The key point in your statement is the first few words: “To sufficiently large or technically skilled organizations…”
Note that my comments are primarily focused towards companies outside the software industry. I’m not talking about Silicon Valley startups but rather Wall Street banks.
I’ve worked with hundreds of large enterprises each of which had hundreds or thousands of IT personnel and software engineers employed. The problem is not that they didn’t know what they were doing at any given point in time (most of them did) but that the employees and the strategy change over time. Each generation ends up with a few (or many) warts in the architecture of their on-prem stack. They standardize on vSphere for a few years only to switch to KVM or Xen at the whim of an executive who wants to “innovate” or because the finance department wants to save money. First they buy NetApp, then they buy EMC, then they buy HP Storage Servers, then they buy vSAN, etc. Don’t forget that they also need management tools, backup tools, firewalls, etc.
Every single enterprise I’ve ever worked with has had this problem. Invariably, the IT department finds out that they don’t have the time or money or expertise to move all workloads to the new environment so they end up with multiple silos, most of which they’re afraid to touch. These guys still have some apps running on 1960’s IBM mainframes!
You could argue that the same is true for the cloud — they’d start with AWS and then switch to Azure or GCP to save money. You’d be correct, hence the other half of my recommendations. Always use the highest level of abstraction possible for the problem you’re trying to solve (IaaS, PaaS, SaaS representing ever increasing levels of abstraction in the cloud, just as HW, OS, App do so in the on-prem world). You’d protect yourself from AWS lock-in, for example, by using a PaaS layer like CloudFoundry. In the case of a bank implementing an HR system, of course, they’d be better off just going with a SaaS solution because HR is not part of their core business value.