It will be great for audio or video processing, robotics and way more
My point is that there are already well supported, affordable, high-performance CPUs with lots of processing power available today for these use cases.
Two years ago, I slaved a Teensy 3.2 to a Raspberry Pi 2 for a robot, and it was great!
This year, I'm slaving a 3.5 to a Jetson TX2, and we'll see how that goes.
T3.6 becomes a little bit slow.
and RPI eats a lot of power
If you think the Pi 3 draws too much power (because: quad 64-bit cores and a GPU!) then you can use the Pi Zero W today. It's a single core, it runs at 1 GHz, it has 512 MB of RAM, it has HDMI out and dual USB and camera input and 3.3V I/O, and it doesn't draw particularly much power, and it's price competitive with the Teensy (even at Amazon scalper prices.) If the difference between Pi W and Teensy 3.6 power draw matters to you, but you need the DSP power, you might not find anything that suits you at all other than some custom DSP board
The Teensy is great because of all the hard work Paul does with the software environment, and because of the robust interface to all kinds of low-level peripherals.
Trying to grow that into a "home computer on a chip" moves away from what it's currently great at and competes with a vast variety of incumbents.
Paul is one guy, with some additional help from the community, as far as I know. Look at how long after 3.5 release we are still looking for USB updates and Ethernet to get started.
Adding more peripherals (GPU, anyone?) will just add to the backlog of features to add support for, for incrementally less gain to the majority of Teensy customers.
I've seen companies I care about develop the wrong product and waste their resources before, which is why I'm being vocal about this, as a customer who appreciate the work and care about the value they deliver. If I were running a company like PJRC, I would look very hard at what the majority of customers are actually doing, and then look at what options exist for the extensions customers want, and then move in the direction where the market isn't already full of well supported options.
My guess is that more software support for actual products would land better and differentiate better than walking the "a desktop computer on a chip" path. Also, frankly, the software has a lot more value for the future than the hardware designs, because software is reasonably easy to port forward. Companies get acquired for their great software tooling/product all the time, so there's clear value there.