All of the above; being a mid-career IC/TL with visibility into actual outcomes in prod has organically matured a deep respect for boring technology---warts and all---especially when cumulative NRE is well into 10-digit territory.

I'll be glad when legacy support hits EOL though, if only for the indirect consequence of unambiguously seperating wheat from chaff as we embrace a new tech stack approaching stability. Although it isn't the case, it sure has felt like we've been operating like cost center baggage these past few years.

Equipment of this time frame was highly reliable. The machines were expensive to purchase and very solidly built. I would be surprised if this fleet of systems deals with a hardware failure a year. Especially if they are kept in a space with cold conditioned air.

I'm a R&D plebeian; not my job. We have a lifecycle management team that monitors metrics---e.g. system availability, failure rates, spares inventory, obs, etc.---and postures as risk/budget considerations permit.

In the aggregate, these legacy systems are on the right tail of the bathtub curve quickly approaching EOL, hence their modern replacements already being fielded. As legacy systems are decommissioned, reclamation buffers the spares inventory pool. Push come to shove, cannibalization in the field keeps most balls rolling at the cost of diminished availability/throughput.

Based on a sample size of 2 legacy systems in our lab that saw daily use over the past decade, things definitely broke, but I don't recall the production Alpha hosts every giving up the ghost beyond CMOS battery replacement nuisance.

We still have a few support Alphas laying around just in case, but virtualization[1] on commodity bare metal effectively renders them redundant.

[1] https://www.stromasys.com/solution/charon-axp-alpha-virtuali...