[analog31]

I design my own boards too, but throwing a SOM into my layout is way too attractive to pass up, especially when schedule matters more than COGS.

Make-versus-buy decisions abound in product development. If you aren't a full stack electronics and firmware designer, then you have to manage the design project with its uncertainties, including schedule. A buy-in gets you moving right away, to address other project issues that require a prototype. You can always design it out later.

Dealing with complex chips often involves discovering bugs in the docs, and a second board spin to fix them. Then you'd better have a good scope and some real knowledge.

Despite its hobby orientation, the RPi is legit.

[Turskarama]

Sure but the added cost is not a big factor when you're making a handful of them.. If you're making tens of thousands though? Then it begins to add up.

[arcticbull]

10,000 @ $35 is $350,000 - that's really not that much money. That's like 1 to very generously 3 engineers for a year. Can you really design, build, validate, ramp, stock, fulfill and deploy your own hardware for less than that? I strongly suspect they ran the numbers and the answer was a hard no.

Not to mention the time penalty - as a rule of thumb it takes about minimum 6m average 12m to go from concept to physical units in people's hands at scale. These Pi's are shipping today. As a startup, that's hard to pass on.

[analog31]

And then the designer up and quits, leaving you with a sketchy and gratuitously complex design that nobody can fathom.

A volume like 10k puts you in a dreaded "in between" zone. You're not making a one-off widget that a customer is willing to pay an unlimited amount for because it solves a unique problem. And it's not enough volume to make something that just rolls off an assembly line. All of the simplistic mental models for how to proceed, don't work.

I work in such an industry, making specialized equipment. Every component is an expensive buy-in. It's ridiculously hard to find cost savings every component is up against the same math. Then again the competition faces the same problem.