The internet is much bigger, more diverse and complex today. You need a lot of storage to get any meaningful impact. Caching the http of Wikipedia won't get you much. You need to cache lots of YouTube videos. Or you just get them from the data center you peer with over the fat link you built.

With bandwidth usage the diversity of the data retrived over the internet has also gone up. You can't just cache the few most popular websites and save most bandwidth. But bandwidth capacity has scaled a lot so you probably also do not need to.

> You could have some hash check to prevent hijacking. The old method would be naive today.

But how do you know that the cached site is up to date? How does the ISP know that? What about dynamic content? What about consistency between different requests that are part of the same page load?

> Sure but they are some switches away.

My point is that this does not matter much. Usually, at least in non sparsely populated parts of the world with modern infrastructure, these switches are close and there is lots of bandwidth capacity.

I just don't think it makes sense for ISPs to save bandwidth on these links by building their own local data centers when they peer with a CDN data center anyway.