How Starlink Actually Works, From Flat Dish To Space Laser

Picture a Starlink Mini on a boat, giving it internet out on the water. The panel doesn't move. The satellites it talks to never stop moving. The connection has to hold while the thing carrying it flies across the sky.

How does a flat, still panel keep up with that? And why does a much bigger rocket change what this service can do for you? I'm going to walk the chain in order: satellites, dishes, factories, launch. Follow it all the way through and the long-term story gets a lot bigger than better Wi-Fi.

Why fly at 550 kilometers

One familiar Starlink orbit sits about 550 km up, roughly 340 miles. That's close for a communications satellite. A geostationary satellite, the kind that stays over one spot on Earth, sits about 35,800 km up, roughly 22,200 miles. That's about 65 times the Starlink height.

Why so close? Your signal has less distance to cover. The wait between asking for something online and getting the response is called latency, and a shorter trip through space helps cut it. On a video call, latency is that awkward delay before the other person responds.

Bandwidth is a different thing: how much data the connection carries per second. You want a quick response and enough capacity for what you're doing. Those are two separate jobs.

The price of flying low is that a satellite won't sit still in your sky. It crosses overhead, and the network hands your connection from one satellite to the next. You do nothing. The call keeps going while Starlink handles the swap underneath it.

The satellites fly in groups at similar heights called shells, with paths tilted at different angles to reach different parts of Earth. Some of the network flies lower than 550 km, which I covered in an earlier piece.

What's inside one satellite

Each of those moving connections needs a whole spacecraft behind it. The flat main body is called the bus, and it's where the hardware comes together. The flat shape also lets SpaceX stack satellites inside a rocket, which matters when you're putting up a lot of them.

Solar panels make electricity. When the satellite is out of sunlight, batteries supply stored power so the radio link keeps working through that part of the orbit.

To know which way it's facing, the satellite uses cameras that look at stars. Spinning wheels inside, called reaction wheels, control its orientation. The design has four, including a spare.

Pointing the spacecraft and changing its orbit are different jobs. For orbit, the V2 Mini satellites carry an electric engine called an argon Hall thruster, with argon as the propellant. It raises the satellite toward its working orbit and moves it when needed. At the end of its service, propulsion helps lower the orbit for retirement, and SpaceX aims that return through the atmosphere over open ocean.

Take away any one of those jobs and a radio in space never turns into an internet service.

A dish that aims without turning

Now your end. You position the Mini using guidance from the software. Once it's set up, the housing stays still while the satellites move. The steering happens electronically inside the panel.

The panel holds lots of small antenna elements working together. That arrangement is called a phased array. The name sounds complicated. The useful part is timing.

Each element sends a radio wave. The electronics adjust the timing of those waves relative to one another. In the direction you want, the waves line up and add together, creating a stronger signal aimed at the satellite. Change the timing and you change where the combined signal points. The panel keeps facing the same way while its beam follows the satellite.

Why lots of little elements? Because the useful signal is what they produce together. One element is just one part of the pattern. Controlling how their waves line up is what lets the whole array aim.

Receiving works the same way in reverse. The dish combines what its elements pick up so it listens in the direction of the satellite. Sending and receiving both run through the array.

I think that's the coolest part of the whole system. It looks like a passive slab, yet the direction of its radio link keeps changing without the panel chasing anything across the sky.

It still has to be positioned properly. The Mini's field of view is 110 degrees, wider than a right angle: that's the patch of sky it can work with from where it sits. Older Starlink dishes had motors. The Mini and the V5 home dish use manual positioning with electronic steering. The physical setup gets the panel facing the useful part of the sky, and the electronics track within that view.

The satellite runs phased arrays too. The V2 Mini, the satellite design that flies on Falcon 9, has five for its user connections, so both ends of your link are aiming radio beams. A lower orbit with the same antenna size shrinks each beam's footprint on the ground, which lets the network serve more customers in a given area. I covered that in my earlier piece on the lower shells.

From the satellite to the website

Reaching a satellite only gets you partway. The website you want lives somewhere on the internet, and your data still has to get there.

Think of it as a relay. Your dish is the local entry point. The satellite carries your traffic to the next connection. A ground station called a gateway receives the satellite's radio signal with its antennas and brings your traffic down. Starlink then connects with other internet networks at a point of presence, which is where your traffic enters the wider internet. The response comes back through the network to you, so the request and the answer both need a working route.

What happens when a satellite needs to send traffic farther across the network before it comes down? That's where the lasers come in. Satellites send data to other satellites over laser links, and your traffic can move across those links until it reaches a satellite with a suitable connection to the ground. The V2 Mini has three of these lasers, each rated for up to 200 gigabits per second. A gigabit is a billion bits, so that's 200 billion bits every second on one link.

Not every trip takes the same route. A satellite can pass traffic to a gateway or send it across laser links first. When you hear about Starlink's space lasers, that's the job they're doing.

The radio bands differ by leg. Your dish talks to the satellite on Ku band. Gateway links use bands including Ka and E. Those names describe the radio frequencies each part of the trip uses.

Ten spacecraft a day in Redmond

SpaceX builds the satellites in Redmond, Washington. Its factory video shows components being built and satellites being assembled. SpaceX makes critical machined parts and assembles circuit boards there, and it designs a chip that works across two frequency bands for the satellite's gateway connection.

During a five-month stretch ending in April 2026, the factory averaged about 70 satellites a week. That's about 10 a day. Ten spacecraft, every day.

The reaction wheels' spinning disks are aluminum. SpaceX makes a mirror film that directs reflected sunlight away from the ground, plus dark paint that reduces reflections. Sunlight hits the spacecraft, and its surface decides how much of that light heads toward people on the ground. Even the coating has a job.

SpaceX also designs components to break up during re-entry. The end of a satellite's life gets decided on the factory floor before it ever flies.

The dish factory in Bastrop

The other end of your connection comes out of Bastrop, Texas, the user terminal factory. The dish is a manufactured electronics product, and its cost affects how many people can get onto the network. Building it at scale is a huge part of building Starlink.

SpaceX's Bastrop video shows plastic and aluminum becoming terminal hardware and finished kits, through injection molding for the plastic and die casting for the metal. The case is only the outside. Inside, the antenna hardware connects to circuit boards and radio chips, and those chips make the electronically steered beam possible. SpaceX co-designs them with STMicroelectronics. The silicon gets fabricated, packaged and tested before it goes into the hardware that sends and receives your data. In the Mini, even the Wi-Fi router sits inside the terminal, so one small unit handles the satellite link and your local Wi-Fi.

By December 2025, Starlink said it was producing more than 20,000 terminals a day across all its sites. Over a seven-day week, that's more than 140,000.

The cost curve moved just as hard. In 2021, SpaceX had cut the cost of building a terminal from about $3,000 to under $1,500. By 2023, it had stopped subsidizing terminals. That matters for growth: adding a customer no longer needs a hardware subsidy, so each new person costs less to bring onto the network. The network has to grow at both ends, and this is the ground end keeping pace.

The ride up

Falcon 9 does most of the launching. In footage of the first Starlink deployment, you can see the stack spreading out as the satellites leave the rocket. They aren't ready for customers yet. Each one uses its own propulsion to climb toward its working orbit, so the rocket and the satellites split the trip.

Weight decides how many fit. The optimized V2 Mini is about 575 kg, roughly 1,270 pounds, nearly 22% lighter than the original V2 Mini. That allows as many as 29 on a Falcon 9, six more than the original version. Same launch, more machines providing service.

The V3 jump

Starship lets SpaceX design around a much bigger ride. SpaceX says its V3 broadband satellite is designed for a terabit per second of download capacity, about 10 times the V2 design. Does that mean a terabit for your house? No. That capacity is shared across all the connections the satellite serves.

More capacity changes the hardware around it. SpaceX says V3's solar arrays are designed for about twice the power of the V2 arrays, and each V3 carries six lasers designed for 400 gigabits per second apiece, twice the earlier link rate. The upgrade widens the links across space as well as the links down to users.

SpaceX plans full Starship loads of up to about 60 V3 satellites. On design numbers, that's about 60 terabits per second of download capacity in one trip, roughly 20 Falcon 9 Starlink launches' worth. I covered the launch math in an earlier piece.

Has Starship actually carried them? Yes. Flight 14 put the first 26 V3 satellites into orbit. In the deployment footage they come out through the ship's slot, and one satellite's camera looks back at Starship as it pulls away. The ship fired an engine again to enter orbit, then released them. SpaceX made contact with all of them through laser links, ground stations and user terminals, the same chain this piece just walked through. Next came checks and orbit raising before customer service.

Who pays for it

Starlink went from 12 million to 13.6 million consumer subscribers in the third quarter of 2026, another 1.6 million in three months. SpaceX's connectivity business also serves enterprise and government customers on aircraft, on ships and at fixed sites.

In the first nine months of 2026, connectivity made about $5.2 billion in operating profit, more than the combined operating losses of SpaceX's space and AI businesses over the same stretch. My read: connectivity generates cash while the rocket program builds the means to add capacity, and Starship carries the satellites that expand what connectivity can sell. The internet business helps pay for the rocket.

Phones, then machines

The other path onto the network skips the dish. Direct to Cell takes the satellite link straight to a compatible phone over mobile frequencies, and I covered that side in two earlier pieces. In PCMag's field test, a Pixel 10 made a WhatsApp call over T-Mobile's T-Satellite. V3 is the broadband family; V2 mobile is the phone family. SpaceX plans to begin deploying V2 mobile on Starship in 2027, targeting more than 100 times the bandwidth of the first mobile system.

Who uses all this bandwidth? I made the machine-traffic case before, so one line: Elon Musk expects AI agents to send a thousand times more data per second than all humans within a few years. The view I laid out in the video is that physical AI, including humanoid robots, makes that demand much bigger, and that Starlink's commercial business becomes much bigger than its consumer business.

That's why I think of Starlink as a nervous system for AI. The computers do the thinking. The network carries information between them and the devices using them.

It ties straight into the companies. Terafab is the SpaceX and Tesla chip plan, covering chips for Optimus, Cybercab and data centers in space. SpaceX plans to put AI computers in orbit and connect them with lasers. Computers up there need a way to exchange information and get results down to people and devices. The chip supplies the computing, the network supplies the connection, and Starship is how SpaceX plans to carry more of that hardware up.

Two hypotheticals from the video. Your business runs a humanoid robot at a remote construction site, and updated plans have to reach the site while results come back. Or an AI system helps run a mission to Mars, sending information home and taking instructions from Earth. Either way you need a communication system built for the job. NASA has selected Blue Origin to provide a Mars telecommunications network. The longer-term Starlink idea follows the same need: extend the connections alongside the intelligence doing the work.

Back to the boat

Go back to that little panel. For the person using it, the result is simple: a connection out on the water. Behind it sit star-watching cameras, argon thrusters, lasers rated in the hundreds of gigabits per second, chips co-designed with STMicroelectronics, and factories building roughly 10 satellites and more than 20,000 dishes a day.

My read is that the long-term goal is getting intelligence connected wherever there's useful work for it to do, and building enough bandwidth is part of getting there. That panel on the boat is the piece of it you can already see.