There's also a very very "fat head" in front of the long tail of what customers pay FB et al for their ads.
https://www.wordstream.com/articles/most-expensive-keywords
Suggests the 10 most expensive categories to advertise in are dominated by insurance, financial, and legal services.
I suspect when that curve is overlaid with the similar long tail curve of the disposable income of FB's 2 billion users - I suspect it'd be obvious that the vast majority of FBs income comes from super expensive PPC ads shown to people with a (globally) super high disposable income.
If you let the bay area tech crowd who're struggling to find a place to rent on their own because they're "only" making $160k all opt out of getting profiles and shown advertising for $25 or $50 per month, you'd probably make a existential dent in FB's profitability - much more so than if they just turned advertising off altogether for the lowest earning billion and a half people on their platform.
The minority of people on Facebook that click on ads and buy things, are making it possible for the other 2 billion people to use the service without paying a monthly fee.
This distribution is identical to advertising pretty much everywhere else throughout history.
Everyone that drives by a billboard can't afford to buy what the billboard is selling. That doesn't make it a con. The buyers you do derive, out of the very large number of people that have no interest or can't afford what you're selling, pay for your billboard (hopefully).
Everyone that watches an ad on TV - the past half century - can't afford to buy the thing being sold (a $1,000 workout machine), that doesn't mean the advertising is a con.
The sole difference is the heightened targeting based on information that Facebook or Google know about your habits or interests. In the past, it was far more difficult. A TV advertiser for example, might discover through research and trial & error that TV watchers are more likely to buy that $1,000 workout machine at 1am via an hour long infomercial. The experimentation loop was dramatically slower.