Just to comment on one part of the previous post. I think @bxr140 implied this, but for benefit of others I'm going to spell it out
Latency is an issue, but its about a quarter second round trip, not one way.
the latency to
geostationary satellites is indeed about 1/4 second at best. This is because the geostationary satellites are on ~35700 km (about 22K miles) orbit above equator. So, to reach them one way, light takes slightly longer than 1/10 second (exact delay depends on the client station location). Then the signal needs to get to the ground station (another 1/10th), then routed to the server (depends how far the server is from the ground station). The server response comes back via the same channel. In the very
best case the signal covers 4 * 35700 km, which means the very best round trip time possible due to the speed of light is
1/2 second.
Some satellite internet providers cut this time in half by using dialup for upstream. So, the first part of the process I described above, doesn't go to space, but instead is routed through the ground-based phone network. Since most of the traffic is usually on the down leg, the slow dialup connection is used to send a small request (which works fast even via dialup), but the wide satellite connection is used to download pages, pictures and movies. This cuts the round-trip time in half. This is why the satellite internet is usually thought of as 1/4 sec roundtrip. Compare it with ~1/15 sec round-trip up and down Eastern or Western seaboard and up to 1/10 sec cross-country ping.
All the above applies to geostationary satellites. What Elon is proposing is much-much lower orbit constellation. With 4000 satellites on 1200km orbits there's always one overhead and the light flight time to it is just 1/250 sec. If the constellation does the routing without help of the ground stations (that's an IF), then we only pay ~4/250 sec (about 16ms) per roundtrip for using this network as opposed to the ground-based network. However, this network is MUCH faster: it's medium is vacuum and the signal always goes straight (as opposed to follows the route of the cable on the ground, which is never straight). I think this network architecture has potential to be faster than existing internet. This is going to be very big. Routing is a very interesting problem to solve in this case. And security, too. And stationkeeping.