I care little about what the product/impact is and mostly care about how interesting the work is on a technical level. I am pretty fond of theoretical research even when practical usage is very unlikely and did a math major mostly for fun. One thing the companies you listed share is large amounts of data and as I work in ML having large datasets is really nice to play with. Also large tech companies tend to have better data infrastructure and more time can be spent focused on ml tasks vs data engineering tasks. What the model actually does is of little interest to me and I've worked in a couple different ml areas now.

I do have an eventual interest to do research focused role, but interesting impactful research I think requires either high management (professor leading a large lab/research director) or making your own thing. The first I lack the experience for and the second doing a startup feels too risky to me today. If I was wealthy enough I would probably try making a research themed ai startup.

It depends on whether you care about the overall product or just the puzzles involved in the product.

I get my overall high level fix from outside of work activities, so work is enjoyable mostly for the little puzzles it generates every day. I am sure that there are some fascinating considerations for a mega e-billing system, even if it isn't all that interesting in itself.

That's the thing, it is something companies have control over, if they have a dynamic and passionate CEO, and are willing to take on risk. Or at least, if it's a big established company, create pockets where groups are able to work on entirely new initiatives.