Google AMP was never great as a standard. It's too fuzzy and random and nobody _really_ cares about it beyond the search ranking benefits.

I understand where they were coming from, but creating a reduced-interaction, faster-loading HTML subset would require way more spec work than they put in to be self-consistent and long-term maintainable.

This is the same sort of reason PNaCl didn't get traction. Both were great science projects, but neither was appropriate as a long-term standard. Hopefully we'll see the WebAssembly equivalent to AMP from the ashes of that project.