For a time, I tech-led the creation of the AMP site for a major news publisher. The technical choices of AMP, excluding the CDN-aspect, are I think a great fit for publishing websites with tens-hundreds of developers who are all tempted to write bespoke JS and in so doing create performance and maintenance hell. In many respects, philosophically, I think AMP was not far of HTMX. In AMP, developers are able to construct relatively sophisticated dynamic/interactive features using simple markup (and pre-built JS components). The page is managed through a single JS runtime which helps manage performance issues. As components have a standard HTML interface, it is possible to migrate the backend to different rendering technologies partially over time unlike (for example), isomorphic JS which forces a large-scale rewrite down the line.
I tried to advocate for an in-house AMP-like solution for our main website, but it was ultimately re-written in React -- a process which took several years and resulted in a codebase of much greater complexity. (Performance was better than the old website but I'm not sure React really contributed to the gains here.)
While AMP is rightly dead, I think the technical choices it made live on (or at least, they should).