That sounds like the opposite of nuts. What would be nuts is to do a whole bunch of work to upgrade something that's already working for literally no reason at all.
The problem is, once you ossify, you can't upgrade for whatever reason or extend functionality because the build tools will get lost or won't run on modern stacks any more.
Or, say, some industrial machine - you might wish to allow network connectivity so that the vendor/service teams can get notified once a sensor detects signs of upcoming failure such as vibrations. But you can't connect it to your network, or have to take it off your network once the vendor's security support has passed, because otherwise you're just asking hackers to establish persistence on the system.
How about spare parts? Sure, IDE disk drives (aka the thing that fails the most often) can be replaced by CF cards, but if your software hasn't been touched in 20 years, good luck getting it to run on a "modern" embedded board. You'd have an easier time if you had invested into keeping the embedded OS, the build environment and the boards reasonably up-to-date.
There are two industries the are leading in obsolescense managememt: Aerospace and railways. But yes, this is a seriousbissue if service love of your product is measured in multiple decades.
At least in Europe, railways got the hint (or rather, the EU forced their hand)... they are switching over to ETCS and digital signalling control based on standardised interfaces. Will take a few decades to fully roll out, though.
Hence obsolescence management to litterally keep the trains running, sometimes even on time. And no, regardless of what smart phones do, you do not update a running, and certified, piece of equipment with a real technical need to do so. One of the reasons rail, and air, travel are as safe as they are.
And it is not just signaling, it is rail car and engine systeks as well, down to the last mechanical and electronic component.
