Perhaps worth remembering that they don't have to engineer the final system right now. Consider that:
- Building ground station gateways costs money, but as the number
of subscribers grows, so does the income stream that allows for
ground station construction.
- Gateway ground stations, if generally kept close to users, at
least where there are lots of users, don't have the problem that
there's no land to put them on as there aren't normally lots of
users in the middle of the sea.
- Where there are more than 10k users, someone will generally put
a fibre optic cable there or at least think hard about how they
can get that done. We've seen this widely in the Pacific, where
such islands now generally have fibre (with some notable
exceptions). Once you have fibre, there's no reason why you can't
have a LEO ground station there. So you can generally always build
more
- Constellations that are short of capacity can be added to
incrementally. Remember Starlink is at a very early stage of what
they're planning.
- Satellite capacity depends on a lot of things, from on-board
processing capacity to the bandwidth of the uplinks and downlinks
and the received signal levels on those. Current Starlink gateway
antennas are tiny by satellite infrastructure standards, but that
doesn't preclude bigger antennas and more gateways for the future,
which allows more satellites to be provided with direct gateway
links.
In some ways, this evolution mirrors that of mobile networks. In
the early stage of mobile network evolution, the providers aimed
at getting coverage with the least number of base stations from
the highest hills and buildings available. These never had the
potential to provide the capacity that today's networks have, so
as the networks evolved, they moved down from the lofty heights to
the bottom of the valleys, cashing in on improved frequency re-use
potential (you could re-use the same frequency a valley over, you
see, because the ridge between yours and theirs meant that signals
wouldn't interfere). It also meant less path loss to the end users
(so more battery life for them). Not to mention fewer lawsuits
from people who were worried about large towers irradiating their
kids. The cost of those lawsuits, by the way, was seen simply as
something to factor in when making engineering decisions.