First, and less importantly, source. Because phone companies love to toss out coverage numbers that are pure, unadulterated bullshit. But second that's less important because 3G and 4G are both completely uncompetitive for broadband. LTE is basically the minimum there. And awful data caps and congestion remain a serious issue all over the place.

>A fiber PoP / tower has much lower OPEX than launching, operating a constellation of satellites.

Is it? I mean, seriously. A lot of SpaceX's plans are based on estimated launch costs via Super Heavy, so necessarily somewhat speculative. But the target is sub-$100/kg, and at least in terms of fundamentals (like cost of fuel, Raptor effectiveness, estimated cadence) that doesn't look ridiculous. Each Starlink satellite apparently has a mass over around 230 kg. Add a bit of extra mass for some bigger sats in the mix, double the launch cost to be a bit more conservative, and we're still talking a launch cost per satellite of well under $60k a piece. They necessarily have everything needed to operate their design lifetime, because there is no maintenance that can be done on them. They're operating low enough that solar weather shouldn't be too constant a concern, and of course will never face OPEX costs like vegetation clearing or weather damage or corrosion or the like. Also doesn't face NIMBY tower politics either. Is it actually more expensive then running fiber and building/maintaining a tower? For very high density locations probably, but there's a pretty sizable population in the US alone for whom it seems on napkin math it could be very competitive.