Not so sure.

If you look around, you'll find that a huge percentage of all electronics in production contains a number of single-source components.

Since a number of years, the sharing of designs between manufacturers have almost totally died away.

So the electronics may contain single-sourced DC/DC converters from Linear. Single-sourced ARM chips from NXP. Single-sourced CAN transceivers from xx. Single-sourced FECs from ...

30 years ago, you produced a product for 10 years or more with zero hardware changes.

Today, you redesign the same product every 12 to 24 months, because you need to make use of the latest components just to cut costs. Producing the same hardware for 10 years will not let you cut the production prices with 20-30% each year. Something you need to do today, just to be able to keep your market shares without having to sell at a loss.

It is fewer and fewer niches where you can charge stiff overprices for a product just because you call it "industrial".

So what does it matter if the CAN controller has multiple sources? The CAN controller gets thrown away and moved into the processor, just requiring a trasceiver chip outside. The network controllers gets moved into the processor. The memory gets moved inside. The external UARTS gets integrated. The prioritizing interrupt controller.

In the end, we end up with very few chips driven by C or C++ programs (to make sure that as few lines of code as possible needs to be rewritten when changing target hardware). And the main difference in quality is the design of the power supply and how well all inputs/outputs are isolated and protected.

But in the end - todays product cycles are not so affected by second source supplies. A far worse problem is delays in availability of the new components, since each new hw revision brings in one or more components that are in alpha or beta stage when the development starts, and that will go "live" about the same time as the product is about to start being manufactured. So while backup plans for phased out components was very important 20 years ago, backup plans for delayed introduction of components may be more important today.

One thing that follows from the above, is that you need to have quite large market shares, since you can't live 10 years on your product certifications. You need a large yearly volume to take the cost of constantly having products tested according to different standards.

About Atmel - my worst problem with them was that we had to make do with Mega48 chips because their Mega88 chip got significantly delayed. No fun to have to squeeze the program in to 50% flash, 50% RAM, 50% EEPROM when you had originally selected the Mega88 as the best fit... It seriously hurt the maintainability of the software.