Re: The whole premise of this article is bullshit
The problem with these sorts of analysis is that it always looks at raw compute and storage and ignores the hidden cost. Take some simple web app, its got frontend resources, a backend api, a cache, probably some file storage, some sort of message queue, background workers, and at least one database. If you look at the raw compute power to run this, no shit, its going to be bundles cheaper on prem.
But you now have other concerns. You need people who know how to run that database, back it up, restore it, resize it, make it HA, scale it. You need people who can operate a message broker service, keep it up, monitor it, replay events. Its the same for almost all parts of the stack. With cloud, you don't need those people, you can concentrate on your core business logic - the frontend, the api, the background workers - and have a cloud ops team who rely on the provided cloud resources for maintaining all those essential services that are not your core business.
It's the same with most IT SaaS solutions. You can run datadog, sentry, gitlab, grafana, etc self hosted and on prem if you want - most choose not to, most choose to use the managed services, because investing time and resources into running these services yourself is not what your business does - its just another cost centre. Same with cloud services.
Build it on the cloud, concentrate on providing what your business needs, and price the cloud costs in to your service. Hopefully if you are successful enough to grow to a point where your product is big enough that bringing it on-prem provides enough saving to pay for the expertise you need at that point.