At that point, you'd have to cache the JSON responses just the same, so the complexity of cache invalidation would be there in the exact same way.
But you would still not have the additional load and bloat on the user browser.
Load and bloat? Well that depends. If you're doing a realtime collaborative texteditor with a realtime chat, I just don't see how you'd get anything even remotely nice without what you call "load and bloat".
However, to return to the original topic. Which is simpler: a) creating a performant data API b) creating a performant server that handles data stuff just like a) AND all the rendering & UI stuff
Surely the latter is more complex since it has more responsibilities?