If latest chips were unavailable (war, civilisation collapse), we'd do just fine with existing chips for decades or maybe even centuries.

With clever software, you can eek much more real world use out of hardware. For example, your PS3 is probably sitting with its CPU idle right now, when it could be doing useful computations for someone else who needs more computing power.

> existing chips for decades or maybe even centuries.

Chips experience solid-state diffusion, so I'd expect the expected survival probability of chips (ignoring any infant mortality effects) to go something like feature_size / sqrt( time ). If this is true, then I'd guess modern process nodes would have very low survival rates out at 100 years. Though, it's been over 20 years since I had any solid state chemistry, and most of my experience with diffusion modeling is in financial models, so take my wild guess with a grain of salt.

But the failure rate in observed time for non-nand components is very very low.

Take a 5 year old phone, and the chances of the main board being bad might be 10%... But the chances of the main CPU being bad is probably under 0.1%. The other 99% of failures are mostly water damage, physical damage, bad soldering, fatigue failure, bad nand, etc.

Maybe they'll all fail at once, but I somehow doubt it. IC's from the 1970's are still going strong, also with a low failure rate.

You don't have to be clever (unless you're talking about video games), just not completely reckless.