Based on historical trends it seems more likely to me that $90 computers of the future will simply be able to run the most advanced consumer software locally and thin clients will just be unnecessary complexity.
The hardware will probably have the capabilities to do so, but I can see consumer software reaching a point where much of the compute is performed remotely, and we're just being streamed a rendered interface.
That would require networks to be massively more reliable than they are currently. It's all well and good if you live somewhere where you have a rock-solid connection, but most of the world doesn't. And is unlikely to anytime soon.
Some very popular CAD software already works this way (Autodesk Fusion 360 and OnShape, possibly others). The people with money to spend on such things tend to have good network connectivity.
If the hardware has the capability of doing all my calculations locally, then why would I want to do it all remotely and just receive a rendered interface? So that I can have the "added convenience" of having my software unavailable during the occasional network outage? And the pleasure of network latency, even when there isn't an outage?
We've gradually been having control of the software we use wrestled away from us.
The (unfortunate) idea is that network latency will be low enough, and rendered interfaces will be good enough, that companies can make "streaming software" a compelling enough experience for the masses, to the point where it becomes normalised.