Another way it might go is something like this:
User always downloads everything to a proxy browser (maybe in the cloud), renders everything, runs all JavaScript, and lets it send back tracking info etc. But the User doesn't see that.
What happens next is basically AI-powered filtering: all ads and tracking and such is filtered out using classification (something like how spam filters work today, i.e. machine learning). The filtered version is what gets sent to the User.
This isn't a new idea. Amazon's Silk browser/system [1] is in the same ballpark. No doubt there are others.
The ironic aspect of this approach is that some of the leaders in classification (machine learning) are Google and Facebook, the two companies profiting the most from ads.