Some kinds of complexity increases do significantly change the cost. It used to be that we didn't need matched-length pairs on PCB traces back when timings and frequencies were low. That means all sorts of flexibilities as to what components you use, how you lay them out and what kind of PCB will work. That suddenly got much more expensive when you needed to use larger or additional layers on your PCB to get exactly length-matched traces. But between that being a requirement and the "old way" was a decade+ gap when plenty of other upgrades were happening.
Some major jumps come to mind:
- Single layer to double layer
- DIP to SMD (mixed first, where they actually glued SMD components first and then did the solder bath)
- Signal layers, integrity layers
- Microstrip vs. normal trace (luckily for us, not happening as frequent as it could be)
- Metal purity impacting signal integrity so now we need more expensive materials
- Higher frequencies with smaller buffers/preambles mean big performance increases but also less tolerances in design errors and manufacture imprecision
Even something seemingly insignificant can do this and drive up the cost: VGA to DVI wasn't all that expensive at first, but when DVI started to surpass the frequencies required for VGA the cables needed to be of a better material with less room for error, which makes both the raw materials more expensive and the process to make this stuff more expensive. This happened with USB2 (which would generally work with the same tolerances as USB 1.1 and 1.0) and USB3 too, and then again with USB3.2 (or whatever they call it now) and USB-C. And again with HDMI 1 to HDMI 2 (and the steps in between) and with Displayport 1 to 1.2a and again to 1.4a.
The big price increase generally is a combination of:
- Stricter requirements on materials (more expensive to make, more expensive to buy because validation is now more complex)
- More steps in the manufacturing process (sometimes linearly, but sometimes exponentially, i.e. when you do something special with your PCB layers, you end up with 1 extra layer step causing N layers extra steps)
- Fewer people are actually capable of designing this
- Fewer people are capable of designing for production with enough sophistication
- Fewer factories can make the investment to actually setup production lines
- You generally can't use those newer processes with older designs unless you want to be reconfiguring your machines all day (which you don't), so now you have to also keep all the old lines, old workforce, old processes next to the new stuff... bleh
As for the introduction of new stuff, that is indeed a problem on its own, because you generally need to keep old and new alongside each other for a while, and also work out the kinks in your designs and processes so you can increase your yield and not trash 30% of the stuff coming out of the production process. But once something new has been adopted widely enough (i.e. PCIe 5) the demand for the old stuff goes down, the knowledge might have spread a bit (making knowledge workers more readily available), the processes might have been optimised and fixed up, so now price can come down a bit (after making fat profits of course) and yields can go up. Stuff like this generally happens during the 'pong' after the initial 'ping' when a new bump in a standard is released, just like it does with Intel.