Hard hardware is hard
There has been a resurgence in people building hardware. This is good.
However, a lot of it is quite simple stuff. I'm getting a bit tired of endless Arduino-style projects, involving an ATMega, a few bits and pieces wired to GPIOs, and a pile of C code. Sure, it's hardware, and sure there are plenty of gadgets out there that are like that (what does your smoke alarm do?) but how far can you go with this approach?
Are these startups building phones or laptops or servers or basestations or fancy RF things...? The hard engineering (20 layer PCB, 10GHz signalling, DDR3/4, PCIe gen 3) isn't happening outside large companies. The one place it is happening is China - the Shanzhai are making phones and tablets, which requires some decent engineering.
Maybe this is a function of commoditisation - BeagleBoards and Raspberry Pis already exist as components so we don't have to do that tedious work. But if you hit the limit of what's possible with them, you have a very steep wall to climb.
So IoT is the latest buzzword. Maybe you can do all of that with an Arduino. But if the volume's there it's almost certain that someone can do it cheaper and lower power with an ASIC - and which of these startups is doing ASICs? And if the volume isn't there, is there enough revenue to make it worthwhile except for niche products?
Someone recently said 'heavy semi[conductor development] is like steel and railroads' - in other words, needs lots of investment of money and time. Board-level stuff is less, but to do anything complex still ain't cheap or easy.