OP's point was that those are all techniques that you don't need AMP to implement. So their question is: if those are the reason for you to use AMP, why use AMP and not just implement them directly?

Why not use AMP? If you don't use the special caches, it's just another library.

I really question this "inline all CSS". Consider:

- 1K of HTML with 40K of CSS in a file with a long term cache. Clicking a different page on the site downloads another 1K of HTML.

- A 41K file with everything inlined. Clicking a link downloads another 41K.

The recommendations are to inline the important (render-blocking) CSS, so at most only the part of the CSS that applies to the 1K of HTML should be inlined.

The recommendation you describe is a violation of the AMP standard, which strictly disallows this for performance reasons.

Sorry, I was commenting generally. Don’t know about AMP. How does that make sense, though? What performance reasons could there possibly be for inlining CSS that doesn’t apply to any element in a page?

Likewise, I'm sorry, I didn't mean "you have to inline CSS that doesn't apply".

However, if you have non-render blocking CSS, or CSS that's used for below the fold or generally lower down the page content, "only render critical CSS inline" is usually coupled with "and then have the rest of your CSS in an external stylesheet". Which you are not allowed to do.

Accordingly, it's ALL inline, all the time.

When surver push of resources gets better supported by browsers, I bet AMP will be updated to change this restriction.

Ok, but that does not sound like a real world example.

Why not? The Medium link we're all looking at right now has a 43KB CSS stylesheet. It's sent with browser headers giving it a long cache.

In the AMP world, this whole stylesheet would end up inlined, and you would download it again the next time someone posted a Medium link on HN.