You're confusing cause and effect. There was a movement that posited semantic markup would be good, and we all (mostly) accepted it. CSS's technical shortcoming

You're confusing cause and effect. There was a movement that posited semantic markup would be good, and we all (mostly) accepted it. CSS's technical shortcomings were very pronounced in the beginning and mostly abated soon after. Achieving the perfect 3-column layout was still hard, as was really centering something, but most other stuff wasn't that far out of reach. Now there are new problems having to do with bloat, and of course many people use Bootstrap wrongly, but nobody is honestly suggesting you go back to table-driven layout. I'm saying, this is the philosophy that got us here. I'm not saying it works well or that most content is semantically tagged—it isn't. Your concluding remarks also seem off-base to me. TeX to some extent and LaTeX to a much greater extent put a lot of emphasis on semantic markup. It's just locked up in TeX-land and hard to parse outside of it, so the utility is limited, but every edition of the TeXbook and Guide to LaTeX2e expends great effort encouraging you to introduce semantic macros into your documents so that changing the style does not require extensive rework, as it would a Word document without stylesheets. Even the old Nroff/Troff macros had semantic-ish macros you were supposed to use. And DocBook really could not be further in this direction, where all the styling happens in CSS or XSL, and you have tags like `<computeroutput>` and `<funcprototype>`. I don't see how semantic meaning is an imposition, and anyway, the document authors and the document readers are not the same people.