Selfish reason for not building accessible UIs: it's extra work that doesn't justify the cost compared to the market share.
We would all be better off spending that investment on improving the dozen or so screen readers instead of billions of web sites. We aren't far from an LLM powered screen reader making aria tags moot.
It takes the same amount of work to use an <aside> or <button> as a <div>. I have never accepted the argument that accessibility is more work. It is the same amount of work but done correctly. Aria attributes are only necessary when extending HTML semantics.
In the best (and thankfully pretty common) cases the work to do the accessible answer can be the same, or even less, than the work to do it otherwise. That certainly doesn't extend to all cases though and that's where it can snag. E.g. making complex charts or visualizations accessible to those using screen readers often involves creating a second version of the information only suited for that modality. A ton of other things are a bit in-between, such as prefers-reduced-motion being somewhere between "pretty easy" to "complex" depending on how you implement it in your site but, either case, adding some amount of effort.
Complex cases are indeed complex! Data viz is an interesting example since visuals are often misleading and time sinks regardless of ability. It's less effort to provide an HTML table or CSV export than gluing D3 and React together (or whatever stack your team uses).
I try my hardest to push back on canvas viz so we're not reinventing every wheel. SVG is easy to extend and make interactive
We would all be better off spending that investment on improving the dozen or so screen readers instead of billions of web sites. We aren't far from an LLM powered screen reader making aria tags moot.