other day about designing a board for the RPI compute module 3. At $25/ unit, that's okish, and apparently can plug right into whatever board you design. It's odd that there isn't a plugin form factor for the RPI Zero which comes in cheaper at $15.

Then there's the actual RPI 3 that's $35/ unit which I guess could be used directly and then have a plug for the for the GPIO header to plug into whatever custom hardware the device integrates with. It feels like cheating doing it that way but supporters might almost prefer that method so they can more easily extend and repurpose the hardware. Lacking the eMMC that the compute module offers is a little unfortunate too in some ways.

Someone helped me print a board via JLCpcb and they turned out great, though they were really simple boards and didn't include component assembly at all. If anyone's played around with ARM mass assembly, I'd love to hear additional insights around the process.