I understand that X Does Not Support Screenreaders is an imposition for folks who have issues with standard browsers, particularly because there are an awful lots of Xes. At the same time, X Does Not Support Screenreaders Because Of Y, and there are an awful lot of Ys, too.
You might think that scanning a menu and formatting a menu in HTML are approximately equally as easy. Not so much, no. There is an actual cost here. For a restaurant menu, that is likely to only be in the couple hundreds of dollars range. (<joke>I will entertain complaints about this from people who don't bill in six minute increments.</joke>)
For more complex applications, this can -- easily -- run into the tens or hundreds of thousands of dollars. It is easy to underestimate how expensive it is because you only see the element from the set of Ys that caused non-compatibility in the present instance, not the entire set of Ys which you have to a) avoid and b) rigorously test against.
Getting down to brass tacks, the population which benefits from any particular engineering effort to address a Y is generally very small and, often, far smaller in marginal value to the business than the cost of addressing that Y. It has occasionally been the public policy of the US that businesses are supposed to subsidize that population (see: Americans with Disabilities Act, etc). Reasonable people can disagree on whether that is a justice-maximizing policy.