Build an application in Python, then as soon as it becomes pushed by load you're going to need a rewrite. A few small load balancers and virtual web servers doing PHP can handle tens of thousands of requests per second, the database is almost guaranteed to be the bottleneck.
Python is a (according to some) nicer bash, it's not a good alternative for implementing non-trivial systems and web services.
Python took some getting used to, but the more I used it the more strongly I began preferring its lack of things like curly braces, semicolons, even parenthesis in some cases. Because of that, python feels more concise, human-readable, and more efficient to write, than PHP.
I like both languages, and still rely on PHP for a lot of web work, but for data science or ML I do end up enjoying my time with python.
Right, Python has a purpose as a glue language.
I'm more interested in your claim about "usable for any application". How come all those fancy ML types and learned academics refuse to use Python for the core computations they perform when they try to invent witchcrafts, instead opting for C, C++ and Fortran libraries?
Sometimes the algolians are called general-purpose programming languages, as opposed to SQL, functional programming languages, logic programming, and so on. Sometimes anything it's used as an opposite to domain-specific languages.