The price point in the long term is the cost of electricity to run your model locally and the amortized cost of buying some hardware that can run it. The fact that it isn't streamlined today doesn't mean it won't be in X years.

Investors are blindly banking on everyone perpetually going to the theater to see the talkies and missing the vision that we'll all have TVs shortly thereafter...

They're banking on "bigger computers = smarter AI".

If this is true, then even if you can do cool things with small local models, the godlike superintelligences will still be in the datacenters.