I think the hardest part of hardware is that its development is essentially an optimization problem - find the cheapest way a factory can make a widget with the

I think the hardest part of hardware is that its development is essentially an optimization problem - find the cheapest way a factory can make a widget with the features you're trying to sell. The hard part of this problem is that there's an immense amount of information asymmetry between factories, designers, and marketers. This will get better with factory automation. Right now you can upload CAD files to get a quote. The issue for designers is you want the quote to tell you why something is expensive (eg, a feature requires three tools for machining a part instead of one, or it's slightly too large/small for a process that is way cheaper, etc). Normally you work through this with engineers, designers, and factory staff iterating together on designs before tooling. That process is expensive and time consuming - it's a very human problem that can and will be automated soon. I don't see "hardware as a service" coming into vogue for markets that are a race to the bottom (except if the service is sold to advertisers, like with Rokus). I see the service being manufacturing the hardware itself. It's already turning that direction.