How Starlink Actually Works: Satellites, Lasers, and Ground Stations Explained

Why Starlink is fundamentally different

For two decades, “satellite internet” meant one thing: a massive spacecraft parked in geostationary orbit at 35,786 km above the equator, serving an entire continent from a single point in space. HughesNet and Viasat still operate this way. The physics are simple and brutal: a radio signal traveling at the speed of light needs roughly 240 ms for the one-way trip to geostationary altitude and back, which translates to a round-trip latency of 600–800 ms once you add ground routing. That is why video calls over HughesNet feel like talking to someone on Mars.

Starlink threw out the entire model. Instead of one satellite very far away, it uses thousandsof small satellites very close — orbiting at just 540–570 km. That single architectural decision, reducing altitude by roughly 65x, cuts the speed-of-light delay from 240 ms one-way to about 3.6 ms one-way. The result is a round-trip latency of 20–60 msin practice, which is 10–20x better than legacy satellite and close enough to terrestrial cable internet that most users cannot tell the difference in daily use.

But low orbit creates a new problem: a satellite at 550 km moves across the sky in about 4–6 minutes instead of sitting still. That means you need a constellation— enough satellites so there is always one overhead, no matter where you are on Earth. And that requirement cascades into every other design choice: the phased-array dish, the ground station network, the laser inter-satellite links, and the software that orchestrates handoffs between all of them multiple times per minute. Understanding each piece explains why your Starlink speed is what it is — and why it varies.

The constellation: 6,000+ satellites in coordinated shells

As of mid-2026, SpaceX has launched over 6,000 Starlink satellites, of which roughly 4,500 are operational. The rest are either raising their orbit after a recent launch, undergoing controlled deorbit at end of life, or being held as on-orbit spares. The constellation is organized into orbital shells— rings of satellites at specific altitudes and inclinations:

• Shell 1 (~540 km, 53° inclination):the largest and earliest shell, covering latitudes between roughly 53°N and 53°S. This is where most V1 and V1.5 satellites operate.
• Shell 2 (~570 km, 70°): extends coverage into higher latitudes, reaching Alaska, Scandinavia, and southern Chile.
• Shell 3 (~560 km, 97.6° polar): polar orbits that fill in coverage at extreme latitudes, including Arctic shipping routes and Antarctic research stations.

Each satellite covers a ground area called a “cell” roughly 15 miles (25 km) across. Within each cell, all subscribers share the satellite's available bandwidth. This is why cell congestion is the single biggest factor determining your speed at any given moment — a rural cell with 50 users and a suburban cell with 2,000 users might be served by the same satellite, but the per-user bandwidth is radically different.

There are two main satellite generations in the constellation today:

V1.5 satellitesweigh about 306 kg, use hall-effect thrusters for station-keeping, and communicate with ground stations via Ku-band and Ka-band radio. They do nothave laser inter-satellite links — each satellite must be in line of sight of a ground station to route traffic. A V1.5 satellite has a throughput capacity of roughly 17–23 Gbps shared across all users in its coverage area.

V2 Mini satellitesare the current generation being launched. They weigh about 800 kg (requiring SpaceX to reduce from 60 to 23 satellites per Falcon 9 launch), but carry roughly 4x the throughput capacity of V1.5. Critically, V2 Mini satellites include laser inter-satellite links that allow them to relay data between each other in orbit without touching the ground. This is a generational leap that reshapes the entire network's architecture.

You can watch the constellation in real time on starlink.sx, which tracks every satellite's position, altitude, and operational status. If you want to spot them visually, findstarlink.com predicts visible passes for your location.

Laser inter-satellite links: the space backbone

This is the piece of Starlink's architecture that has the most far-reaching consequences, and the one most people underestimate.

Each V2 Mini satellite carries multiple laser terminals that establish optical point-to-point links with neighboring satellites. These lasers operate in near-infrared wavelengths and can sustain data rates above 100 Gbps per link. A satellite typically maintains links with four neighbors: two in the same orbital plane (one ahead, one behind) and two in adjacent planes (one left, one right). This creates a mesh network in space — a backbone that can route data across thousands of kilometers without ever touching the ground.

Why does this matter? Two reasons that compound on each other:

First, it eliminates the ground-station dependency. Without laser links, a V1.5 satellite must be in direct line of sight of a ground station to route your traffic. If there is no ground station within range (over oceans, remote regions, or countries where SpaceX lacks regulatory approval for gateways), that satellite cannot serve you. With laser links, a V2 Mini satellite over the middle of the Atlantic can relay your traffic hop by hop through space to a ground station on the US East Coast or in Portugal. This is what enables Starlink service over open ocean for maritime customers and in countries where SpaceX has no local gateway.

Second, laser links can actually beat fiber on latency for long-distance routes. This sounds counterintuitive, but the physics are real. Light travels through the vacuum of space at 299,792 km/s. Light traveling through a glass fiber optic cable moves at roughly 200,000 km/s(about 67% of vacuum speed, because of the glass's refractive index). That means a photon in space is roughly 47% fasterthan a photon in fiber. On short routes like New York to Chicago (1,100 km), the difference is negligible. But on intercontinental routes like London to Tokyo (9,500 km), the accumulated speed advantage plus the ability to route in a straighter great-circle path (instead of following undersea cable routes that detour around continents) can give Starlink a real latency advantage over terrestrial fiber. This is already attracting interest from financial trading firms where single-digit millisecond advantages are worth billions.

You can track which satellites overhead have active laser links through community tools like starlinkstatus.space. As the constellation shifts from V1.5 to V2 Mini, the percentage of laser-equipped satellites grows with every launch. By late 2026, the majority of operational satellites will have laser capability.

Ground stations: where space meets the internet

Starlink operates over 200 ground stations (SpaceX calls them “gateways”) across six continents. Each gateway is a cluster of large radome-enclosed antennas, typically 4–8 dishes, that communicate with passing satellites overhead via Ka-band radio. The gateway connects to the terrestrial internet backbone through fiber optic lines to major internet exchange points (IXPs) and peering partners like Google, Netflix, and Cloudflare.

The signal path without laser links works like this: your dish transmits to a satellite overhead, that satellite relays the signal down to the nearest ground station, the ground station routes it onto the public internet, the response travels back through the ground station, up to a satellite, and down to your dish. The whole round trip takes 20–60 ms depending on geometry and congestion.

Gateway density directly determines total available bandwidth in a region. The United Stateshas the densest gateway network — over 50 stations, concentrated in the northern states where orbital geometry provides the best satellite coverage. Europe has roughly 30+, with major clusters in France, Germany, and the UK. Regions with fewer gateways (sub-Saharan Africa, Southeast Asia) rely more heavily on laser inter-satellite links to reach distant gateways, which adds a few milliseconds of latency but still stays well under 100 ms.

Each gateway can serve multiple satellites simultaneously as they pass overhead. A single well-connected gateway with 8 antennas can handle roughly 40–80 Gbps of aggregate throughput. When SpaceX lights up a new gateway in an underserved area, users in that region see an immediate improvement in speed because the traffic no longer needs to route through a distant gateway.

You can see gateway coverage relative to your location on our coverage map, which shows satellite footprints and ground station proximity.

Your dish: a phased-array antenna with no moving parts

The Starlink user terminal — the “dish” or “Dishy” — is the most sophisticated piece of consumer electronics most people will ever own, and it looks like a plain white rectangle. Inside is a phased-array antenna containing over a thousand individual antenna elements that work together to electronically steer a radio beam toward whatever satellite is currently overhead.

Traditional satellite dishes (like the ones used for DirecTV or old HughesNet) are parabolic reflectors that must be physically pointed at a single, stationary satellite. They have motors to adjust aim. Starlink's phased array has no moving parts at all in the Gen 3 design. Instead, it adjusts the timing (phase) of the signal across its antenna elements to electronically sweep its beam across the sky. It can switch from one satellite to another in milliseconds, and it does this continuously — handing off to a new satellite every 15–90 seconds as each one transits the visible sky and the next one rises.

The dish requires a clear view of the sky spanning roughly 100° field of view(a cone from about 25° above the horizon in all directions to straight overhead). Anything blocking that cone — trees, buildings, chimneys — reduces the pool of satellites the dish can see, which means fewer handoff options, more interruptions during transitions, and lower effective throughput. This is why our obstruction planner exists: even a 5% obstruction in the wrong part of the sky can cause disproportionate connection drops.

The Gen 3 Standard dish draws roughly 50–75W in normal operation and up to 150Wwhen the snow-melt heater is active. It performs an initial sky scan on boot to map obstructions, aligns itself automatically (earlier Gen 1 dishes had a motor for initial positioning; Gen 3 skips this entirely), and begins tracking satellites within 2–5 minutes of power-on. The dish communicates with satellites using Ku-band(12–18 GHz) for user traffic.

Inside your home, the dish connects to the Starlink router (or your own router in bypass mode) via a proprietary cable. The router provides WiFi 6 (Gen 3) or WiFi 6E (Gen 3 Premium) and a single Ethernet port (with an optional adapter for older models). If you are testing Starlink performance, always test wired via Ethernet first — use our speed test to separate dish performance from WiFi limitations.

Why your Starlink speed varies

Knowing how the system works explains every speed variation you will ever experience. There is no mystery — each factor maps directly to a piece of the architecture:

Cell congestion (the biggest factor).Every satellite allocates bandwidth across its ground cells. When more users are active in your cell, each user gets less. A cell with 100 active users at noon might deliver 200 Mbps per user; the same cell at 9pm with 800 active users might deliver 50 Mbps. This is exactly the same dynamic as a cell tower — shared wireless capacity divided among concurrent users.

Time of day.Peak hours (7pm–11pm local) drive congestion up because residential usage concentrates in the evening. Off-peak hours (midnight–6am, midday) typically deliver the fastest speeds. If you run a speed test only at 9pm and see 60 Mbps, you are sampling the worst-case window.

V1.5 vs V2 satellite overhead. If the satellite currently serving you is a V1.5 unit, your maximum potential speed is lower than if a V2 Mini is overhead. The constellation is mixed right now, with V1.5 satellites being gradually replaced. As V2 Mini density increases, baseline speeds in most cells will rise.

Gateway proximity.If no ground station is within direct line of sight of your serving satellite, your traffic must hop through laser links to reach a distant gateway. Each hop adds latency (typically 1–3 ms per hop). Users near dense gateway clusters (northern US, Western Europe) generally see lower ping than users in gateway-sparse regions.

Weather.Heavy rain causes signal attenuation (“rain fade”), typically 10–30% throughput reduction. Snow on the dish, if the heater cannot keep up, blocks signal entirely. Extreme heat triggers thermal throttling.

Obstructions.Trees, buildings, or terrain blocking part of the dish's 100° field of view reduces available satellites and forces more frequent handoffs. Every handoff is a potential brief dropout. More obstructions mean more dropouts and lower average throughput.

Starlink LEO vs legacy satellite vs fiber

The architecture differences between satellite systems and terrestrial fiber create fundamentally different performance profiles. Here is how they compare on the numbers that actually matter:

The table makes the tradeoffs clear. Starlink loses to fiber on raw speed and latency but wins on availability — you can get it virtually anywhere with a view of the sky. It crushes legacy satellite so thoroughly on latency that they are not in the same category for interactive use. The practical question for most buyers is not “is Starlink as good as fiber?” but “is fiber available at my address?” If the answer is no, Starlink is the best option that exists.

The full signal path: putting it all together

Here is what happens when you click a link on your Starlink connection, start to finish:

1.Your device sends the request over WiFi (or Ethernet) to the Starlink router, which forwards it to the dish via the proprietary cable. Time: under 1 ms.

2.The dish's phased-array antenna encodes the request into a Ku-band radio signal and transmits it to the Starlink satellite currently being tracked overhead at 540–570 km. One-way travel time: about 3–4 ms.

3a. If the satellite has direct line of sight to a ground station, it relays the signal down to that gateway via Ka-band. One more hop of 3–4 ms.

3b. If no ground station is in range (ocean, remote area), the satellite uses its laser links to relay the signal through one or more neighboring satellites until it reaches one that can see a gateway. Each laser hop adds roughly 1–3 ms.

4. The ground station receives the signal, converts it to standard internet protocol, and routes it onto the fiber backbone to the destination server (Google, Netflix, your company VPN, etc.). Backbone routing takes 5–30 ms depending on the distance to the server.

5. The response travels back through the same chain in reverse: backbone to gateway, gateway to satellite, satellite (potentially via laser hops) to the satellite above your dish, satellite to dish, dish to router, router to your device.

Total round trip: 20–60 ms typically. This is the ping number you see in a speed test, and it maps directly to the physical architecture described throughout this article. Every millisecond is accounted for by the physics of the signal path.

FAQ