Hmmm... not so sure about that.
Latency for geosync satellites is atrocious...at about 750ms for the bounce to the bird 24K miles up. Of course Starlink sats are supposed to be very low-earth-orbit, at somewhere in the neighborhood of 750 miles up, but that implies a round trip distance of more than 1500 miles added to the packet latency in the best case. Even if it's a coast-to-coast session, that's going to be something like 40% of the duration. If your session is to the nearest cache point 10-1000 miles away the latency will be far worse than fiber.
I suppose trans oceanic or really long haul sessions might start to show some advantage, which might help the developing countries scenario.
For certain routes Starlink can have lower latency than ground based fiber. Speed of light in a vacuum versus glass (or anything else not a vacuum, to include the atmosphere). Of course the ground to sat and sat to ground will be going through atmosphere, but sat to sat is actual speed of light. The longer the route, the bigger the advantage in speed of light, so over a certain threshold it will be faster than fiber.
Possibly fewer router hops depending on where point A and point B are, but I wouldn't assume that to be the case.
I don't recall if this depends on the VLEO constellation to reach better-than-fiber latencies though. Tintin A&B have been quoted as having a 25ms latency - and even if that's just ground-sat-ground and not ground-sat-sat-ground, that's pretty good. They're not exactly in either a LEO or VLEO orbit at the moment, closer to VLEO, but 25ms is nothing to sneeze at, especially considering that a home user might be lucky to get that over a physical distance of tens of kms rather than hundreds.
Of course, it won't likely be too useful in highly built up areas, so AT&T and friends don't really have too much to worry about in their denser areas. But don't be surprised if AT&T and so on are using Starlink to backhaul whatever-G cellular in the future, rather than running ever bigger fiber connections.
Starlink basically has two markets : the underserved (developing world, rural areas, etc, where oversaturation of user terminals isn't an issue), and aggregate backhaul (for cellular tower to telco's network, datacenter to datacenter, datecenter to last mile, etc). Actual "last mile" service in well served areas won't be as big of a market since there's a limit to how many user terminals can exist in one area (even with a zillion spot beams). They might still sell such "last mile" service but it'd end up having to be first come, first serve.
While I wouldn't be surprised to some day see Starlink terminals built into Tesla vehicles some day, they would only really be useful when not in cellular coverage.