At what point is MTBF (mean time between failure) so large that it doesn't justify a support contract and/or shelf spare?

I realize this is a purely subjective post, but I was recently noting the MTBF for some devices and they are like half a century. Cisco 2960X 48-port switch is 442,690 hours between failure, which according to my calculations is over 50 years!

I realize there are mission critical cases where you can never be without a spare or support contract, but how do you make that judgement call? There's no way these things will be in service 50 years from now, and that's just the mean, some could last far longer. Is redundancy even needed (for other than load balancing)? Is there other information that can be used in combination with MTBF to determine probability of failures?