Satellite internet used to come with an asterisk, and the asterisk was physics. The new generation removed the asterisk by moving the satellites - and everything interesting about LEO - Low Earth Orbit - constellations follows from that one decision. This is the fourth last mile, and it works nothing like the first three.
The geometry problem GEO couldn't escape
A geostationary satellite orbits at 35,786 km above the equator, where its period matches Earth's rotation - it hangs motionless in the sky, so a dish points once and never moves, and three satellites can see almost the whole planet. The price is the distance itself: radio travels at light speed, and up-and-down twice (query out, answer back) across that altitude costs roughly half a second of round-trip time before any network equipment adds a microsecond. For broadcast television, irrelevant; for TCP handshakes, TLS negotiations, and anything interactive, structural. GEO internet was always fighting its own altitude.
LEO's inversion, and the engineering it forces
LEO constellations - Starlink being the operating example at scale - place satellites at a few hundred kilometers instead, collapsing propagation delay into the tens of milliseconds, competitive with terrestrial paths. But a satellite that low cannot hang motionless: it crosses the sky in minutes, so coverage now requires thousands of satellites in coordinated orbital shells, and every property that GEO made trivial becomes engineering. The ground terminal becomes a phased-array antenna - steering its beam electronically to track satellites and hop between them with no moving dish. Sessions survive constant handoffs as satellites rise and set. And routing gains a third dimension: modern constellations carry inter-satellite laser links, letting traffic travel between satellites in vacuum and come down near its destination - which is why a traceroute through such a network shows a first hop that is, delightfully, in space.
The honest trade-offs
The engineering candor this site owes every technology: LEO's price list is real. Thousands of satellites must be launched, replaced on short lifetimes (atmospheric drag at low altitude is a feature for debris management and a cost for operators), and deorbited responsibly; astronomers document the streaks; the terminal is a small power-hungry computer rather than a passive dish; and capacity is shared per cell of sky, so density of subscribers matters the way it does on a DOCSIS neighborhood node. None of this is disqualifying - it is the shape of the bargain: latency bought back from physics by fleet logistics.
Where it actually fits
The design consequence is placement. Where fiber exists, fiber wins on capacity and stability. Where nothing exists - rural areas, oceans, disaster zones, the mobile and the remote - a delivers something previously impossible: low-latency broadband with no local infrastructure beyond the terminal. And in the multi-path household this series keeps returning to, it earns a specific job: the independent path - sharing no duct, no pole, no headend, and no failure domain with the terrestrial lines it backs up.