Using the amp cache requires pre-loading content in the end user's browser. That makes things really fast (there is zero request latency to "load" and AMP page), but has security implications. If I stick `window.onload(send a bunch of user analytics to my personal server);` in a preloaded page, than anyone who does a google search where my resource is a result unintentionally loads my page and then has data leaked to me, a random nefarious internet person.
So given these two requirements:
1. 0 request latency
2. Prevent data leakage to third parties
Come up with a scheme that is different from AMP. Or you can argue that either of these requirements is wrong: that data leakage is okay, or more likely that some low-but non-zero level of latency is alright. But my understanding is that the options are either use something amp-like and get approximately 0 latency, or use something non-amp like, and have latency of 50-100ms minimum on good, wired, HTTP/2 connections for well designed websites, and that becomes really bad, really fast, if you're on a 2g or 3g HTTP/1 connection in rural $wherever that's dropping some packets.