Hasn’t OCR been around for ages now, and doesn’t need dedicated hardware? I think the main limitation is putting the infrastructure in place to allow a browser to speak to an accessibility tool (because this is an accessibility issue) and say “give me all the text on screen which is embedded in images”.
Personally, I don’t really want any AI tools. I don’t want my system trying to preempt what I want to do and half arse it, when I could always just do it myself. I want to know exactly what my system is doing and not have it “helpfully” do or change things it thinks I might want it to.