I'm on a flight somewhere west of Ireland, and the Wi-Fi is actually working. I know that shouldn't be remarkable, but here we are.
It's remarkable when you're over the Atlantic and you're not staring at a loading spinner.
And that brings us to what Daniel sent in. He's on an Emirates flight from Dubai to the USA. The crew announced Starlink shortly after takeoff. He says it's the best in-flight Wi-Fi he's ever used, period. Right now he's over the oceanic leg where traditional satellite internet tends to fail, especially up at the more northerly latitudes, and it's holding steady. He doesn't have a terminal to measure latency — even sys admins take breaks — but he can feel the difference. His questions are pretty direct. First, how has Starlink provided superior connectivity compared to traditional satellite internet? And second, what unique advantages does it bring to serving a fast-moving receiver at thirty-seven thousand feet?
Two excellent questions, and they're related but distinct. The first is about the architecture. The second is about the receiver. Let's dig in.
So let's dig into the physics and engineering of what's actually happening above that aircraft.
The core shift here is that traditional in-flight Wi-Fi is what engineers call a bent pipe. The aircraft has an antenna that points at a single geostationary satellite parked at about thirty-five thousand, seven hundred eighty-six kilometers above the equator. The signal goes up from the plane to that satellite, the satellite bounces it down to a ground station, and the reply comes back the same way. The latency is dominated by the round-trip distance, not the bandwidth. That distance is roughly seventy-two thousand kilometers just for the space portion. At the speed of light, that's a minimum of two hundred forty milliseconds before any processing overhead. Real-world, you're often looking at six to eight hundred milliseconds.
So when you click a link, your packet is taking a half-second vacation to space and back before anything even starts loading.
And that's before the server responds. Add another half-second for the reply. You're at over a second for a round trip. That's why traditional in-flight Wi-Fi feels like dial-up even when the bandwidth is technically adequate. TCP, the protocol that underpins most web traffic, interprets that latency as congestion and throttles itself. You can have all the bandwidth in the world and TCP will still slow to a crawl because it thinks the network is clogged.
That's the part most people miss. They assume it's slow because there's not enough pipe. The pipe's fine. The problem is the pipe is too long.
Right. Starlink flips the entire architecture. Instead of one satellite at thirty-five thousand kilometers, you've got a constellation of thousands of satellites at roughly five hundred fifty kilometers. The round-trip distance to the satellite is about eleven hundred kilometers, which at light speed is about four milliseconds. Four. The real-world latency Daniel is experiencing is twenty to forty milliseconds, and most of that is ground-side routing and processing. The space portion is negligible.
That's the difference between "this is impressive for a plane" and "I forgot I was on a plane."
And it's not just latency. The throughput story is completely different too. A traditional in-flight system shares a single GEO transponder across the entire aircraft. Often across multiple aircraft on the same beam. You've got three hundred people on a 777 all competing for a slice of one transponder that's also serving other planes. Starlink's constellation has thousands of satellites, each with multiple phased-array beams. Each aircraft gets a dedicated, steerable beam. No sharing, no contention with the plane next to you.
So it's the difference between a party line and a direct connection.
And Daniel mentioned something specific that's worth pulling on. He said he's on the oceanic leg where traditional satellite internet tends to fail, especially at more northerly latitudes. That's not a coincidence. GEO satellites are parked over the equator. If you're flying a polar-adjacent route — Dubai to the US west coast, say — the satellite is very low on the horizon. The signal has to travel through much more atmosphere at a shallow angle, which attenuates it. And the distance is actually longer because of the angle. LEO satellites pass directly overhead at all latitudes, including polar routes. The coverage is uniform.
So the geometry of the constellation itself solves the polar problem.
And there's another piece that makes oceanic coverage possible at all. Inter-satellite laser links. Traditional satellite internet requires a ground station within the satellite's footprint. In the middle of the Atlantic, there are no ground stations. With a GEO satellite, that's fine if the satellite can see both the aircraft and a ground station on either side of the ocean. But with LEO, a single satellite's footprint is much smaller — a few hundred kilometers across. If that satellite can't see a ground station, it can't relay the traffic. Unless it can talk to other satellites.
Which is what the laser links do.
Right. Starlink satellites have optical inter-satellite links — lasers that connect each satellite to its neighbors in the constellation. Your data hops from the aircraft to the satellite overhead, then across the mesh from satellite to satellite, until it reaches one that's over a ground station. The entire Atlantic crossing, the aircraft never needs to be in view of a ground station. The constellation routes the traffic internally.
So the network is doing the long-distance travel in space, where there's nothing to slow it down, instead of bouncing down to a ground station and back up.
And light travels faster in a vacuum than in fiber. About forty percent faster. So the space hops are actually lower latency than the terrestrial fiber segments. It's a genuine architectural advantage, not just a workaround.
Alright. So the architecture explains why it's better in general. But the specific challenge of a fast-moving receiver — that's a whole other set of problems. Daniel's plane is moving at about nine hundred kilometers an hour.
Two hundred fifty meters per second. And the satellite it's connected to is moving at about twenty-seven thousand kilometers an hour — seven and a half kilometers per second — relative to the ground. The relative motion between the aircraft and the satellite is enormous. A satellite that's directly overhead one moment is below the horizon five to ten minutes later. With a GEO satellite, this isn't a problem because the satellite is effectively stationary in the sky. The antenna locks on and never moves. With LEO, the aircraft has to hand off from one satellite to the next every few minutes.
And it has to do it without dropping the connection.
That's the core engineering challenge Starlink solved. And the key is the phased-array antenna on the aircraft. Let me explain what that actually is, because the term gets thrown around a lot. A traditional satellite antenna is mechanically steered. It's a dish that physically rotates on gimbals to track a satellite. It has motors, bearings, moving parts. It takes seconds to re-point. A phased array has no moving parts at all. It's a flat panel with thousands of tiny antenna elements. Each element transmits the same signal but with a slightly different phase — a tiny time delay. By controlling those phase differences, you can shape and steer the beam electronically. The beam can be redirected in microseconds.
Microseconds versus seconds. That's the difference between a seamless handoff and a dropped connection.
And it gets better. The phased array can track multiple satellites simultaneously. It maintains a connection to the current satellite while pre-acquiring the next one. It knows the orbital parameters of the entire constellation — every satellite's position is predictable — so it can anticipate which satellite will be overhead next and start forming a beam toward it before the current satellite drops below the horizon.
So the handoff sequence is: connected to satellite A, already locked onto satellite B, switch when A goes below the horizon.
The switch happens in milliseconds. The network layer handles the routing, so from the user's perspective, there's no interruption. The IP session stays alive. The VPN tunnel doesn't drop. The video call doesn't freeze. That's what Daniel is experiencing and not even noticing — the handoffs are invisible.
There's another piece here that I want to get into. Doppler shift. At these relative velocities, the frequency of the signal shifts as the satellite approaches and recedes. That's a real physical effect. How does Starlink deal with it?
This is one of those problems that sounds esoteric but is actually critical. When a transmitter and receiver are moving toward each other, the frequency increases. When they're moving apart, it decreases. At seven and a half kilometers per second, the shift is significant — enough that if the modem didn't compensate, the signal would drift right out of the receiver's passband and the connection would drop.
The modem is continuously adjusting.
In real time. It's measuring the Doppler shift on the incoming signal and pre-compensating on the outgoing signal. This is a non-trivial engineering problem that GEO systems never had to solve because the relative motion between a GEO satellite and an aircraft is essentially zero. The satellite is stationary. Starlink's modems are designed from the ground up for high-mobility applications. They expect Doppler shift and they handle it continuously.
That's the vertical integration advantage, isn't it? SpaceX builds the satellites, the antennas, the modems, and the network software. They don't have to negotiate with a modem vendor about whether the firmware supports Doppler compensation for aviation.
That's a huge part of it. Traditional in-flight connectivity is a stack of components from different vendors. The antenna is from one company, the modem from another, the satellite capacity is leased from a third, the ground infrastructure from a fourth. Getting all those pieces to work together for a fast-moving aircraft is a nightmare of integration. Starlink owns the entire stack. They designed the phased array, the modem, the satellite, and the routing software as a single system. When they needed to add aviation support, they didn't have to convince a supply chain to cooperate. They just... did it.
There's a software dimension here too that's easy to overlook. The network has to treat the aircraft as a node that's constantly moving, and the routing has to continuously optimize the path through the constellation.
Right. In a traditional network, you have routers with fixed connections. A packet arrives, you look up the destination in a routing table, you forward it. The topology changes occasionally — a link goes down, a new router comes online — but it's mostly static. In Starlink's network, every satellite is moving. Every aircraft terminal is moving. The topology is changing continuously. The routing software has to compute optimal paths through a dynamic mesh where every node is in motion, and it has to do it fast enough that packets aren't delayed waiting for a route.
How fast are we talking?
The constellation updates routing tables on the order of milliseconds. Every satellite is constantly sharing its position, its link states, its available capacity with its neighbors. The routing algorithm — and Starlink has been pretty tight-lipped about the specifics — has to be something like a distributed, real-time version of a link-state protocol. It's pathfinding on a graph where the graph itself is in constant motion.
The software problem is at least as hard as the hardware problem.
Arguably harder. The phased array is impressive engineering, but electronically steered antennas have existed in military applications for decades. The Aegis radar system on Navy destroyers uses phased arrays. What's new is making them cheap enough to put on a commercial aircraft and pairing them with a network that can handle continuous handoffs at scale. That's a software achievement.
The practical result is what Daniel is experiencing. Usable video calls, streaming, VPNs, at thirty-seven thousand feet over the ocean. That's never been possible on a commercial flight before.
The qualitative difference he's feeling is the difference between a network designed for mobility and a network that's a fixed relay with a moving antenna bolted on. Traditional in-flight Wi-Fi takes a GEO satellite link — which was designed for fixed ground terminals — and tries to make it work on a plane by adding a mechanically-steered antenna. It works, sort of, for email and messaging. But the architecture was never designed for mobility. Starlink was designed from day one for moving receivers. The entire system — the satellites, the antennas, the routing — assumes the terminals are in motion.
There's an interesting knock-on effect here. Once you have low-latency, high-throughput connectivity on an aircraft, what else becomes possible? It's not just about passenger entertainment.
That's the question the airlines are asking. Real-time telemetry from the aircraft to maintenance crews on the ground. Continuous engine performance monitoring. Weather data streaming to the cockpit. The flight deck gets access to the same connectivity as the passengers, which means pilots can receive updated weather radar, turbulence forecasts, rerouting information in real time. Today, a lot of that comes over ACARS — a slow, text-based datalink that's been around since the seventies. Starlink makes it possible to stream full-resolution weather radar to the cockpit.
That's a safety improvement, not just a convenience.
It changes the economics for the airline. If the same connection that serves passengers also reduces fuel costs through better routing and reduces maintenance costs through real-time monitoring, the business case gets a lot stronger. It's not a passenger amenity anymore. It's an operational tool.
Which brings us to Hilbert. Hilbert, you've been quiet on this one. You have a history with in-flight connectivity, don't you?
Hilbert: I was a flight attendant for a now-defunct airline in the late nineties. We had the in-flight phone. The seat-back handset. Seven ninety-nine a minute. Connected through a GEO satellite. You'd pick up, dial, wait. Person on the other end would say hello. You'd hear it three seconds later. You'd say hello back. They'd hear it three seconds after that. It was unusable and everyone used it anyway because they were bored.
Seven ninety-nine a minute in nineteen ninety-eight dollars.
Hilbert: Twelve dollars in today's money, something like that. The pilots had one in the cockpit too. Satellite phone for calling ahead to operations. Same latency problem. I remember one captain hanging up and saying, I could send a letter faster than this.
That's a man who had given up on the twentieth century.
Hilbert: He wasn't wrong. The physics of LEO versus GEO has been understood since the sixties. Telstar, the first active communications satellite, was LEO. Launched in nineteen sixty-two. We knew the latency problem and we knew the solution. The delay wasn't about the satellites. It was about the business model.
Say more about that.
Hilbert: Airlines didn't invest in better connectivity because there was no revenue pressure. Passengers accepted bad Wi-Fi as the norm. It was a captive audience. What were you going to do, switch to a different in-flight internet provider at thirty-seven thousand feet? The airlines could charge ten dollars an hour for a connection that barely worked and people paid it. Starlink changed the economics by making the hardware cheap enough and the network scalable enough that an airline can offer it as a standard amenity. Emirates is including it with the ticket. That's a different model entirely.
The technology was always possible. It just wasn't profitable to build it.
Hilbert: Not until someone built the whole stack themselves and made the antennas cheap. A phased array used to cost a hundred thousand dollars. Military hardware. SpaceX got it down to something an airline can bolt onto a fuselage and forget about. That's the part that impressed me. Not the physics. The manufacturing.
Now that the hardware is cheap enough, the business case flips. It's not a premium add-on. It's table stakes.
Hilbert: Which is what the captain with the letter comment would have wanted. He just wanted to call ahead without waiting three seconds for a dial tone.
Hilbert, you've been waiting twenty-five years to see this moment, haven't you?
Hilbert: I've still got one of those seat-back handsets in a box somewhere. The cord is frayed. The keypad is sticky. I keep meaning to throw it out.
Don't. That's a museum piece now.
Hilbert: It's a paperweight. But yeah. Twenty-five years.
To leave on. So where does this go from here?
The open question is whether in-flight connectivity becomes a commodity. Starlink is the first mover in aviation, but OneWeb is targeting the same market, and Amazon's Kuiper is coming. If every airline has low-latency broadband from one LEO constellation or another, the connectivity itself stops being a differentiator. The competition shifts to what you do with it. Content, services, integration with the airline's app. The ground-side experience becomes the product, not the pipe.
The same technology is spreading. The phased-array and handoff solution that works for an aircraft at nine hundred kilometers an hour also works for a ship at thirty knots, or an RV parked in the desert, or a tractor in a field. The fast-moving receiver problem was the hardest version of the problem. Now that it's solved, the slow-moving and stationary applications are straightforward.
The maritime market is actually ahead of aviation in some ways. Starlink Maritime has been available for a couple of years. Cruise ships, commercial vessels, oil platforms. The economics are different — a ship can afford a larger, more expensive antenna, and the latency requirements for cargo vessels are less stringent than for passenger video calls. But the core technology is the same.
The thing that sticks with me from this whole discussion is that Daniel's experience on that Emirates flight is the first time a passenger has felt like in-flight Wi-Fi is just Wi-Fi. Not "impressive for an airplane." Indistinguishable from the ground. The technology has crossed that threshold, and the threshold is everything. Once it's good enough that you stop thinking about it, it's not a feature anymore. It's infrastructure.
That's the real story. We've gone from "I can check my email if I'm patient" to "I forgot I was on a plane" in the span of a few years. The engineering that made that possible — the phased arrays, the laser links, the handoff logic, the Doppler compensation — is invisible to the passenger. Daniel can't see any of it. He just knows it works. That's the mark of a solved problem.
If you take one thing from this episode, it's that the latency is the story, not the bandwidth. The reason Starlink feels different isn't that it's faster in megabits per second. It's that the signal path is shorter. The round trip to a GEO satellite is seventy-two thousand kilometers. To a LEO satellite, it's eleven hundred. That's the whole game.
The phased array is what makes it work on a moving aircraft. No moving parts, electronic beam-steering, seamless handoffs every few minutes. The antenna is doing more engineering in microseconds than the old mechanically-steered dishes did in seconds.
One thing I keep wondering about. As these constellations grow and the capacity increases, does the latency eventually hit a floor? We're at twenty to forty milliseconds now. Can that go lower, or are we bumping up against the speed of light and the processing overhead on the ground?
The light-time from the aircraft to the satellite and back is about four milliseconds. Add routing, processing, and the terrestrial fiber hops to the destination, and you're probably looking at a practical floor around ten to fifteen milliseconds for most routes. We're not that far from it. The gains from here are in reliability, coverage density, and throughput — not latency.
The captain with the letter was closer to the truth than he knew. The physics hasn't changed. We just finally built the thing.
Thanks to our producer Hilbert Flumingtop for keeping the show running, and for the seat-back handset that should probably be in the Smithsonian.
This has been My Weird Prompts. Find us at my weird prompts dot com, and if you've got a question you want us to dig into while you're hurtling across the Atlantic at thirty-seven thousand feet, email the show at show at my weird prompts dot com.
We'll be back soon.
