I love this, it's a very clever and funny way to solve the problem. Makes me think about how there are infinite routes from A to B, some more scenic and whimsical than others.. as well as all the people I've met along the way who would be so pissed and pedantic about how this isn't a "real solution" LOL

A shorter solution is possible with an ordered list (<ol>) if we're willing to ignore the untidy output:

  li:nth-child(3n), li:nth-child(5n) { list-style: none }
  li:nth-child(3n)::before { content: "Fizz" }
  li:nth-child(5n)::after { content: "Buzz" }

Example: https://susam.net/code/web/css-fizz-buzz-ol.html

  $ curl -sS https://susam.net/code/web/css-fizz-buzz-ol.html | sed -n '/none/,/after/p' |  tr -d '[:space:]' 
  li:nth-child(3n),li:nth-child(5n){list-style:none}li:nth-child(3n)::before{content:"Fizz"}li:nth-child(5n)::after{content:"Buzz"}
  $ curl -sS https://susam.net/code/web/css-fizz-buzz-ol.html | sed -n '/none/,/after/p' |  tr -d '[:space:]' | wc -c
  129

But I don't quite like how misaligned the numbers and the words look in this version. Correcting that would call for extra code that would cancel out the bytes saved.

    list-style-position: inside;

  li { list-style-position: inside }
  li:nth-child(3n), li:nth-child(5n) { list-style: none }
  li:nth-child(3n)::before { content: "Fizz" }
  li:nth-child(5n)::after { content: "Buzz" }

Ignoring the size of the HTML in addition to the CSS, it’s fun, but not really fair when talking about code golf. Beyond a few numbers, you need to include some JavaScript and generating a million list elements. But those bytes count …