These days based on the current gpu offerings, from an efficiency perspective it doesn't make sense to try and use one unless you are targeting the high end which puts you at about 10+ liters, 1000w psu, and heavy. For me, thats too big and not portable, or of it is portable its going to be a toaster. I would do m-ATX at that point. I am only interested in something portable and under 5L while still being quiet and efficient.
After seeing the efficiency of the latest integrated systems (apple silicon, and AMD AI Max 395) I switched from my 3 liter Gigabyte LP 4060 build to a m4 laptop which is able to run anything I want either natively or through Crossover. Ill just trade it in for the latest silicon every year or 2.
I remain excited for the next proper AM5 APU release to make deskmini spicy again (with a good quantity of GPU CU which will probably be in 2027. the coming AI 400 isnt really a proper upgrade), as well as the framework ITX mainboard future updates with the Ryzen mobile processors.
I still find SFF and mini PCs incredibly useful for homelab type things and am refocusing on their remaining strengths - creating plug and play NAS kits for nuc/deskmini platform (power distro and load switch allowing use of multiple 3.5 disks with minipc) and 10 inch rack integration so people can run their own self hosted media server / security / home assistant. Self hosted private AI will also be huge and power efficient SFF platforms are the ideal way to go about it.
I have RTX 4070 and RX 9060 16GB in 5L, just slightly bigger than PS4, console case prototypes. They work fine. Also B580 Limited edition did barely fit, and worked it fine.
I would expect something like 5060TI 16GB or RX 9060 16GB made into triple-fan 270mm long reference-sized card would be the perfect for this case. And these cases should be even more portable than Sentry cases are.
As for laptop and APUs - you're buying whole thing each time, so this is a choice. And not only about the money, but also ease of upgrade. The now GPU you can just pop in and still use the old system without having to bother with whatever happens to your software licenses or drivers.
I don't expect APUs to actually get ahead of configurations of multiple components simply because we've been there over and over again - laptop vendors will not let the optimal APUs land in cheap SKUs. The current question would be whether it makes sense for you to grab AI Max 385 laptop over something with RTX 5050, but the APU-based laptop will be like two times more expensive. Yes, the APU laptop will be more efficient, but will you be still able to cash in on more expensive laptops each generation? The framework laptops could be interesting here though if they'd have something like AI Max 385.
Separate topic though is that the graphics fidelity of games isn't really pushing forward that much now and you can be fine for years with mid range card if you don't necessarily need to have 200FPS or 4K screen etc. And also growing out of playing all the newest games on release is a thing...
Apple / Arm though might be something that will be a completely different thing that will change the status quo seeing proton-based/like libraries for arm doing magic there for running games.