It's more fundamental than this, IMO. No matter how much you invest into faster iteration cycles, physical world just takes a long time, and the worst part is, you either have to get some very complicated thing right the first time, or add a sufficient buffer to your plans. Discovered a bug in your chip after initial tapeout? Add months to the schedule. Too much RF crosstalk in the PCB? Add a couple of weeks at least. Take Apple for example. Do you think they didn't invest into faster iterations? I'm sure they did. And each chip still takes 3+ years there, too. Qualcomm? About 3 years. Google? Also about 3 years (I heard of the upcoming 192GB TPU being developed over 2 years ago when I was there; it was announced this week). The "innovations" that they are talking about today were actually decided 3 years ago. 3 years ago was a very different time, of course.
There's also another problem when it comes to chips. On the higher end, you have to design for the technological processes that don't yet exist. E.g. 2nm did not exist 3 years ago, yet the design of the chips that use it was done back then. You're also doing it with bleeding edge tooling.
There's also another problem when it comes to chips. On the higher end, you have to design for the technological processes that don't yet exist. E.g. 2nm did not exist 3 years ago, yet the design of the chips that use it was done back then. You're also doing it with bleeding edge tooling.