Back when I was involved with this, the answer was: "it is hard to answer". A number of years ago, I was working with a Global Foundaries fab on a cutting edge process node and they had no idea how to estimate the lifetime degradation due to high temp. The issue was it was unknown on how to accelerate the effects of high-temp for some of the failure mechanisms. I seem to recall hot-carrier injection being one such mechanism, but I don't recall for sure.

If you can't accelerate the failure mechanism in a well defined way, you cannot lab test long-term reliability other than just letting the system run. So if you want to guarantee 10 yrs of lifetime, you need to run it for 10 yrs. Obviously, that doesn't work for most products. Instead you have to rely on models, which, for a new node, might not be proven very well.

Bottom line was this vendor shipped products that nobody could really be sure would last the promised lifetime. This was regarded as a top silicon fab in the world. Back then, finer geometries were much harder because of new failure mechanisms. I don't know state of art today, but I would not be surprised if measuring reliability is still not a really hard problem for new nodes.

Edit: I am just referring to silicon issues, the rest of the PCB has its own set of issues totally separate. Generally however, those are better understood than a new process node.