Engineers, CAD, musicians, 3D artists all need more power than it provides and increasingly need CUDA cores.
I'm running a Logic set up on my 2012 MBP, and I've never felt limited by its capabilities. I've been running Macs for music production for almost 20 years, and in my experience they have all at some point not had quite enough in the bag to let me do what I wanted of them. This is the first one (now fours years old) that has stayed ahead of my needs. I can run multiple copies of Massive, and rows of Waves plugins no problem.
I have hit limits with real-time 3D stuff, like gaming and Houdini, but music production is still well within its capabilities for me.
The main gripe musicians have with apple is the new OSX breaking their setup every year. The sandboxing in el capitan was particularly disastrous.
https://www.izotope.com/en/community/blog/product-news/2016/...
Personally I had to spend a few hundred dollars upgrading my audio software that worked in Mountain Lion so it would run on El Capitan. I'm still running into a few glitches in places.
I've been running software on Sierra that is only certified by the developer for Snow Leopard through Yosemite, and it still runs just fine for my needs. Apple puts a lot of effort into binary compatibility.
This is beyond hyperbole.
1. People whose needs are satisfied by almost any laptop with an SSD and semi-recent CPU/GPU.
2. People who need the absolute top-end system but can fit their work on a single machine (e.g. if you had a model which uses 20GB of RAM, a 16GB machine is unsuitable but a 32GB machine is fine).
3. People who have so much data / computation that no normal computer can handle it.
My gut feeling is that #1 covers most of the market and the question is really how many people fall into group #2 but not group #3, especially in the context of laptops where the ceilings are smaller on both the Mac and PC side. I would further expect that a fair number of the people in the second group are not running those workloads 24x7 and thus have a practical, often cheaper, option of renting an hour of time on AWS/Google/Azure/etc. when they need to do something and get the results faster rather than leaving their laptop running for a day or two — even the best laptop GPUs are smaller capacity than what you can rent on a server.