It’s still used in PNG, it is part of the spec.

There’s a trade off, Huffman is relatively fast and easier to get right, and the gains from increased compression may be considered marginal.

This is a classic engineering trade off.

My understanding why arithmetic coding isn't used as much as Huffman coding is because of the patents that IBM had on arithmetic coding. I'm not familiar with any actively used compression algorithm that uses arithmetic coding.

I have explored using arithmetic coding on some data at work (mainly large amounts of XYZ points). My attempts did not work better than standard zip compression.