Correct. My market as a buyer is high-end, not entry-level. They need to differentiate the high-end models since neural compute is becoming so important, and I think it will happen with M4 to tie in with their ML software stack that is going to improve enormously over the next year and a half. Apple is already contributing to a few open-source projects that leverage this and those people will want more power so it makes sense to offer it, especially with unified memory being such an asset as an architecture vs. the competition. Apple's AI teams just got dozens if not hundreds of engineers to add to those efforts. I'd be shocked if by M5 there's no change in the Max line for MBPs.
The vast majority of pro buyers purchase laptops despite what older users used to desktop computers will tell you.
From an enthusiast POV I hope the Mac Pro tower gets a Quad eventually but that isn't slated to happen for another 3 generations, if it ever does. With stacked dies the yields might be enough for them to do it but I don't think we'll see that until 2027+, and the "AI" story they need to tell for both users and investors can't wait for a hypothetical $10,000 machine that has only 4x the performance of a $1299 MacBook Air in this specific domain.
They could do more with FPGA cards but I suspect won't, given how non-existent the developer support was for the Afterburner (and I mean from Apple's end, there was no real devkit at all which was stupid and wasteful). Afterburner was a total anomaly and not something they'd sell as a product again, unfortunately, because an FPGA development platform would be pretty great.
I agree with others that having hardware AV1 and H.266 will be nice too, and should be coming with M4/M5.