You’re becoming a regular here

The high initial cost of hardware integration is definitely a barrier to entry and also a very good way to weed-ou those companies who can’t cut it.

However when hardware integration is done properly the business model produces fantastic returns. Once a “chip” is done the costs associated with the R&D are then amortized….after amortization the “chip” margins tend to skyrocket.

Obviously software is part of the R&D and in a system some of that software cost burden is more or less constant but the overall result can undercut companies who are cleverly connecting discrete components.

I’m not sure I agree with all of your conclusions. At least not for the “IP” equipment vendors. My guess is you are right when the equipment is in fact mostly hardware, e.g. optical equipment.

Is integration in Silicon cheaper? I doubt for the complex Silicon that todays NPs (Network Processors) are. They are still processors, i.e. you have to write code. It is notoriously difficult and error prone, which drives costs. It is somewhat inflexible. They are produced in small numbers compared to standard parts. If anything then it is costly. But it’s the only way to achieve the throughput, jitter etc, so you do it.

From what I can see in my company the main costs are not in the hardware. It’s the software. The larger part is control plane and infrastructure code, i.e. independent from the hardware. But then I’m working for an IP/MPLS 400k gorilla company, not for a lean optical startup

You mention an important detail: “made up of cleverly connecting simple components”. We have to face the fact that the Chinese do an excellent engineering. Many western vendors have lost the idea that a good solution is – relatively – simple.
(but when it comes to cars we admire the puristic design of Lotus and a few other driving machines