Google's primary source for their knowledge graph is semi-structured data on web pages, not APIs. Notably, that claim 1200M "facts" (of which 8% have "high confidence") extracted by understanding web page DOM structure. That compares to 140M "facts" from human annotations on web pages, with 0.2% high confidence (ie, "semantic web").
Given that the premise of your point is wrong it seems your conclusions are likely to be too.
Google read the CSS as well, and uses the visual features as signal what humans will see as important on the screen.
That's impossible with APIs.
There are specialised cases where Google does use APIs: public transport and airline times. They are very special cases though, and for broad web search it seems unlikely Google will follow that model.
Google Now or Siri are killing the page as a medium for certain types of content, and I would not be surprised if the info providers transition to an API-first model where the primary target is layout-agnostic and possibly supported by micropayments.