This is why Microsoft never seriously pursued piracy of Windows and this is also why Windows was never a market leader on servers. This is why Postgres won databases even though it’s clearly an inferior product to Oracle, MSSQL and DB2. This is why CUDA is the defacto standard in GPGPU. This is why every saas business must have a free tier. Etc.
Postgres won out because it was better than the others considering the money you pay and the features you (in the end) don't need or couldn't afford with the others. If it just were down to learning, MySQL/MariaDB would have won. Back in the days, everyone knew MySQL, nobody knew Postgres.
With CUDA, it also isn't what people know, it rather is the existing heap of software that only runs properly, quickly, efficiently in CUDA. People buying Nvidia cards and CUDA-based software don't care about CUDA and don't know any CUDA, they are usually higher level, but the availability of software is what drives the popularity there.
With CUDA, you're even highlighting my point:
> the availability of software is what drives the popularity there
Like, there isn't anything more to it. That's all that matters. Again, a free, good enough product that evolved into a best in class software and hardware package together with a generation of GPGPU developers who don't know and don't really care about anything else.
I've over-interpreted the quotation in your post about what people have learned to mean that it is only about the tools you know, nothing else. In hindsight, I should have read your post more carefully.
Oh, also, MySQL just updates in place without bitching. Postgres wants you do install both versions side by side and migrate the data directory. That is annoying with docker.
Also, vacuuming.
But indeed. MySQL's great weakness and great strength is that it's a somewhat limited SQL engine over a variety of storage backends. It can not be too smart, due to the sheer variety of what it supports.
The side effect is that it is quite predictable.