But I see the criticism here as just personal desires to get it for free just because many others do it, rather than a valid business argument.
Free as in free beer, or free speech?
The free-beer compiler (without optimizations) is plenty for what it does. And if more is needed, the paid compiler is reasonably affordable. From a business perspective this is the easiest calculation to make. However, as I stated, I doubt whether it's of much value to sell a few k$ (max) of licenses to customers that buy literally tens of k$ of chips.
But you cannot pay to get the free-speech away. If one is happy with a closed ecosystem, then that's fine. But for engineers that puts down certain requirements (like C++ support in a 16-bit project, or newer compiler features for 32-bit parts), then it removes Microchip's PIC parts as a potential candidate. A business isn't as quickly going to fuss around with unofficial XC16 compiler builds or manually pulling MIPS GCC from upstream. Those open source projects are too unmaintained as of now to be worth the time.
I think this discussion is largely a hardware vs software engineering perspective. In hardware world, every tool costs money, and dare I say that not every hardware engineer likes to spend a ton of effort/research on the most sensible tools that can be used in the long run. So an "all-in-one" package is arguably best for those use-cases. In the software world, every developer keeps a close eye on any dependency a project has to include. Engineering debt in software under active development is a huge burden. And these include the tools being used alongside large libraries or frameworks.
I regularly read stories here about spinning up old VMs to continue working on an old embedded firmware project. Those are the horrors of needing legacy compiler versions that only work on EOL operating systems.
Probably no company roadmap is going to give a rat's ass about academia, but in many of the papers that I read, I seldomly see a PIC part being used. The last one I saw was from 2016, where a PIC12F was programmed to serve as an OOK wake-up radio. I find this fascinating, as I'm sure that PIC is/was a very popular part family used in education, and probably the first MCU that many learned to program on (e.g. a PIC16F84, PIC16F877 or PIC16F628A). But that is many moons ago. From what I can see, MSP430 is incredibly popular, but so is AVRs (even used in the Arduino form factor), ARM (in particular STM32) and ESP32's. Now AVR is part of Microchip so that's their slice of the cake; but I think that success can be largely contributed by the works of the pre-Microchip era, and in particular because of projects like Arduino, avrdude and avr-gcc. Atmel always gave good vibes to the people that liked writing firmware, even 1 - 1.5 decade ago.
But at this point I'm making arguments out of sentimental value, and indeed, that's of no use in doing business. I just meant to say that these "free-speech" choices often steer away people from using PICs (again).