I mean that when you click a search result in Google you should go to the publisher's site, not to Google's cache of the content. It may have changed since, but that's how it was implemented originally by Google:

https://www.wired.com/2016/02/googles-amp-speeding-web-chang...

"…but there’s a big catch: if readers decide to share a link to an AMP page they’ve clicked on through a Google search, the link points to Google.com (for example, google.com/amp/yoursite.com/yourpage/amp), not to your site. A Google spokesperson confirms that there isn’t a way to both have your AMP-optimized appear in Google’s prioritized search listings without having that content hosted on Google’s AMP Cache servers."

I was under the impression that the publisher could elect to use their own servers for their AMP content. Is this not possible anymore?

There's a lot of doublespeak around this.

1. You can (and must) host AMP content on servers of your own choosing.

2. Google's crawlers ingest your AMP content, validate it, and re-host it in Google's AMP Cache. Bing and other folks run similar crawlers and caches.

3. Google search results will only link to AMP content on Google's own cache.

So yes, you can host your content on your own servers. And you can give people direct URLs to that content. But Google users will only ever see links to the copy of your content that Google hosts; your servers never see that traffic.

There is no way to opt-out and still be considered a valid AMP page.

Publishers already host their own AMP content. Google caches that content, but anyone can go to the publisher's URL and get it directly.

Disclaimer: I work for Google and have some involvement with AMP.

If by "anyone" you mean "my tech savvy friends at google" then yes. I've seen many friends end up on a google-hosted AMP version of a webpage, unable to complete whatever ticket purchase etc they needed because whatever autoAMPifyer the origin site uses produced half broken pages. These users have no idea what AMP is or why the page doesn't work (or that they could have reached the origin site directly if they click around on several unlabeled half-invisible icons in the fake-address-bar on top). In fact, they don't even realize they aren't browsing the origin site. They just give up.

Conceptually this isn't all that different from having a website that works in browser A but not B due to insufficient testing. Why did they not fix it?

Because how many website operators continuously google their own web site on mobile devices to click through to experience their google-hosted AMP editions? I would wager it's a fairly small % of the number of websites running an auto-AMP-ifying wordpress plugin, for example.

I've seen plenty of badly designed mobile websites over the years and these sites seems to have turned out okay - most mobile browsers keep the option of "Request Desktop site" very accessible for a good reason.

The worse case I've seen is a site in which every page crashed Mobile Safari without fail regardless of which version you ask for. It was eventually fixed but I never figured out why. If the sites are just running some script without checking then the admins have failed their line of duty.

Well, I suggest you start giving AMP content that isn't in Google's cache the same preferred treatment in search results that you give cached content.

I'm just going to file another complaint to the EU Antitrust committee otherwise, as this is a simple and clear violation. (Although I doubt my own complaint would be very relevant — all large publishers already have filed such complaints, and Google will be fined for it).

My pages actually increased their loadtime tenfold when I tried them with AMP — they're tested on a HUAWEI IDEOS X3 on 64kbps GPRS. AMP increases load time to over a minute. (On a modern phone, with a modern connection, my pages are obviously also faster — I'm also testing with a Nexus 5X on 100Mbps 4G)

I have the choice between a massively worse user experience, or worse search ranking. And that's a choice that's just not acceptable to me.