20 odd years ago I had a couple of quite drunken conversations with a small group of clever types from what was then Royal Ordanance.
They were discussing a new weapon concept then in planning, when I asked how they figured out how much computing power they would have to work with in 15 to 20 years when the weapon was intended to be in use the response was "we guess"
1970s, the answer was "whatever we can get today", when 32 kilobytes of magnetic core storage was a lot.
1990s, the answer was "the best we can get today, and design it to be easy to change because if we specify an Intel 486/33 at the start of design, odds are it'll be obsolete and unavailable by the time we're trying to produce five years downstream. How much memory do we need? Estimate what we're using today, and stick a zero on the end, I'm sure someone will find a way to use it". You know you're going to be wrong about the future, so make it easy to change for whoever has to do the mid-life upgrade.
Trivia like finding the "right" solution to store, handle and export large quantities of trials data at something approaching affordable prices: anyone remember magneto-optical discs, for instance?
Trying to design and build hardware for a quarter-century ahead, when the commercial lifecycle of many components is a fraction of that time, can be... interesting.