Your dish sends data to a satellite, which relays it to a ground station connected to the internet; replies come back the same way.
Satellite internet has three parts: a dish at your home, a satellite, and a gateway ground station plugged into the fibre internet. When you open a web page, your request goes up to the satellite, down to the gateway, out to the website, and the answer returns along the same path.
The big design choice is how high the satellite flies.
A geostationary satellite at 35,786 km covers a huge area and stays fixed, so a home dish can point at it once. India's GSAT-N2 works this way. It is a Ka-band high-throughput satellite with 32 user beams and a capacity of about 48 Gbps. ISRO says it is meant to improve broadband and in-flight internet across India. It was launched on 19 November 2024 on a Falcon 9 rocket from Cape Canaveral, Florida, and then drifted towards its assigned slot at 68° East.
The drawback is delay. Every request makes four trips of about 36,000 km: up and down to reach the gateway, up and down to come back. That adds roughly half a second, which you notice on video calls.
A satellite in low Earth orbit, under 2,000 km, is far closer, so the delay is small. But it sees only a small patch of Earth and crosses the sky in minutes. To give non-stop coverage you need very many satellites, each handing your connection to the next.
That is why SpaceX's Starlink is so large. Jonathan McDowell's count on 2 November 2026 showed about 11,200 working Starlink satellites. In India, OneWeb and Reliance Jio received satellite-communication approval from the Department of Telecommunications before Starlink did.
Look at how densely the satellites cover the globe compared with the handful of geostationary satellites on the ISRO page. Watch the Starlink constellation
The delay you feel is set by the speed of light and the path length.
Geostationary. For a user near the point under the satellite, one round trip (request and reply) crosses the 35,786 km gap four times:
path = 4 × 35,786 = 143,144 km delay = 143,144 / 299,792 ≈ 0.48 s
Add processing and the terrestrial internet, and a geostationary link rarely feels quicker than about half a second.
Low orbit. Take a satellite directly overhead at 550 km. The same four crossings are 4 × 550 = 2,200 km, which light covers in 2,200 / 299,792 ≈ 0.0073 s, or about 7 ms. Real links are longer because the satellite is usually not overhead and the gateway may be far away, but the physics advantage is a factor of about 65.
Why so many satellites? From height h, the horizon is at an Earth-centred angle θ = arccos(R / (R + h)). For h = 550 km, θ = arccos(6,378 / 6,928) = arccos(0.9206) ≈ 23°. That is the absolute edge, where the satellite sits on the horizon; useful service needs the satellite well above the horizon, so the real footprint is much smaller. For a geostationary satellite, θ = arccos(6,378 / 42,164) ≈ 81°, which is why one satellite can see almost half the Earth. Covering the whole globe from 550 km therefore takes thousands of satellites instead of three.
The signal must travel about 35,786 km up and down twice for each request and reply, adding roughly half a second; LEO satellites are only hundreds of kilometres up.
Each low satellite sees only a small area and passes overhead in minutes, so many are needed to keep one always in view.
A ground station that links the satellite to the terrestrial internet.