The One Sentence That Explains Everything

A satellite’s entire life runs on a single, finite budget — not money, but speed.

Every move it makes — climbing to its orbit, holding its position against the forces trying to drag it down, dodging a piece of debris, and finally diving home to burn up — spends from the same limited tank of propellant. When that tank runs dry, the satellite can no longer fight, no longer steer, no longer save itself. A satellite is born with a fixed amount of “go,” and the day it runs out is the day it begins to die.

Everything else in this post is the detail behind that one idea: how the “go” works, what it’s spent on, and where satellites end up when it’s gone.

You Don’t Steer a Satellite, You Change Its Orbit

The first thing to unlearn: there is no steering wheel in space. A satellite doesn’t turn left or hit the gas the way a car does. To understand why, remember what an orbit actually is (from the previous post on finding your way in space): a satellite isn’t flying, it’s falling — moving sideways so fast that it keeps missing the Earth. Its path is set entirely by its speed and direction at any given moment.

So the only way to move a satellite somewhere new is to change its speed. Speed it up and your orbit swells outward into a bigger loop. Slow it down and your orbit shrinks inward. Give it a sideways shove and you tilt the whole loop. There’s no friction to coast against and nothing to push off of — so every change of course is a deliberate, measured kick of velocity.

That’s the mental shift: a satellite navigates not by turning, but by carefully spending little bursts of speed change. And those bursts aren’t free — each one costs propellant. Which brings us to the most important number in all of spaceflight.

Delta-v: The Currency of Space

Engineers have a word for that budget of “go”: delta-v (written Δv, meaning “change in velocity”). It’s the total amount of speed change a satellite can produce over its whole life with the propellant it carries. Every maneuver has a price tag measured in delta-v, and the satellite spends from a fixed wallet.

How big is the wallet? It’s governed by one elegant equation — the Tsiolkovsky rocket equation, written down in 1903:

Don’t let it scare you. In plain words it says: your speed budget depends on two things — how efficient your engine is (Isp), and how much of your satellite is fuel versus everything else (the mass ratio: full mass m₀ divided by empty mass m_f). The “ln” is a natural logarithm, and its presence delivers a harsh truth: to get twice the delta-v, you don’t need twice the fuel — you need exponentially more. Wanting a lot of maneuvering ability means carrying a satellite that is mostly propellant tank.

The practical consequence is everything: fuel equals lifespan. A geostationary satellite might carry 15 years of delta-v for holding its position. When that propellant is spent, the satellite is still perfectly healthy — working radios, charged batteries, good computers — but it can no longer hold station or save itself. It’s out of “go.” That, more than any broken part, is what usually kills a satellite.

How You Push in a Vacuum

Here’s a puzzle: in space there’s nothing to push against — no road, no air, no water. So how does an engine move anything at all?

The answer is Newton’s third law: every action has an equal and opposite reaction. If you throw something backward, you get pushed forward — exactly like a skateboarder throwing a heavy ball and rolling away in recoil. A rocket engine does precisely this, just very fast and continuously: it hurls mass (hot gas, or charged particles) out the back, and the satellite recoils forward. Since there’s nothing external to push on, a satellite must carry its own stuff to throw — that’s what “propellant” really is. Not fuel to burn for energy, but mass to throw overboard.

This reveals what makes an engine “good.” Two different things matter:

• Thrust — how hard it pushes right now (how much mass, how fast, out the back this second).

• Efficiency — how much push you extract from each kilogram of propellant you brought. Engineers measure this as specific impulse (Isp), basically the “miles per gallon” of a rocket. Higher Isp means you wring more delta-v out of less propellant.

These two pull against each other, and the history of propulsion is the story of trading between them. Early satellites used crude, low-efficiency cold gas (literally a puff of pressurized gas) and chemical engines (high thrust, thirsty). Then came the electric-propulsion era: NASA’s Deep Space 1 (1998) flew the first ion engine on an interplanetary mission, and Europe’s SMART-1 (2003) sipped so little propellant it slowly spiralled all the way to the Moon on a thruster you could hold in your hands. Let’s look at the full menu.

The Propulsion Menu

A satellite’s engine is chosen by trading thrust against efficiency — like choosing between a drag racer and a hyper-mileage commuter car. Here’s the working set:

The key tradeoff jumps out of the table: chemical engines push hard but drink propellant; electric engines sip propellant but push so gently it can take months to change orbit. A chemical thruster can shove a satellite in seconds; an ion engine produces about the force of a sheet of paper resting on your hand — but it can run for years, and that patience adds up to enormous total delta-v.

The real-world poster child is Starlink: every satellite carries a Hall-effect thruster to raise itself to its operating altitude, hold station, and dodge debris. Early Starlinks used krypton as propellant; the newer V2 satellites switched to argon (cheaper and more abundant). Nearly ten thousand of these little electric engines are quietly humming in orbit right now — the largest fleet of thrusters in history.

But here’s the surprising flip side: most small satellites carry no engine at all. Over 80% of CubeSats launch without any propulsion — they can’t dodge debris, can’t fight drag, and can’t deorbit themselves on purpose; they simply ride the orbit they were dropped into until the atmosphere claims them. As orbits fill up and disposal rules tighten, that’s becoming untenable — which is why a wave of new companies (including our own work at Player One Space on a water-plasma pulsed thruster) is racing to give the smallest spacecraft a safe, affordable engine. Water is an appealing propellant precisely because of everything this post covers: it’s non-toxic, stored unpressurized, and rideshare-friendly — no hydrazine handling, no high-pressure tank sitting next to someone else’s satellite.

What Satellites Actually Do With That Fuel

So where does all that delta-v go? A satellite spends its budget on a handful of recurring jobs:

Orbit raising. Rockets usually drop satellites off in a low “parking” orbit. The satellite then fires its own engine to climb to its real operating altitude. For an electric-propulsion satellite this can be a months-long, gentle spiral upward.

Station-keeping — the biggest lifelong expense. An orbit is never left in peace: the thin upper atmosphere creates drag, the Moon and Sun tug on it, and the Earth itself isn’t a perfect sphere (it’s slightly lumpy, which nudges satellites off course). Left alone, a satellite slowly drifts out of position and, in low orbit, eventually falls. So it makes constant tiny correcting burns to stay exactly where it belongs — like a swimmer treading water against a current. This quiet, endless fight consumes most of a satellite’s propellant over its life.

Catching up and rendezvous — and here’s a delicious paradox covered in the navigation post: to catch up to something just ahead of you in orbit, you do not speed up. Firing forward raises your orbit, lengthens your lap time, and makes you fall behind. To catch up, you fire backward, dropping into a lower, faster orbit to close the gap. Orbital motion is gloriously counterintuitive.

Plane changes — the most expensive maneuver of all. Tilting your orbit to a different angle (say, from equatorial to polar) means fighting your entire orbital speed sideways. It can cost more delta-v than reaching orbit in the first place, which is why satellites are launched into the right orbital plane rather than changing it later.

The deorbit burn — the final expense, saved for the very end: one last burn to drop out of orbit on purpose. More on that soon.

The Traffic Problem: Space Is Huge, but Filling Up

Before we follow a satellite to its grave, we have to talk about the danger it spends its whole life dodging: everything else up there.

Low Earth orbit is getting crowded — and not just with working satellites. Decades of launches, dead spacecraft, spent rocket stages, and shattered fragments have filled the most useful orbits with junk. The numbers (as of 2026) are sobering:

And it’s not the size that’s terrifying — it’s the speed. Objects in low orbit move at roughly 7.8 km/s, and two of them on crossing paths can meet at a closing speed of 10–15 km/s (about 10× faster than a rifle bullet). At those velocities, a fleck of debris just 1 cm across hits with the energy of a hand grenade, and a 10 cm bolt can obliterate an entire satellite. Worse, anything smaller than about 10 cm is too small to track from the ground — present, lethal, and invisible.

Some orbital “shells” are far worse than others:

• 400–500 km (the ISS zone): relatively self-cleaning — air drag drags junk down within years.

• 500–600 km: the densest active traffic (Starlink, Kuiper); debris here clears in ~25 years.

• 700–900 km: the worst existing debris band; junk lingers for decades to centuries.

• 1,000–1,200 km: any mess here is essentially permanent.

We know it’s dangerous because it has already happened. China’s 2007 anti-satellite missile test shattered its own Fengyun-1C satellite at 865 km, adding 3,000+ trackable fragments and bloating the global debris catalog by a quarter. In 2009 a dead Russian satellite, Cosmos 2251, slammed into the working Iridium 33 at 11.7 km/s — the first accidental collision of two intact satellites. And in 2021, Russia’s Cosmos 1408 ASAT test forced the ISS crew to shelter. Each event scattered thousands of new bullets into orbit.

This is the nightmare scenario scientists Donald Kessler and Burton Cour-Palais described back in 1978: the Kessler Syndrome, where collisions create debris that causes more collisions, in a cascade that could one day make whole orbits unusable. Which raises the obvious question…

So Why Don’t They All Crash?

If orbit is so crowded and the speeds so deadly, why don’t we hear about collisions every day? Three reasons work together.

1. Space is genuinely vast. This is the most underrated answer. Even ~10,000 satellites are spread across a shell that stretches around the entire planet and hundreds of kilometers thick — a volume of trillions of cubic kilometers. On average, satellites are hundreds of kilometers apart. And orbits aren’t random chaos; they’re carefully organized into planned altitudes and planes, like floors in a parking garage. Most of the time, “crowded” still means “mostly empty.”

2. Everything big is tracked. A global watch keeps a catalog of every object larger than about 10 cm. The US Space Force’s 18th and 19th Space Defense Squadrons run the master catalog (shared publicly via Space-Track.org), and a growing commercial industry sharpens the picture — LeoLabs with a worldwide network of radars, Slingshot Aerospace with optical telescopes. Together they function as the air-traffic control of orbit, predicting where every tracked object will be.

3. Operators get warned, and they move. When two objects are predicted to pass dangerously close, operators receive a Conjunction Data Message (CDM) — a formal “heads-up, you two might collide.” Each warning comes with a calculated probability of collision, and a common rule of thumb is: if that probability climbs above roughly 1 in 10,000, you maneuver out of the way. A small burn hours ahead of time turns a near-miss into a comfortable gap. Starlink alone processes thousands of these warnings every week and dodges autonomously — its satellites decide and maneuver without a human in the loop.

Tracking finds the threat; a little delta-v dodges it. That’s the system that keeps orbit working. But it has a gap — and the ISS shows how we handle it.

Why the ISS Doesn’t Get Hit

The International Space Station — the most precious object in orbit, with humans aboard — is the best illustration of the whole defense system in action. It survives by layering three tactics:

Dodge the big stuff. When trackers predict a close approach, the ISS fires its thrusters (or its docked vehicles’ thrusters) to shift its orbit out of harm’s way. It has performed roughly 40 of these debris-avoidance maneuvers since 1998 — a few every year.

Hide from the surprises. Sometimes a threat is spotted too late to maneuver. In that case the crew shelters in their docked spacecraft (the Soyuz or Dragon capsules that brought them up), sealing themselves into their “lifeboats” until the object passes — ready to undock and flee home if the worst happens.

Armor against the invisible. For the untrackable specks smaller than ~1 cm that you cannot dodge, the station wears armor: the Whipple shield, invented by astronomer Fred Whipple in the 1940s. It’s deceptively simple — a thin outer “bumper” plate held a gap away from the real hull. A hypervelocity fleck hits the bumper and shatters and vaporizes itself, so that only a harmless spray of dust — not a solid bullet — reaches the wall behind it. The ISS carries over a hundred enhanced shields stuffed with ceramic and Kevlar.

Notice the strategy: tracking handles the big objects, shields handle the tiny ones — and the scary gap is the middle, the 1–10 cm fragments too small to track but too big for any shield to stop. That gap is precisely why keeping orbit clean matters so much.

Permission to Fly: Who Lets You Launch a Satellite?

Here’s something many people don’t realize: you can’t just build a satellite and launch it. Space is regulated, and getting a spacecraft to orbit legally means collecting a stack of permissions first. In the United States, for example:

• A radio license (FCC). Almost every satellite transmits radio signals, and you can’t use the airwaves without authorization. This is the big one — and it’s where the regulator checks your whole plan.

• Spectrum and orbital-slot coordination (ITU). Internationally, the International Telecommunication Union coordinates who broadcasts on which frequencies and (for high orbits) who parks in which slot, so satellites don’t jam each other.

• An imaging license (NOAA). If your satellite photographs the Earth, you need a separate commercial remote-sensing license.

• A launch license (FAA). The rocket itself needs approval to fly.

• Plus your own national space regulator, depending on the country.

Crucially, modern licenses now demand an end-of-life plan before you’re allowed up. You must show a credible deorbit plan (how you’ll remove the satellite when it’s done) and commit to passivation — venting all leftover fuel and draining batteries at the end so your dead satellite can’t later explode into a cloud of debris. Permission to launch is now tied to a promise to clean up.

Does all this apply to tiny CubeSats and “pocketsats”? Absolutely — size is no exemption. A satellite the size of a loaf of bread, or even a deck of cards, needs the same authorizations. And the most famous lesson in the industry proves it: in 2018 the FCC fined a startup called Swarm Technologies $900,000 for launching four “SpaceBEE” picosatellites — each just 0.25U, about the size of a sandwich — on an Indian rocket without authorization. The FCC had specifically denied their application beforehand, partly because the satellites were judged too small to be reliably tracked from the ground — exactly the “invisible debris” risk that keeps regulators up at night. Swarm launched anyway, and paid for it. The message to the whole industry was clear: no satellite is too small to follow the rules.

The Good Death: How a Satellite Is Disposed Of

Eventually every satellite’s mission ends — fuel runs low, hardware ages, or newer technology makes it obsolete. A responsible operator doesn’t just abandon it as a hazard. There are two proper ways to die, and which one you get depends entirely on how high you are.

Low orbit → the fiery reentry. Down low, you spend your last delta-v on a deorbit burn that drops the satellite into the thin upper atmosphere — where it slams into air at orbital speed, heats to thousands of degrees, and burns up in a streak of fire. (Below about 500 km, atmospheric drag will eventually drag a dead satellite down on its own — no burn needed.) For years the guideline was to clear out within 25 years of mission’s end, but in 2022 the FCC tightened it dramatically to a 5-year rule for satellites it licenses. Engineers even practice “design for demise“ — deliberately building satellites so that every part fully burns up and nothing dangerous reaches the ground.

High orbit → the graveyard. A geostationary satellite sits at 35,786 km — far too high to drop into the atmosphere; the delta-v to come home from there would be enormous. So instead, near end of life, GEO satellites spend a small final budget to boost themselves about 300 km higher, into a designated “graveyard orbit” above the busy belt (an international IADC guideline), then passivate and go silent. They never come home — they just retire to a quiet parking lot in the sky, drifting forever.

But what about the big low-orbit satellites whose parts don’t fully burn up? They get the most dramatic ending of all.

The Satellite Graveyard in the Ocean: Point Nemo

Big, dense satellites — and especially space stations — don’t burn up completely. Chunks of titanium and steel (fuel tanks, reaction wheels, dense structural parts) can survive the inferno of reentry all the way to the surface. Since the global safety standard demands that the risk of a reentry hurting anyone on the ground stays below 1 in 10,000, you can’t let those survivors fall just anywhere. You aim them at the emptiest place on the entire planet.

That place has a name: Point Nemo — the “oceanic pole of inaccessibility,” at roughly 48°52′S 123°23′W in the remote South Pacific. It is the single spot on Earth farthest from any land in every direction; the nearest people are often the astronauts on the ISS passing 400 km overhead, closer than any coastline. Because nothing and no one is there, it became the world’s spacecraft cemetery.

Somewhere between 260 and 300+ spacecraft now rest on the seabed beneath Point Nemo — defunct satellites, cargo ships, and spent stages, deliberately steered to splash down there. Its most famous resident is Russia’s Mir space station, guided to a fiery, controlled reentry over the Pacific on March 23, 2001.

And the cemetery is about to receive its largest occupant ever. When the International Space Station is retired around 2030–2031, it will be deorbited to Point Nemo — a structure the size of a football field, far too massive to risk an uncontrolled fall. To do it, NASA selected SpaceX in 2024 to build a purpose-made “US Deorbit Vehicle” — a beefed-up Dragon capsule packed with extra thrusters and propellant — under a contract worth up to $843 million, whose only job is to grab the station and aim humanity’s greatest orbital outpost at that lonely patch of ocean.

Cleaning Up: The Future of Orbital Housekeeping

Here’s the catch with all these disposal rules: they only slow the creation of new junk. They do nothing about the 35,000+ pieces already up there — the dead satellites and spent rockets that will keep colliding and fragmenting for centuries. ESA’s models deliver a stark verdict: just to stabilize the worst orbital bands, we’d need to actively remove about five large objects every year — and right now we remove essentially zero.

So a new industry is being born: active debris removal (ADR) — spacecraft that chase down dead objects and drag them out of orbit. The pioneers are flying:

• ClearSpace-1 (Switzerland/ESA) — originally contracted to capture a ~112 kg leftover rocket adapter with a robotic arm. In a grim twist, that adapter was itself struck by debris in 2023 — the problem underlining its own cleanup mission — so the rescoped mission now targets ESA’s retired PROBA-1 satellite later this decade.

• Astroscale (Japan/UK) — demonstrated magnetic capture with ELSA-d in 2021; its ELSA-M servicer aims to deorbit several client satellites per mission.

• D-Orbit (Italy) — space tugs that deliver satellites and then responsibly remove themselves.

• RemoveDEBRIS (UK) — back in 2018, actually tested catching debris with a net and a harpoon in orbit.

• Starfish Space (USA) — its Otter servicer is already winning US Space Force contracts to deorbit satellites.

The frontier beyond cleanup is in-orbit refuelling — companies like Orbit Fab want to set up “gas stations” in space. Refilling a satellite’s delta-v changes everything: a debris-removal vehicle could top up and bag many targets instead of one, and satellites could be kept alive for decades instead of dying when the tank runs dry. Combine refuelable tugs with the AI-piloted rendezvous from the navigation post, and orbit stops being a one-way graveyard and becomes something we actually maintain.

Summary: A Satellite’s Whole Life in One Breath

1. You don’t steer in space — you change your orbit by changing your speed. Every maneuver is a measured kick of velocity.

2. Delta-v is the currency of space — a fixed budget of speed change set by the Tsiolkovsky equation. Fuel equals lifespan.

3. Engines push by throwing mass backward (Newton’s third law); they’re judged on thrust vs efficiency (specific impulse).

4. The propulsion menu trades the two: cold gas and chemical (high thrust, thirsty) vs ion and Hall-effect (sip propellant, push gently). Starlink runs the largest fleet of electric thrusters ever.

5. Satellites spend delta-v on orbit raising, lifelong station-keeping against drag, rendezvous (with the catch-up paradox), costly plane changes, and the final deorbit burn.

6. Orbit is crowded and lethal — tens of thousands of tracked objects, millions of untracked specks, closing at 10–15 km/s, with the Kessler cascade as the long-term fear.

7. They mostly don’t collide because space is vast, big objects are tracked (Space Force + LeoLabs + Slingshot), and operators dodge when collision odds top ~1 in 10,000.

8. The ISS survives by maneuvering (~40 times), sheltering the crew when it’s too late, and wearing Whipple shields against the specks it can’t dodge.

9. You need permission to fly — FCC, ITU, NOAA, FAA, and a mandatory deorbit-and-passivation plan — and it applies even to sandwich-sized picosats (just ask Swarm and its $900K fine).

10. Satellites die two ways: low ones burn up on reentry (now within 5 years); high ones retire to a graveyard orbit ~300 km above GEO.

11. The survivors splash down at Point Nemo, the remotest ocean on Earth — soon to receive the ISS itself, via a purpose-built SpaceX deorbit vehicle.

12. The future is housekeeping — active debris removal and in-orbit refuelling, turning orbit from a one-way graveyard into something we maintain.

A satellite is born with a fixed tank of “go.” It climbs, it holds its ground against the forces dragging it down, it dodges the debris trying to kill it — and when the tank finally runs dry, it takes the long fall home. How well we manage that whole arc, from launch permission to final splashdown, decides whether orbit stays usable for the next generation.

Up next: how satellites get to orbit in the first place — the story of rockets, rideshares, and the brutal climb uphill.

Published by Player One Space | dima@playeronespace.com

The One Sentence That Explains Everything

Every satellite, every moment it’s alive, is answering two completely separate questions:

“Where am I?” (its position — its dot on the map of space) and “Which way am I facing?” (its attitude — its orientation, the direction each side is pointed). It measures both, constantly, and then nudges itself to keep its attitude exactly where it needs to be.

That’s the whole job. Everything else in this post is how it pulls off those two measurements — and why they’re far harder, and far more different from each other, than they sound.

The Two Questions People Always Confuse

Imagine you wake up blindfolded in a strange city. Two different questions need answering, and they’re not the same:

1. Where am I? Which street, which intersection — your dot on the map. This is position.

2. Which way am I facing? North? Toward the alley? This is orientation — what satellite engineers call attitude.

You could know your exact GPS coordinates and still have no idea which way you’re looking. And you could know you’re facing north while being hopelessly lost. They are independent facts, and — crucially — you measure them with totally different tools. Your phone’s GPS gives you position. A compass gives you orientation. One does not give you the other.

Satellites live with this same split every second:

• Position is solved by orbit determination — GPS receivers, ground radar, and the steady clockwork of orbital physics.

• Attitude is solved by attitude determination — sun sensors, star cameras, gyroscopes, and magnetic sensors.

And once a satellite knows its attitude, it has to change it — turning to point a camera at a city, an antenna at a ground station, or its panels at the Sun. That third job is attitude control. Put the determination and control together and engineers call the whole system ADCS (Attitude Determination and Control System), or more broadly GNC (Guidance, Navigation, and Control).

Keep the two questions separate in your head — where am I vs which way am I facing — and everything that follows clicks into place.

A Short History: From Spinning Bullets to Star Cameras

The story of how satellites keep their bearings is a story of escaping chaos.

Spin stabilization (the football trick). The earliest satellites used the same trick as a thrown football or a rifle bullet: spin. A spinning body resists being knocked off-axis — gyroscopic stiffness. Set the satellite spinning and it holds its orientation in space like a top. Simple, passive, no computer needed.

Explorer 1’s embarrassing surprise (1958). America’s first satellite, Explorer 1, was designed to spin neatly about its long, pencil-like axis. Within a single orbit, it did something nobody expected: it flipped into an end-over-end flat spin. The culprit was its four flexible whip antennas — they flexed as it spun, quietly bleeding off energy. And here’s the iron law of physics that bit them: a flexible spinning object will always drift toward spinning about the axis with the most inertia (the “fat” tumbling axis), not the least. Explorer 1 taught engineers the hard way that real satellites aren’t rigid — and that attitude is not something you can just set and forget.

Gravity-gradient booms (let the Earth do it). A clever passive fix: extend a long boom, and Earth’s gravity — very slightly stronger on the near end than the far end — gently tugs the satellite into pointing at the planet, like a weighted buoy standing upright in water. Free, but only roughly accurate.

3-axis stabilization (the modern standard). Today almost all satellites are 3-axis stabilized: instead of spinning, they hold a fixed, controlled orientation on all three axes using active sensors and actuators (which we’ll meet soon). This is what lets a telescope stare at one galaxy for hours or a camera lock onto a precise spot on Earth.

The ongoing saga. Even the best systems fail. The Hubble Space Telescope has fought gyroscope failures for decades — gyros died in 1997, 1998, and 1999; astronauts replaced all six in 2009; and it has dropped into safe mode over gyro trouble as recently as 2023. Keeping a satellite’s bearings is not a solved problem you install once. It’s a fight that lasts the whole mission.

“Where Am I?” Starts With Physics, Not Sensors

Here’s the first surprise about position: a satellite mostly doesn’t need to measure it constantly — because physics predicts it.

A satellite isn’t flying; it’s falling. An orbit is just a fall that keeps missing the Earth: gravity pulls the satellite down, but it’s moving sideways so fast (about 7.8 km/s in low orbit) that the ground curves away beneath it just as fast as it falls. The result is a stable, endlessly repeating loop.

And falling is predictable. If you know a satellite’s exact position and velocity at one instant — its “state” — the laws of gravity let you calculate where it will be hours or days later, the same way astronomers predict eclipses centuries in advance. This calculation is called propagation.

So the real task of “where am I” isn’t continuous measurement — it’s: (1) pin down the state accurately once in a while, and (2) let the math carry it forward in between. The catch is that the prediction slowly drifts — thin atmospheric drag, the Sun’s pressure, the Moon’s tug, the Earth’s lumpy gravity field all nudge the satellite off the ideal path. So you have to re-measure periodically to correct the drift. The next sections are about how that measurement is actually done — and, critically, how it changes depending on how high you’re orbiting.

The Positioning Sensors: How a Satellite Fixes Its Own Location

When a satellite does measure its position, it has a small toolkit:

• GNSS receiver onboard (space GPS). The most common method today: a satellite carries a GPS receiver, the same fundamental tech as the one in your phone. It listens to the fleet of navigation satellites (GPS, plus Europe’s Galileo, Russia’s GLONASS, China’s BeiDou — collectively GNSS) and trilaterates its position from their timing signals. The twist: a satellite is screaming along at ~7.8 km/s, so its receiver has to handle huge Doppler shifts and lock onto signals far faster than a car navigator ever would. A space-rated GNSS receiver can fix a satellite’s position to within a few meters.

• Ground tracking (the world watches you). Even without GPS, the ground can find a satellite. A global web of radar stations and optical telescopes — the US Space Force’s Space Surveillance Network — tracks tens of thousands of objects in orbit. Their measurements are distilled into a compact data format called a TLE (Two-Line Element set): two lines of numbers describing an object’s orbit, which anyone can feed into a propagation model (the standard one is called SGP4) to predict where that object will be. This is how hobbyists know when the ISS will fly over their backyard.

• Ranging. A ground station can also send a signal up and time how long the round-trip takes — distance equals time times the speed of light. Combine ranging from a few stations and you get a precise fix. This is the backbone of deep-space navigation.

• Inertial dead-reckoning. Between fixes, a satellite can estimate how far it’s moved using onboard motion sensors (an IMU — inertial measurement unit) — the same principle as counting your steps in the dark. It drifts over time, so it’s a stopgap between real fixes, not a standalone answer.

Positioning Changes With Altitude: LEO vs MEO vs GEO vs Beyond

Here’s something most people never realize: how a satellite finds its position depends enormously on how high it’s orbiting. The GPS trick that works perfectly in low orbit barely works at all up high. Here’s the full ladder:

The key insight is about where the GPS constellation lives: the GPS satellites orbit at about 20,200 km (in MEO), and they beam their signals down toward Earth.

• In LEO, your satellite is below them, looking up into a sky full of GPS — it works beautifully.

• At GEO, your satellite is above the entire GPS constellation. The GPS satellites are now mostly on the far side of the Earth, beaming away from you, with the planet blocking the view. So how does a GEO satellite use GPS at all? With a clever trick: it catches the faint edges of the GPS signal beams that spill past the curve of the Earth — the main beam plus weak “side-lobes” that leak around the planet’s limb. It takes a specialized, super-sensitive high-altitude GNSS receiver, but it works: NASA’s GOES-16 weather satellite was the first to operationally use GPS this way at GEO.

• Beyond GPS reach — out toward the Moon and into deep space — GPS is useless, and spacecraft fall back on ground-based ranging from giant antennas (NASA’s Deep Space Network), plus the celestial and pulsar methods we’re about to meet. (This is exactly why “GPS for the Moon” is an active frontier — there’s no satnav out there yet.)

So “how does a satellite know where it is” has no single answer. It depends on which shell of space it’s flying in.

Can a Satellite Navigate by the Stars?

Here’s a question almost everyone asks: the old sailors used the stars — can a satellite just take a picture of the stars and know where it is?

The answer is beautifully subtle, and it’s worth getting exactly right: a picture of the stars tells a satellite which way it’s facing — but NOT where it is.

Why? Because the stars are unimaginably far away. Whether your satellite is over Canada or over Australia, the star field looks identical — the stars don’t visibly shift at all across the width of an orbit (their parallax is far too tiny to measure). So a star photo is a perfect compass: match the pattern of stars in your camera to a catalog, and you know your orientation to within an arc-second. But it’s a useless map: the same star pattern is what you’d see from anywhere in orbit, so it can’t pin your location.

To get position from the sky, you need a second, nearer reference — something close enough that its apparent position actually shifts as you move:

• Celestial navigation (the sextant trick). Measure the angle between a star and a nearby body — the edge of the Earth (its “limb”), a planet, or a moon. The stars stay fixed, but the near body slides against them as your position changes — and that shift reveals where you are. NASA’s deep-space probes do exactly this: Deep Space 1 flew an autonomous optical navigation system called AutoNav (the first spacecraft to fix its own position without help from Earth), and OSIRIS-REx used optical navigation, photographing the asteroid Bennu against the background stars to triangulate its approach.

• Pulsar navigation — “GPS from dead stars.” This one sounds like science fiction but it’s real. Certain collapsed stars (pulsars) spin and flash beams of X-rays past us with the regularity of an atomic clock. By timing those flashes from several known pulsars, a spacecraft can triangulate its position anywhere in the solar system — no Earth contact needed. NASA proved it in November 2017 with an experiment called SEXTANT, running on the NICER instrument aboard the Space Station: using four millisecond pulsars as beacons, it pinned the station’s position to within about 7 kilometers, purely from starlight.

The mental model to keep: Stars alone = a compass (orientation). A star plus a nearby body, or the ticking of pulsars = a map (position).

And that “compass” use — matching stars to a catalog for orientation — is so good it became the gold standard for attitude. Which brings us to the next question.

“Which Way Am I Facing?” The Attitude Sensors

Now the second question — orientation. A satellite carries a suite of sensors, each with its own strengths, and it blends them. Here’s the lineup, from coarse-and-cheap to precise-and-pricey:

A few of these deserve a word:

• The gyroscope is special because it doesn’t measure orientation directly — it measures how fast you’re turning. Add up (integrate) that rate over time and you get your change in orientation. The problem: tiny errors accumulate, so a gyro’s estimate slowly drifts away from truth — like trying to track your position by counting steps with your eyes closed. Accurate for a few seconds, hopeless after an hour.

• The star tracker is the hero. It’s a small digital camera pointed at deep space. It photographs whatever stars are in view, then matches that pattern against an onboard catalog of thousands of stars — pattern-recognition, like face ID for the sky. Once it identifies the stars, it knows the camera’s exact orientation to within an arc-second (1/3600th of a degree). This is the same star-camera idea from the navigation section — but used purely as a compass for orientation, which is what stars are actually good for.

No single sensor is enough, though. The star tracker is precise but can be blinded by the Sun and is slow to recover if the satellite is tumbling. The sun sensor and magnetometer are crude but always give something. The gyro is fast but drifts. The solution is to combine them.

No Single Sensor Is Enough: Sensor Fusion

Real satellites don’t pick one sensor — they fuse all of them into a single best estimate, playing each one to its strength:

• When a satellite first wakes up tumbling and lost, it leans on the crude-but-reliable sun sensor and magnetometer to get a rough orientation — enough to stop tumbling and find the Sun for power.

• Once stable, the star tracker delivers pinpoint orientation.

• Between star-tracker updates, the fast gyroscope tracks moment-to-moment turning — and because the gyro drifts, the star tracker periodically resets it back to truth. The gyro fills the gaps; the star tracker corrects the drift. Together they’re better than either alone.

The mathematical machinery that does this blending has a famous name: the Kalman filter. In one plain sentence: a Kalman filter combines a prediction of where you think you’re pointing with a fresh sensor measurement, weighting each by how much you trust it, to produce an estimate better than either. If the gyro says “you’ve turned 10°” and the star tracker says “you’re now at this exact attitude,” the filter merges them intelligently — trusting the gyro for smoothness and the star tracker for truth. It’s the same kind of math behind your phone knowing its orientation by blending its gyroscope, accelerometer, and compass.

“How Do I Turn?” The Actuators

Knowing your attitude is only half the battle. To point at something, you have to physically rotate the satellite — and there’s no ground to push against in space. So how does a satellite turn itself? Three main tools:

The reaction wheel is the elegant workhorse, and it runs on one of the most beautiful tricks in physics: conservation of angular momentum. There’s nothing to push against in space — so the satellite pushes against itself. Spin a heavy internal wheel clockwise, and the satellite body must rotate counter-clockwise to balance it (Newton’s third law, in spin form). Speed the wheel up or slow it down, and the satellite turns with exquisite precision — no fuel burned. Three wheels (one per axis) let a satellite point anywhere.

But there’s a catch: the saturation problem. Every time the satellite fights a persistent nudge (like atmospheric drag or sunlight pressure always pushing the same way), its wheels have to spin a little faster to compensate. Eventually they hit their maximum RPM — “saturated” — and can’t speed up any more. They’re full. To empty them, the satellite uses magnetorquers or thrusters to hold itself steady while it slows the wheels back down — a maneuver called desaturation or “momentum dumping.” It’s like a figure skater who can only spin up so fast before they have to reach out and touch the wall.

Magnetorquers are simple electromagnets. Switch them on and they push against Earth’s magnetic field — the same way a compass needle twists to align with it. Free and reliable, but weak, and only useful in LEO where the field is strong (which is also why they can’t help a deep-space probe).

Thrusters are the brute-force option: fire a jet, get a fast, powerful turn or an actual change in orbit. But every puff spends propellant the satellite can never refill, so thrusters are rationed carefully — used for big maneuvers and saved for when wheels and torquers aren’t enough.

Putting It Together: The Control Loop

Sensors tell you where you’re pointing. Actuators let you turn. The thing that ties them into purposeful behavior is the control loop — the satellite’s sense of balance, running silently many times per second:

1. Determine — fuse the sensors into a best estimate of current attitude.

2. Compare — measure the error: the gap between where it’s pointing and where it wants to point.

3. Command — tell the actuators how hard to push to shrink that error.

4. Act — wheels spin, torquers pulse, the satellite rotates a hair.

5. Repeat — measure again, correct again, forever.

This is the exact same loop as balancing a broom on your palm: your eyes sense it tipping (determine), your brain notes which way and how far (compare), and your hand slides to catch it (command + act) — over and over, faster than you consciously think. A satellite does this with its sensors, its computer, and its wheels, holding a telescope dead-steady on a galaxy a billion light-years away. Determine, compare, command, act, repeat — that loop is the heartbeat of every pointed satellite in space.

Why Any of This Matters: Pointing Modes

All this machinery exists so a satellite can point precisely — and what it points at changes by the minute. Common pointing modes:

• Sun-pointing — keep the solar panels square to the Sun. Without it, the battery dies. This is the first thing a satellite does after deployment and the safe fallback if anything goes wrong.

• Nadir-pointing — aim straight “down” at the Earth. Essential for imaging cameras, weather sensors, and communications antennas that must stay locked on a ground station.

• Target tracking — slew to keep a specific spot (a city, a storm, a wildfire) centered in a camera’s view as the satellite races overhead at 7.8 km/s. The satellite has to physically rotate to counter its own motion.

• Inter-satellite laser pointing — the most extreme case of all. As covered in the previous post on getting data down from space, two satellites linking by laser must hold a pencil-thin beam on a target thousands of km away, both moving at ~7.8 km/s. That’s the “Acquisition, Tracking & Pointing” problem — and it’s pure attitude control pushed to its absolute limit.

Get pointing wrong and everything fails at once: panels miss the Sun (no power), antennas miss the ground (no data), cameras smear (no usable images). Attitude control isn’t a nice-to-have — it’s the difference between a working satellite and an expensive tumbling brick.

When It Goes Wrong: Tumbling and Safe Mode

Things break in orbit, and a satellite that loses its bearings is in mortal danger — drifting panels mean a dead battery within hours. So satellites are built to fail gracefully.

The universal fallback is safe mode: if the satellite gets confused, detects a fault, or starts tumbling, it abandons whatever it was doing, uses its crudest reliable sensors (sun sensor, magnetometer) to find the Sun, and rotates to keep its panels charged — then waits, like a flower turning to the light, for operators on the ground to diagnose and recover it. Survival first; mission later.

The hardware fails too. Kepler, the planet-hunting telescope, relied on four reaction wheels to hold its ultra-steady stare; one failed in 2012, a second in May 2013, and with only two working wheels it could no longer point precisely enough — ending its original mission. (Engineers brilliantly resurrected it as “K2” by using the pressure of sunlight itself as a third virtual wheel.) Hubble’s decades-long gyroscope battles, mentioned earlier, are the same story. This is why satellites carry more sensors and actuators than they strictly need — redundancy is survival.

AI Eyes: How Inspector and Servicing Satellites Navigate by Camera

Everything so far has been about a satellite locating itself against the Earth and the stars. But a new class of satellites faces a harder problem: finding another satellite and flying right up to it.

These are inspector and servicing satellites — spacecraft that rendezvous with another satellite to photograph it, repair it, refuel it, or drag it out of orbit. And here the old tools fall apart. GPS tells you your position, not your target’s. A star tracker tells you which way you’re facing, not how far away that tumbling, uncooperative satellite is or which way it’s spinning. For the last few hundred meters of a close approach, there’s only one good option: look at the target with a camera, and let AI make sense of what it sees.

Vision-based pose estimation. A neural network watches the camera feed and continuously estimates the target’s 6-DOF pose — its three position coordinates and its three orientation angles — frame by frame. It’s conceptually like your phone detecting how far away a face is and which way it’s turned, except the “face” is a tumbling rocket body and the lighting is brutal: blinding direct sunlight on one side, pitch-black shadow on the other, against a star field. Training AI to nail pose under those conditions is the core challenge.

Visual odometry. The same camera stream also tells the chaser how it is moving — by tracking how the scene shifts between frames, it dead-reckons its own motion, like how you sense your speed from the world streaming past a car window.

AI-planned maneuvers — and the orbital paradox. Knowing where the target is, the satellite has to plan how to get there — and orbital motion is wickedly counterintuitive. To catch up to something just ahead of you in orbit, you do not speed up. Firing your thruster forward actually raises your orbit, lengthens your lap time, and makes you fall further behind. To catch up, you fire backward, dropping into a lower, faster orbit to close the gap, then rise back up to meet the target. These backwards-feeling dynamics are exactly why you can’t just hand an astronaut a joystick — an autonomous system that has internalized orbital mechanics plans the burn sequence far better than human intuition.

Why AI, not remote control? Because a human on the ground often can’t fly it: the satellite may be out of contact for most of its orbit, the final approach happens too fast and demands too much precision for a laggy remote link, and for anything beyond Earth orbit the speed-of-light delay makes joystick control impossible. The satellite has to decide for itself, in real time.

This is real and flying today:

• Starfish Space’s “Otter” servicing vehicle uses a computer-vision relative-navigation system called CETACEAN to track a target’s pose in real time, and an autonomous guidance system called CEPHALOPOD to plan safe approach trajectories to satellites that were never designed to be docked with — even tumbling ones — grabbing on with a docking mechanism called Nautilus.

• Astroscale’s ADRAS-J (launched February 2024) flew up to a discarded Japanese rocket upper stage — a non-cooperative object with no GPS, no markers, no help — and inspected it up close (to within about 15 meters) using cameras and angles-only navigation.

• Northrop Grumman’s MEV-1 made history in February 2020 as the first vehicle to dock with a live commercial satellite (Intelsat 901) in geostationary orbit, extending its life.

• And the frontier: Vision-Language-Action (VLA) AI models — like RT-2 (Google DeepMind), OpenVLA (Stanford and UC Berkeley with others), and pi-0 (Physical Intelligence) — fold perception, planning, and motor control into a single neural network. Originally built for robot arms on Earth, they point toward satellites that could be told in plain terms to inspect or repair a target and figure out the rest themselves.

  

The Future: Satellites That Find Themselves

Where is all this heading?

• Fully autonomous navigation. Satellites that determine position and attitude, plan maneuvers, and recover from faults entirely on their own — no ground in the loop.

• Smarter star trackers. AI-powered “lost-in-space” star trackers that can identify their orientation from a single frame even while tumbling, recovering in seconds instead of minutes.

• Formation flying and swarms. Fleets of small satellites holding precise formation relative to each other — flying as a single distributed instrument, like a telescope kilometers wide, which demands relative navigation good enough to keep them centimeters in step.

• Routine servicing. The camera-and-AI rendezvous of today becoming standard: satellites refueled, upgraded, and de-orbited on demand, turning orbit from a graveyard into something maintainable.

• Navigation beyond GPS. As humanity returns to the Moon and pushes deeper, the celestial and pulsar techniques move from experiments to infrastructure — and a dedicated lunar navigation system (a “GPS for the Moon”) is already on the drawing board.

The arc is clear: from Explorer 1 tumbling helplessly in 1958, to satellites that wake up lost, find themselves against the stars, and fly with intention.

Summary: How a Satellite Finds Itself, Start to Finish

1. Two separate questions. A satellite must always know where it is (position) and which way it’s facing (attitude) — different problems, different tools.

2. Position starts with physics. An orbit is a predictable fall; you fix the state occasionally and let the math propagate it forward.

3. Positioning sensors: onboard GPS/GNSS receivers, ground radar/telescope tracking (TLEs), ranging, and inertial dead-reckoning.

4. Position depends on altitude. GPS is excellent in LEO, partial in MEO, a special spillover trick at GEO, and useless beyond — where ranging, celestial, and pulsar nav take over.

5. Stars are a compass, not a map. A star photo gives orientation, not location; for position you need a near reference (celestial nav) or pulsar timing.

6. Attitude sensors: sun sensor, horizon sensor, magnetometer, gyroscope (drifts), and the arc-second-accurate star tracker.

7. Sensor fusion. A Kalman filter blends them — gyro for speed, star tracker for truth — into one best estimate.

8. Actuators turn the satellite: reaction wheels (momentum, no fuel), magnetorquers (push on Earth’s field, LEO only), thrusters (powerful, but burn fuel).

9. The control loop — determine, compare, command, act, repeat — is the satellite’s sense of balance.

10. Pointing modes (Sun, nadir, target, laser link) are why all this exists; bad pointing kills a mission.

11. Safe mode and redundancy keep a confused satellite alive until help arrives.

12. Relative navigation — AI vision estimating another satellite’s pose, plus counterintuitive orbital-mechanics maneuvering — lets inspector and servicing satellites fly right up to their targets.

A satellite is born lost — tumbling, blind, draining its battery. Everything in this post is how it answers two questions well enough to save itself and do its job: position is knowing your dot on the map; attitude is knowing which way you’re looking. A satellite has to nail both, every second, for its entire life.

This is the fourth post in our series on how space technology actually works — following What Is a Satellite?, Brains in Orbit: Edge AI on Satellites, and From Orbit to Your Screen: How Data Gets Down From Space. Up next: how satellites move, maneuver, and eventually die — the story of propulsion, orbits, and the long fall home.

Published by Player One Space | dima@playeronespace.com

The One Sentence That Explains Everything

Here’s the core idea, and everything else in this post is just detail layered on top:

A satellite is a sensor with a radio. It collects data, then shouts that data downward as invisible light (radio waves), and someone on the ground listens with a big ear (an antenna) and writes it down.

That’s it. Sputnik did exactly this in 1957. A $500M Earth-observation satellite does exactly this today. The only things that have changed are how much data we shout, how fast, and on what color of light.

Let’s build up from the simplest possible version to the bleeding edge.

Part 1 — What Does “Sending Data” Even Mean?

Before satellites, let’s get brutally simple. All wireless communication is the same trick: you wiggle something, and someone far away watches it wiggle.

Wave a flashlight on and off — on means 1, off means 0 — and a friend across a field can read your message. That’s it. That’s digital communication. The flashlight is a transmitter, your friend’s eyes are a receiver, and the on/off pattern carries bits.

A bit is the smallest unit of information: a single yes/no, 1/0, on/off. Eight bits make a byte. A short text message is a few hundred bytes. A photo is a few million bytes (megabytes). A high-resolution satellite image is billions of bytes (gigabytes).

The whole game of getting data down from space is: how many of these 1s and 0s can I flash per second, and how reliably will the other side read them correctly? We measure that in bits per second (bps):

• 1,200 bps — a CubeSat beacon (a few sentences of status per minute)

• 1 Mbps (1 million bps) — basic satellite telemetry

• 1 Gbps (1 billion bps) — a modern Earth-observation downlink

• 100 Gbps — a laser link between two satellites

A radio doesn’t use a flashlight — it uses radio waves, which are the exact same thing as visible light, just a color your eyes can’t see. And that’s the next piece.

Part 2 — Radio Is Just Light You Can’t See

Visible light, radio waves, microwaves, X-rays — they’re all the same phenomenon: electromagnetic waves. The only difference between them is frequency: how fast the wave wiggles up and down per second, measured in hertz (Hz).

• 1 Hz = one wiggle per second

• 1 kHz = a thousand per second

• 1 MHz = a million per second

• 1 GHz = a billion per second

Radio for satellites lives roughly between 30 MHz and 40 GHz — far below the frequency of visible light (which wiggles about 500 trillion times per second). We pick radio for space because, unlike visible light, radio waves pass cleanly through clouds, rain, and the atmosphere. You can talk to a satellite in a thunderstorm. (Mostly — we’ll see the limits soon.)

Two facts about frequency drive every design decision in satellite communications, so burn them in:

1. Higher frequency = more room for data. A higher-frequency wave wiggles faster, so you can stuff more 1s and 0s onto it per second. More frequency = more potential speed.

2. Higher frequency = more fragile. Those fast, tight waves get absorbed and scattered by rain, water vapor, and obstacles. Higher frequency = worse weather performance and a narrower, harder-to-aim beam.

That tension — speed versus robustness — is why satellites don’t use just one frequency. They use bands, chosen for the job. Let’s meet them.

Part 3 — The Frequency Bands (and Why Each One Exists)

Engineers carved the radio spectrum into lettered bands. Each band is a trade between speed and robustness. Here’s the working set for satellites, from slow-and-tough to fast-and-fragile:

Notice the pattern marching down the table: as you go faster, the “Toughness” column gets worse. This is the whole story.

A real satellite typically uses several bands at once for different jobs:

• A small S-band radio for TT&C — the satellite’s lifeline. It carries the housekeeping data (”am I healthy? what’s my battery voltage?”) and receives commands from the ground. Low data rate, but rock-solid reliable. You never want to lose this link, so you use a tough, slow band.

• A fast X-band or Ka-band transmitter for the payload data — the actual product (images, sensor readings). High data rate, accepts some weather risk, because if you miss a chunk you can re-send on the next pass.

This split — a reliable slow lifeline plus a fast firehose — is universal, from a university CubeSat to a billion-dollar spy satellite.

Part 4 — How You Put Bits Onto a Wave (Modulation)

A plain radio wave is just a pure tone humming at one frequency — it carries no information. To send data, you have to modulate it: deliberately change the wave in a pattern the receiver can decode. Back to the flashlight: flashing on/off is modulation.

Radios have richer tricks than just on/off. Instead of only flashing brightness, they nudge the wave’s phase (its timing — whether the wave peaks a little early or a little late) or combine phase with amplitude (loudness). Each distinct combination is a symbol, and each symbol can stand for multiple bits:

• BPSK — 2 phase positions = 1 bit per symbol. Simple, extremely robust. Used when the signal is weak.

• QPSK — 4 phase positions = 2 bits per symbol. The general-purpose workhorse for tough links.

• 8PSK — 8 positions = 3 bits per symbol.

• 16-APSK, 32-APSK, 64-APSK — symbols arranged in concentric rings, combining phase and amplitude. 4, 5, or 6 bits per symbol. The dominant scheme in real satellite links today — used in DVB-S2/S2X, the standard behind virtually every modern satellite data downlink.

• 16-QAM, 64-QAM, 256-QAM — also combine phase and amplitude for 4, 6, or 8 bits per symbol, but arranged in a square grid. Fast and spectrum-efficient, but designed for clean linear channels — terrestrial fiber, cable, Wi-Fi.

Why APSK and not QAM for satellites? It comes down to the amplifier. A satellite’s onboard power amplifier runs near its maximum output to squeeze every watt from limited solar power. At that point it becomes nonlinear — it clips and distorts any signal whose amplitude varies widely. QAM’s square grid has symbols at very different distances from the center, so those outer symbols get crushed by the amplifier. APSK deliberately arranges symbols in rings of roughly equal amplitude, giving the amplifier far less variation to distort. You get nearly the same data rate with a transmitter that the satellite can actually run efficiently.

Here’s the intuition for the whole family: imagine sorting marbles into bins. With 4 bins (QPSK) it’s easy to tell which bin a marble landed in even if your aim is shaky. With 64 bins (64-APSK) you cram far more information per throw — but aim must be precise, and for satellites you have to arrange the bins so a slightly shaky arm (the nonlinear amplifier) doesn’t scramble everything. That’s the ring trick APSK uses.

Smart satellites switch modulation on the fly: clear sky and strong signal → step up to 32-APSK or 64-APSK for max speed. Rain or a weak pass → drop to QPSK to stay reliable. This is called adaptive coding and modulation (ACM), and it’s exactly what your home Wi-Fi does when you walk away from the router.

Part 5 — The Hard Limit: Why You Can’t Just Go Infinitely Fast

There’s a famous law that caps how fast any channel can carry data. It’s called the Shannon limit, and in plain words it says:

The speed you can achieve depends on how much spectrum you’re given and how loud your signal is compared to the background noise.

The formula, just so you’ve seen it once:

C = B × log₂(1 + SNR)

• C = max speed (bits per second)

• B = bandwidth (how wide a slice of spectrum you have, in Hz)

• SNR = signal-to-noise ratio (how loud your signal is versus the hiss of the universe)

Two takeaways, no math required:

1. More bandwidth → more speed. This is why high bands (Ka, V) are prized — there’s simply more spectrum available up there to slice off.

2. A louder, cleaner signal → more speed. Boost your power, use a bigger antenna, or get closer, and you can push more bits.

The universe sets the noise floor — there’s always faint hiss from the cosmos, the atmosphere, and the receiver’s own electronics. Your job as an engineer is to make your signal stand out above it. That fight is called the link budget.

Part 6 — The Link Budget: Why Talking to Space Is So Hard

Here’s the brutal physics. Radio waves spread out as they travel, like ripples from a stone. The farther they go, the more thinly that energy is spread. By the time a satellite’s signal reaches the ground, it is breathtakingly faint — often less than a billionth of a billionth of a watt.

The energy spreads out by the square of distance. This is free-space path loss, and at 550 km on X-band it eats roughly 150 decibels — meaning the signal arriving at your dish is something like a quadrillion times weaker than it left the satellite. You’re trying to hear a whisper from someone shouting in another city.

A link budget is the accountant’s tally of every gain and loss along the path, to confirm the whisper still beats the noise. Roughly:

What helps you (gains):

• Transmit power — a louder shout (but satellites have tiny power budgets — often just a few watts for the radio).

• Antenna gain — focusing energy into a tight beam instead of spraying it everywhere. A dish that focuses tightly is like cupping your hands around your mouth. Bigger dish = more focus = more gain.

• A big receiving dish on the ground — a bigger ear catches more of the faint signal.

What hurts you (losses):

• Distance (free-space path loss) — the dominant killer, as above.

• Atmosphere, rain, clouds — especially at high bands.

• Pointing error — if the beam misses the dish even slightly, you lose everything.

• Noise — cosmic background, the Sun, the receiver’s own electronics.

If the gains beat the losses with a margin to spare, you have a link. If not, you slow down (drop to a tougher modulation), get a bigger dish, or fly lower. Every decision in satellite comms is balancing this ledger.

Part 7 — A Short History: From a Beep to a Firehose

The whole arc of this technology is just “shout more bits, faster.” Watch it climb:

1957 — Sputnik 1. The first satellite’s entire “downlink” was a simple repeating beep at 20 and 40 MHz (VHF). It carried almost no data — the timing of the beeps hinted at temperature and pressure. Effective data rate: a few bits per minute. But it proved you could hear a machine from orbit.

1958 — Explorer 1. America’s first satellite carried a Geiger counter and discovered the Van Allen radiation belts — sending science data down on a low-rate radio link. The satellite as a sensor was born.

1962 — Telstar 1. The first satellite to relay live television across the Atlantic. Suddenly the link wasn’t a beep — it was a continuous, high-value data stream. The age of satellites as infrastructure began.

1960s–70s — NASA’s Deep Space Network. To hear whispers from probes near other planets — billions of km away — NASA built 70-meter dishes in California, Spain, and Australia. The lesson: when the signal is impossibly faint, build an enormous ear and slow the data rate to a crawl. Voyager still talks to us this way today, at a few hundred bits per second from 24 billion km out.

1983 — TDRS relay satellites. A clever fix for a stubborn problem (covered in Part 9): NASA put relay satellites in high orbit so spacecraft could talk to the ground continuously instead of only during brief overhead passes.

2000s — X-band Earth observation. Commercial imaging satellites standardized on X-band to downlink hundreds of megabits per second, dumping a full pass of imagery in minutes.

2010s — Ground-station-as-a-service & mega-constellations. Companies stopped building their own dishes and started renting them (more on this in Part 8). SpaceX’s Starlink began launching thousands of satellites.

2020s — Lasers. The newest leap, and the most dramatic: instead of radio, satellites now talk to each other with infrared lasers at tens of gigabits per second. We’ll dig into this in Part 10 — it’s the most exciting change in the field right now.

The throughline: from a few bits per minute (Sputnik) to 200 billion bits per second (Starlink laser mesh) in under 70 years. That’s roughly a trillion-fold improvement.

Part 8 — The Ground Segment: Catching the Signal

Up to now we’ve been in orbit. Now the signal arrives at Earth — and the ground side is half the system. The collection of antennas, receivers, and computers that talk to satellites is called the ground segment or ground station.

A ground station does four things:

1. Tracks the satellite — a LEO satellite races across the sky at ~7.5 km/s, so the dish must physically slew to follow it, like tracking a fast bird with a telescope.

2. Receives the faint downlink and amplifies it.

3. Demodulates and decodes — turns the wiggling wave back into clean 1s and 0s, and uses error-correction (next paragraph) to fix any bits the noise corrupted.

4. Transmits commands up (the uplink), so operators can task the satellite.

Error correction deserves a word. Because the signal is so faint, some bits will arrive wrong. Engineers fight this by sending extra redundant bits using schemes with names like Reed-Solomon, convolutional, turbo, and LDPC codes — a bit like sending “the password is TANGO — that’s T-as-in-Tom, A-as-in-Apple…” so the receiver can reconstruct the message even if static eats part of it. This forward error correction (FEC) is the difference between a usable link and garbage.

The 10-minute problem

Here’s a frustration unique to LEO. A low-orbit satellite is only above the horizon of any given ground station for about 5–10 minutes per pass, and it only gets a handful of useful passes per day over one site. The rest of the time, the satellite is flying over an ocean or a country with no dish, unable to talk to anyone.

So a satellite might spend 90 minutes collecting gigabytes of data, then get a 7-minute window to dump all of it. Miss the window, or fill the buffer, and you’re stuck. This single constraint shapes almost every design choice in LEO — and it’s why two solutions exist: rent more ground stations, or relay through other satellites (Part 9).

Ground-station-as-a-service

You used to have to build your own dishes around the world — hugely expensive. Now you rent them by the minute. Companies like AWS Ground Station and KSAT operate networks of dishes across the globe (including high-latitude sites that see polar-orbiting satellites on nearly every pass) and charge per pass. A startup can downlink from a satellite without owning a single antenna — the data lands directly inside a cloud account, ready to process. This “pay-per-pass” model is one of the quiet revolutions that made small space companies viable.

Part 9 — Two Ways to Beat the 10-Minute Problem

If a satellite can only talk for a few minutes per pass, how do mega-constellations and crewed stations stay connected all the time? Two strategies:

Strategy A — Store-and-forward (the patient way)

The satellite simply buffers data on board like a postal truck collecting mail, then dumps it the next time it flies over a ground station. The data arrives late — maybe 90 minutes later — but for many jobs that’s perfectly fine. A soil-moisture sensor or an asset tracker in the Arctic doesn’t need its reading delivered in milliseconds; “within the next hour or two” is great.

This is cheap, simple, and power-light. It’s the bread-and-butter mode for low-data IoT satellites — including the kind of remote-sensor connectivity small constellations are built around. The trade is latency: your data is always a bit stale.

Strategy B — Relay through other satellites (the instant way)

Instead of waiting to fly over a ground station, the satellite hands its data to another satellite that can see a ground station right now — or that can pass it along a chain until it reaches one. The data hops across the sky and comes down continuously, with no waiting.

NASA pioneered this in 1983 with TDRS — big relay satellites in high geostationary orbit that can always “see” both a low-orbit spacecraft and a ground station, acting as a mirror in the sky. The Space Station and Hubble use TDRS to stay connected almost 100% of the time.

The modern, commercial version of Strategy B is the most exciting development in the field — and it uses lasers.

Part 10 — Lasers: The Game-Changer (Optical Inter-Satellite Links)

For 60 years, satellites talked with radio. Now they’re switching to light — specifically, infrared lasers, the same 1,550 nm wavelength that carries the internet through fiber-optic cables under the ocean. A laser link between two satellites is called an Optical Inter-Satellite Link (OISL), or just “laser ISL.”

Instead of broadcasting radio in a wide cone, a satellite points a pencil-thin beam of laser light at another satellite and holds it steady while both race around the planet. Why bother? Because light at that frequency can carry staggeringly more data:

That’s a thousands-fold speed jump, with no spectrum license needed and near-impossible interception. So what’s the catch?

Aiming. This is the brutal engineering challenge, called Acquisition, Tracking, and Pointing (ATP): you must hold a beam thinner than a pencil on a target moving at 7.5 km/s thousands of km away. Miss by a hair and the link drops. Solving ATP reliably, at low cost, is what took so long — and it’s now finally solved at commercial scale.

What changed in 2024–2026

This stopped being a lab experiment and became something you can buy:

• Starlink’s laser mesh — roughly 9,000 Starlink satellites are stitched together with lasers, moving about 42 million gigabytes per day between them, with intra-constellation links up to 200 Gbps. Traffic crosses entire oceans satellite-to-satellite, never touching the ground — which actually makes it ~40% faster than undersea fiber, because light travels faster in vacuum than in glass and takes a straighter path.

• SpaceX now sells the “Starlink Mini Laser” to other companies’ satellites — a terminal that does 25 Gbps over 4,000 km and lets any equipped satellite plug into the Starlink mesh as an external node. Order it, bolt it on, and your satellite gets near-continuous global connectivity without owning a single ground station. Early adopters (as of 2026) include Muon Space and the orbital-data-center startup StarCloud.

• Independent vendors sell standalone laser terminals for any mission — Mynaric’s CONDOR (up to 100 Gbps), Tesat’s SCOT family (from a 1.7 kg, ~35 W small-sat unit up to constellation-scale terminals reaching 8,000 km). You don’t need SpaceX to go optical.

The market for these terminals is growing ~20% a year — the whole industry is racing to replace radio backbones with light.

The one weakness of lasers

Laser links between satellites are flawless — there’s no air up there to scatter the beam. But a laser link from a satellite down to the ground has a problem radio doesn’t: clouds block light. You can’t shine a data laser through a thunderstorm. The fix is site diversity — build optical ground stations in several dry, clear-sky locations (deserts, high plateaus) so at least one always has a clear view. For satellite-to-satellite links, though, lasers are simply winning.

Part 11 — Once It Hits the Ground: The Data Pipeline

The bits have made it to Earth — through radio or laser, via a ground station or a relay. But raw bits aren’t a map, an alert, or a chart yet. Now begins the data pipeline, the part most people never think about. Here’s the journey from “bits on a dish” to “answer on a screen”:

Step 1 — Demodulate & decode. The ground station’s receiver turns the wiggling signal back into 1s and 0s and runs error-correction to fix corrupted bits. Out comes a clean bitstream.

Step 2 — Frame & packet reassembly. Space data is chopped into standardized packets (the global standard is CCSDS — think of it as TCP/IP for spacecraft). The pipeline reassembles these packets in order, like reassembling a torn-up letter, discarding duplicates and flagging gaps.

Step 3 — Decompress & calibrate. Data is usually compressed on the satellite to save precious downlink bandwidth, so it’s decompressed here. Then it’s calibrated — raw sensor counts become real-world units (a number becomes “412 parts-per-million of methane” or “this pixel is reflecting this much infrared”).

Step 4 — Convert to a usable format. The data is written into standard files: imagery becomes GeoTIFF (a picture that knows its own GPS coordinates), scientific arrays become HDF5 or Zarr (formats built for huge multi-dimensional datasets), telemetry becomes rows in a database.

Step 5 — Store in the cloud. Files land in cloud object storage (e.g. Amazon S3). Because ground-station-as-a-service drops data straight into your cloud account, this can happen seconds after the pass.

Step 6 — Process & analyze. Now the value gets created. This is where the AI we covered in the previous post runs — detecting ships, classifying land cover, flagging changes, spotting anomalies. (Increasingly some of this runs on the satellite before downlink, to shrink what has to be sent — but the heavy lifting still often happens here on the ground.)

Step 7 — Deliver. The finished product reaches the human: a dashboard showing a fleet of assets, an API another company’s software pulls from, an alert texted to a responder, a report filed for regulators. This is the only step the customer ever sees — and everything in this entire post existed to make it possible.

Part 12 — Latency: Why “Real-Time” From Space Is Mostly a Myth

A natural assumption: space is far, so the delay must come from distance. Wrong. Radio waves travel at the speed of light, so the actual flight time of the signal is tiny:

• To a LEO satellite (550 km): about 1.8 milliseconds each way. Blink and you’ve missed it 150 times over.

• To a GEO satellite (35,786 km): about 120 milliseconds each way — noticeable on a phone call, but still under a quarter second.

So if the signal arrives in milliseconds, why does satellite data often reach you hours late? Because the delay isn’t the flight — it’s the waiting and the processing:

• Waiting for a pass — with store-and-forward, the data sits on the satellite until it flies over a ground station. That’s the big one: minutes to a couple of hours.

• Processing on the ground — decoding, calibrating, running analytics: seconds to minutes.

• Scheduling and queuing — your pass competes with everyone else’s for dish time.

This is exactly why lasers and relay networks matter so much: they kill the “waiting for a pass” delay. A satellite on the Starlink laser mesh doesn’t store-and-forward — it hands data off right now and it comes down continuously. That’s the difference between learning about a wildfire in 20 minutes versus 20 seconds.

The rule of thumb:

• Need it cheap and “within an hour or two” is fine? → Store-and-forward on a simple radio. (Most IoT, environmental monitoring, asset tracking.)

• Need it now? → Pay for a laser mesh or relay network. (Disaster response, defense, live broadband.)

Part 13 — Two Complete Journeys, End to End

Let’s tie the whole chain together with two real examples. Follow every hop.

Journey A — A remote sensor reading (the patient path)

1. A solar-powered sensor on a pipeline in northern Canada measures pressure. It packages a few dozen bytes.

2. It transmits that tiny message up on a tough, low band (UHF/L-band) to a small satellite passing overhead.

3. The satellite stores the reading in onboard memory — it’s currently over the Arctic with no ground station in view.

4. ~70 minutes later, the satellite flies over a high-latitude ground station and dumps its buffered readings in a quick downlink.

5. The ground-station-as-a-service provider drops the raw data straight into a cloud bucket.

6. The pipeline decodes, calibrates (”pressure = 642 psi”), and writes a database row.

7. A rule fires: pressure dropped 15% → anomaly. An alert texts the operations team.

8. Total time from sensor to text message: ~75 minutes. Cost per reading: pennies. Perfectly adequate — a slow pressure leak doesn’t need millisecond response.

Journey B — A wildfire detection (the urgent path)

1. An Earth-observation satellite’s infrared sensor captures a frame and, running AI onboard, spots a thermal hotspot consistent with a new fire.

2. Instead of waiting for a ground-station pass, it fires its laser terminal and hands the alert to a neighboring satellite on the mesh.

3. The alert hops across two more satellites and comes down at an optical ground station with clear skies — seconds later.

4. It lands in the cloud, gets cross-referenced with a map, and pushes to a fire-agency dashboard with coordinates.

5. Total time from detection to dashboard: under a minute. The laser mesh collapsed what used to be a 20-minute wait into near-instant.

Same fundamental chain — sensor, radio/laser, link, ground, pipeline, human — but different choices at each stage produce wildly different latency and cost. That’s the entire art of the field: matching the chain to the job.

Part 14 — The Future: An Always-On Nervous System Around Earth

Where is all this heading? Toward erasing the seams entirely:

• Optical everywhere. Radio backbones give way to laser mesh as the default. Satellites become routers in a giant orbital internet, and the “wait for a pass” delay disappears for good.

• Direct-to-cloud. Data flows from satellite to ground station straight into a cloud account and through an automated pipeline with zero human touch — analysis-ready products seconds after capture.

• Process in orbit, send only answers. As on-board AI matures (see the edge-AI post), satellites stop downlinking raw data and downlink decisions — “fire at these coordinates,” not a gigabyte of imagery. This sidesteps the downlink bottleneck entirely.

• Direct-to-device. Your unmodified phone talks straight to a satellite (already beginning with Starlink and others), erasing the line between “cellular” and “satellite.”

• Inter-satellite autonomy. Constellations coordinate peer-to-peer — sharing workloads, rerouting around failures, and tasking each other without waiting for a human on the ground.

The endpoint is a planet wrapped in an always-on data fabric: every remote asset, ship, plane, sensor, and field connected, with information flowing continuously instead of in stolen 10-minute windows.

Summary: The Whole Chain in One Breath

Let’s reassemble everything you just learned, start to finish:

1. Data is just bits — 1s and 0s, measured in bits per second.

2. Bits ride on radio waves — invisible light, chosen by frequency band for a trade between speed and robustness: low bands are slow but tough, high bands are fast but fragile.

3. You put bits on the wave by modulation — more bits per symbol means more speed but less tolerance for noise.

4. The Shannon limit caps your speed based on bandwidth and signal strength, and the link budget is the ledger of gains and losses that decides whether the whisper from orbit beats the noise.

5. A satellite uses several bands at once — a tough slow lifeline (S-band TT&C) plus a fast firehose (X/Ka-band payload).

6. The ground station catches the signal, but a LEO satellite is only visible ~10 minutes per pass — the defining constraint.

7. Two ways to beat it: patient store-and-forward (cheap, late) or instant relay/laser mesh (fast, premium).

8. Lasers now move tens of gigabits between satellites, license-free — the biggest leap in the field.

9. On the ground, a pipeline decodes, calibrates, formats, stores, analyzes, and delivers the data to a human.

10. Latency is dominated by waiting, not flight time — which is exactly why lasers and relays change everything.

A satellite collects data that nobody on the ground could otherwise see. But it’s only as valuable as the chain that gets that data into a human’s hands — through the right band, the right link, the right pass, and the right pipeline.

The sensor sees. The link carries. The pipeline understands. The human acts. That’s how data gets down from space.

This is the third post in our series on how space technology actually works — following What Is a Satellite? and Brains in Orbit: Edge AI on Satellites.

Up next: how a satellite knows where it is and where it’s pointing.

Modern satellites are extraordinary sensors. A single synthetic aperture radar (SAR) satellite can generate 2–5 terabytes of raw imagery per day. A hyperspectral imager captures hundreds of spectral bands per frame. A constellation of Earth observation satellites shooting at full resolution produces more data per hour than a hospital’s entire annual records.

There’s just one problem: you can’t downlink all of it.

Ground station contact windows are short — a typical LEO satellite has 5–10 minutes of contact per pass over a given station. Downlink bandwidth, even with laser optical communications, is improving at roughly 10× per decade. But sensor data volume is growing at 200× per decade.

The math doesn’t work.

On-orbit AI is the answer. Instead of sending raw data to the ground and processing it there, you process it on the satellite — and only send down what matters. AI reduces that 5 TB/day fire hose to a few hundred gigabytes of actionable intelligence. Latency drops from hours to minutes. Storage requirements collapse. And the satellite stops being a dumb pipe and starts being an intelligent edge node.

Use Cases: From Simple Filtering to Full AI in Orbit

Not all on-orbit compute is equal. Think of it as a ladder — each rung requires more processing power, more energy, and a bigger satellite. Here’s the full progression:

Tier 1 — Cloud Detection and Frame Filtering

Complexity: Low | Hardware needed: FPGA or microcontroller

The simplest and most universal application. About 66% of Earth’s surface is cloud-covered at any given moment. Without onboard filtering, an Earth observation satellite wastes most of its downlink quota transmitting white cloud images that are useless to customers. A lightweight neural network or even a classical computer vision algorithm runs on the satellite, detects cloud cover, and simply skips those frames. Only clear-sky imagery gets queued for downlink. Immediate ROI: fewer wasted passes, more useful data per megabyte downlinked.

Tier 2 — Image Compression and Normalization

Complexity: Low | Hardware needed: FPGA or low-power CPU

Neural image compression (replacing legacy JPEG-2000) can achieve 3–5× better compression at the same visual quality. Geometric correction, radiometric normalization, and band calibration can all run onboard, delivering analysis-ready data directly — no ground preprocessing required.

Tier 3 — Land-Cover Classification

Complexity: Medium | Hardware needed: Jetson Nano or Coral TPU

Convolutional neural networks (CNNs) classify each pixel of an image into categories — forest, urban, water, cropland, bare soil — with 92% overall accuracy in published benchmarks. Instead of downlinking a raw image, the satellite downlinks a classified map. Megabytes, not gigabytes. Applications: deforestation monitoring, urban expansion tracking, and agricultural land-use change.

Tier 4 — Change Detection

Complexity: Medium | Hardware needed: Jetson Nano or better

Compare today’s imagery to yesterday’s — or last month’s — and flag only what changed. A construction site appearing overnight. A forest patch that disappeared. A river changing course. 95% recall rates have been demonstrated with multitemporal CNN models. This is the foundation of environmental monitoring, infrastructure surveillance, and disaster response.

Tier 5 — Object Detection: Ships, Aircraft, Vehicles

Complexity: High | Hardware needed: Jetson Orin NX or Jetson AGX Orin

Move from pixel classification to object detection — find specific things in an image and label them. ICEYE, the Finnish SAR satellite company, demonstrated 30-second onboard classification of SAR radar returns into categories like bombers, transport aircraft, air defense systems, and surface ships. This capability is transforming defense and maritime intelligence — instead of analysts reviewing hours of radar data on the ground, the satellite itself tells you what it saw.

Tier 6 — Space Situational Awareness (SSA): Watching Other Objects in Orbit

Complexity: High | Hardware needed: Jetson AGX Orin or dedicated processor

There are currently ~50,000 tracked objects in orbit — active satellites, dead satellites, rocket bodies, and debris fragments. Every object is a potential collision. Traditional SSA relies on ground-based radar networks that track objects and upload conjunction warnings to satellite operators, who then decide whether to maneuver. That loop takes hours.

Onboard AI changes the equation. A satellite equipped with a star tracker or space telescope payload can autonomously:

• Monitor proximity of nearby objects in real time

• Calculate conjunction probability using onboard orbital mechanics models

• Autonomously command a small thruster burn to avoid collision — without waiting for a ground uplink

Companies like LeoLabs, Slingshot Aerospace, and ExoAnalytic Solutions are building AI-powered SSA ground networks today. The next step is moving that intelligence onto the satellite itself, enabling autonomous collision avoidance at orbital speed — where the decision window is sometimes measured in minutes, not hours.

Future vision: constellations that self-coordinate collision avoidance peer-to-peer via inter-satellite links, with no ground involvement at all.

Tier 7 — Ballistic Missile Detection and Trajectory Prediction

Complexity: High | Hardware needed: Radiation-hardened AI processor, large satellite

This is where on-orbit AI intersects directly with national security. The US Space-Based Infrared System (SBIRS) and its successor the Next-Generation OPIR program use infrared sensors in GEO and HEO orbits to detect the heat bloom of a missile launch within seconds of ignition.

Traditional systems use rule-based detection — look for a bright infrared signature matching a known launch pattern. On-orbit AI upgrades this dramatically:

• Classification: Is this a ballistic missile, a space launch vehicle, or a false positive (wildfire, industrial explosion)? ML models trained on thousands of launch signatures can distinguish these in milliseconds.

• Trajectory prediction: Given the launch vector and burn characteristics, compute the likely target region and impact time — before the missile leaves the atmosphere.

• Intercept window calculation: Relay actionable targeting data to ground-based interceptor systems in near real-time.

The US DoD is actively investing in this transition — from rule-based to ML-based missile warning — because AI dramatically reduces false alarms while improving detection confidence. Every second saved in the decision chain matters when the incoming flight time is 20–30 minutes.

Tier 8 — Asteroid, Meteor, and Near-Earth Object (NEO) Tracking

Complexity: High | Hardware needed: Large satellite or space telescope

Earth is constantly bombarded — most of it harmlessly, but ~2,000 potentially hazardous asteroids are known to exist in near-Earth space, and many more are undiscovered. Ground-based telescope networks like Pan-STARRS and Catalina Sky Survey scan the sky nightly. The problem: clouds, daylight, and atmosphere all limit ground coverage.

Space-based AI telescopes change this:

• Continuous 24/7 sky survey from orbit — no clouds, no atmosphere

• Onboard AI distinguishes moving objects (asteroids, meteors) from background stars in real time

• Calculates orbital parameters and impact probability autonomously

• Flags high-priority objects for immediate downlink and ground follow-up

NASA’s NEO Surveyor (launched 2028, infrared space telescope) and ESA’s Hera mission both incorporate autonomous AI navigation and object classification. The vision: a distributed constellation of small AI-equipped space telescopes doing collaborative sky surveys — covering more sky, faster, and cheaper than any single large observatory.

Space weather is the other dimension here. Solar flares and coronal mass ejections (CMEs) can disrupt satellite electronics, GPS signals, and power grids on Earth. On-orbit AI can detect and classify solar events from satellite sensor data, generating automated alerts minutes before the effects reach Earth — critical for protecting infrastructure.

Tier 9 — Full ML Inference Pipeline

Complexity: Very High | Hardware needed: H100-class GPU, large satellite

Full end-to-end ML pipelines running onboard: hyperspectral image analysis across hundreds of spectral bands, multi-sensor fusion (optical + SAR + thermal + infrared), real-time disaster damage assessment, precision agriculture crop stress detection, and methane/carbon emission quantification from industrial sites.

Pixxel’s Pathfinder satellite (200 kg, launching Q4 2026) carries datacenter-grade GPUs and a partnership with Sarvam AI to run foundation models for hyperspectral Earth intelligence onboard. This isn’t just processing an image — it’s running a trained model that understands what the image means.

Tier 10 — Large Language Models and Generative AI in Orbit

Complexity: Cutting Edge | Hardware needed: NVIDIA Vera Rubin Space-1 or equivalent

The frontier: running LLMs, foundation models, and generative AI directly in orbit. VAST Space is building a 15 kW-class satellite bus specifically designed to host NVIDIA’s Space-1 Vera Rubin Module — delivering 25× more AI compute than an H100 for space-based inferencing, with two Rubin GPUs, a Vera CPU, and 1.2 TB/s memory bandwidth. First launch: 2027.

What do you run on an orbital LLM? Autonomous decision-making for complex multi-sensor missions. Natural language tasking interfaces where operators simply say “monitor this region for unusual military activity” and the satellite interprets, plans, and executes. Synthetic data generation for model retraining in orbit.

The Hardware: What Actually Runs AI in Space

Radiation-Hardened CPUs — The Reliable Workhorse

Before AI accelerators, satellites ran on radiation-hardened processors — chips designed to survive the constant bombardment of cosmic rays and energetic protons in space. These aren’t fast. They’re reliable.

BAE Systems RAD750 is the gold standard: 110–200 MHz, 5W power consumption, survives up to 1,000,000 rads of total ionizing dose. It flies on the Mars Curiosity and Perseverance rovers, the New Horizons probe to Pluto, and hundreds of military satellites. Slow by modern standards (about the speed of a 1990s desktop PC), but nearly indestructible.

BAE RAD5545 is the modern successor: quad-core, 466 MHz, 35W, 100 krad tolerance. Capable enough for signal processing and moderate data analysis workloads.

FPGAs — The Flexible Middle Ground

Field-Programmable Gate Arrays are reconfigurable chips that can be rewired in software after launch. They sit between fixed CPUs and full GPUs — more flexible than a CPU for signal processing, more power-efficient than a GPU for fixed-function tasks.

AMD/Xilinx Versal XQR space-grade FPGA: the most capable space-qualified FPGA available today, with integrated AI engines for ML inference. ~5–10W for typical workloads. Used for onboard image processing, AI inference pipelines, and custom algorithm acceleration.

Microsemi RTAX-S: Built-in Triple Modular Redundancy (TMR) — three copies of every computation, majority-vote on the result. Immune to single-event upsets. ~1–5W. The conservative choice for deep-space missions where you can’t tolerate errors.

FPGAs are ideal for Tiers 1–4: cloud detection, compression, and classification at low power.

NVIDIA Jetson — COTS AI That’s Been to Space

Commercial off-the-shelf (COTS) AI modules weren’t designed for space — but they’ve been radiation-tested and are increasingly flying on LEO satellites where radiation doses are manageable.

Jetson Nano: 7–25W, 128-core Maxwell GPU, 472 GFLOPS FP16. Passed Total Ionizing Dose radiation testing beyond 20 krad(Si) — sufficient for 1–2 year LEO missions. Fits in a 3U CubeSat power budget. Entry point for real AI in small satellites.

Jetson Orin NX: 10–25W, 157 TOPS AI performance, 5× improvement over Xavier NX. Radiation-tested for LEO. The current commercial sweet spot for 6U CubeSats and small satellites.

Jetson AGX Orin Industrial: 15–75W (configurable), 248 TOPS, 2048 CUDA cores, 12-core ARM CPU, up to 64 GB RAM. This is the serious workhorse — capable of full ML inference pipelines on smallsats.

Aitech S-A2300 packages the AGX Orin Industrial into a radiation-characterized enclosure specifically for LEO satellite missions — the bridge between a consumer AI module and a space-grade system.

Google Coral Edge TPU — Tiny but Mighty

NASA‘s SC-LEARN (SpaceCube Low-power Edge AI Resilient Node) project integrated Google‘s Coral Edge TPU into a CubeSat-compatible card. The result: trillions of operations per second at just a few watts of power — a remarkable efficiency ratio that makes it ideal for 3U–6U satellites with tight power budgets.

The Coral TPU excels at fixed-function inference: running a trained neural network at high speed and low power. It doesn’t train models — it executes them. Perfect for cloud detection, object classification, and anomaly detection in resource-constrained spacecraft.

NVIDIA H100 in Orbit — The Data Center Goes to Space

In 2026, Starcloud is launching a satellite hosting an NVIDIA H100 GPU cluster — the same chip powering hyperscale AI data centers on Earth, adapted for space with custom thermal management and power systems. At full terrestrial power, an H100 consumes 700W; the space version uses custom engineering to operate within satellite power budgets.

This isn’t a scientific instrument. It’s a commercial orbital compute node — a data center in LEO that customers can rent for AI workloads that benefit from being processed close to where the data is collected.

NVIDIA Space-1 Vera Rubin Module — The Orbital Data Center

The cutting edge of space compute. NVIDIA‘s Space-1 Vera Rubin Module combines two Rubin GPUs with one Vera CPU (88 NVIDIA-designed cores), delivering 25× more AI compute than the H100 for space-based inferencing, with 1.2 TB/s LPDDR5X memory bandwidth. It’s specifically engineered for size-, weight-, and power-constrained satellite environments.

VAST Space has designed its 15 kW-class satellite bus to integrate the Vera Rubin Space-1 Module — creating an orbital data center capable of running large language models, foundation models for Earth intelligence, and full AI training pipelines in space. First VAST launch is planned for 2027.

This is the satellite equivalent of the jump from a pocket calculator to a supercomputer — happening in a single product generation.

Satellite Size vs. Compute: What Fits Where

The satellite you need depends entirely on what compute you want to run. Here’s how the physics maps out:

One important note: power budget is the hard constraint. For every watt of compute, you need roughly one watt of solar generation capacity and one watt of thermal rejection capacity. A Jetson AGX Orin running at 75W requires a satellite with a meaningful solar array area and dedicated radiator panels — that pushes you toward at least a 6U or larger platform.

The Three Hard Problems

Running AI in space sounds straightforward. In practice, three fundamental physics problems make it genuinely difficult.

1. Radiation: The Silent Corruptor

Space is bathed in cosmic rays, solar protons, and trapped radiation in the Van Allen belts. When a high-energy particle passes through a transistor, it can flip a bit — a Single Event Upset (SEU). In a memory chip storing an image, one flipped bit produces a white pixel. In a neural network performing inference, one flipped bit in a weight matrix can silently corrupt the output — and you won’t know it happened.

Modern AI accelerators use sub-10nm transistors that are particularly vulnerable. The smaller the transistor, the less charge it holds, and the easier it is for a cosmic ray to flip it. Standard solutions:

• Triple Modular Redundancy (TMR): Run every computation three times, take the majority vote. 3× power cost.

• Error Correction Codes (ECC): Detect and correct single-bit errors in memory automatically.

• Periodic scrubbing: For SRAM-based FPGAs, periodically rewrite configuration memory to correct accumulated bit flips.

• Radiation-hardened fabrication: Special chip manufacturing processes that make transistors inherently more radiation-resistant. More expensive, less capable.

The Safe-NEureka architecture (a research project) uses a hybrid DMR/TMR approach specifically for deep neural networks in space — detecting when inference results may have been corrupted and triggering recomputation.

2. Thermal: No Air to Carry Heat Away

On Earth, chips cool themselves by heating the air around them. In space, there is no air. Heat can only leave a satellite by thermal radiation from surfaces — a far less efficient process. A satellite’s radiator panels (those flat, often black-painted surfaces you see on spacecraft) are the only mechanism for getting rid of waste heat.

For a small 3U CubeSat, the total radiating surface is a few hundred square centimeters. Running a Jetson Orin NX at 25W in that enclosure is a thermal engineering challenge. Thermal throttling — the processor slowing itself down to avoid overheating — is a real operational concern for onboard AI workloads. Large satellites solve this with dedicated deployable radiators, heat pipes, and active thermal control loops.

3. Power: The Eclipse Problem

Satellites in LEO spend roughly 35% of each 90-minute orbit in Earth’s shadow. During eclipse, solar panels generate nothing. All power comes from batteries. Running a demanding AI inference job during eclipse drains batteries faster, potentially leaving insufficient power for other critical systems.

This means onboard AI workloads must be scheduled around the power timeline — run heavy compute during sunlit periods, throttle back during eclipse, ensure battery reserves for communications and attitude control. Every AI workload has to earn its power budget against competing satellite functions.

Companies Leading the Way

The on-orbit AI industry has moved from research papers to operational satellites in the span of five years. Here are the organizations at the frontier:

Satellogic holds the longest commercial heritage — 13 years of flying GPUs in space. Their next-generation Merlin constellation (projected launch October 2026) is an AI-first platform with onboard edge processing for real-time imagery analysis. They were awarded a $30M contract for AI-first constellation services, validating the commercial model.

Loft Orbital‘s YAM-9 (launched November 2025) was the first commercial satellite to host four networked compute units creating a heterogeneous processing environment onboard. Their “Virtual Missions” platform lets customers deploy and update AI models on the satellite after launch — treating the satellite like a software platform. They are planning the world’s first operational AI/edge-compute constellation, and have partnered with Helsing (defense AI) for a multi-sensor constellation for European defense and security.

Pixxel (India) is building the most ambitious near-term orbital compute capability: a 200 kg Pathfinder satellite (Q4 2026) carrying datacenter-grade GPUs, partnered with Sarvam AI to create India’s first orbital data center. Goal: real-time hyperspectral AI inference in orbit.

ICEYE (Finland/US) demonstrated practical defense AI: 30-second onboard SAR classification of military targets — from satellite tasking to labeled intelligence — a capability that transforms tactical reconnaissance.

D-Orbit partnered with SkyServe STORM for on-orbit ML inference and with AWS for the first demonstration of cloud ML services running on an orbiting satellite. Their ION platform is effectively a software-defined satellite bus with compute as a service.

VAST Space is building the most powerful orbital compute platform yet — a 15 kW bus designed to host NVIDIA’s Vera Rubin Space-1 Module, with a confidential anchor customer for 4 satellites and options for 200+ more. First launch: 2027.

Kepler Communications (Canada 🇨🇦) currently operates the largest active orbital compute cluster in existence. In March 2026, they commissioned distributed cloud infrastructure across their optical constellation — 40 NVIDIA Jetson Orin modules spread across 10 satellites in their “Aether” Tranche 1 series, linked by optical inter-satellite links (OISLs). Unlike most companies on this list, this is live and open for commercial AI workloads today, not projected. Kepler is also the only major player in this space headquartered in Canada.

Starcloud (YC-backed 🇺🇸) launched Starcloud-1 in November 2025 with an NVIDIA H100 GPU and ran the first LLM inference in space in December 2025. Their Starcloud-2 (2027) upgrades to Blackwell GPUs + AWS Outposts integration, and they have filed FCC paperwork for an 88,000-satellite constellation. The purest orbital data center company on this list — no Earth observation, no IoT, just compute-as-a-service from orbit.

Orbital (a16z-backed 🇺🇸) is launching Orbital-1 on a Falcon 9 in April 2027, carrying NVIDIA Vera Rubin Space-1 GPUs. Their model is specifically distributed AI inference — each satellite is a node in a parallel inference cluster, solar-powered and radiatively cooled. A direct competitor to VAST Space but focused purely on inference-as-a-service rather than general orbital data center operations.

Google (🇺🇸) is entering orbital AI through Project Suncatcher, a joint initiative with Planet Labs to launch two prototype satellites by early 2027. Crucially, Google is using TPU chips (not GPUs) — the same custom AI accelerators powering Google’s data centers on Earth. This is a significant architectural bet: TPUs offer better performance-per-watt for inference workloads but require Google’s software stack. CEO Sundar Pichai has publicly called orbital data centers “the new normal within a decade.”

SpaceX / SpaceXAI (🇺🇸) filed FCC paperwork for SpaceXAI — an orbital computing constellation of up to 1 million satellites with optical inter-satellite links, targeting 100 GW of AI compute capacity. Elon Musk’s acquisition of xAI further signals intent to combine the world’s largest satellite network (Starlink) with AI infrastructure at unprecedented scale. Timeline is years away, but the Starlink orbital infrastructure advantage and the stated ambition make this impossible to ignore as the long-term market shaper.

Zhejiang Lab / ADAspace (China 🇨🇳) launched the first 12 satellites of their “Three-Body Computing Constellation” in May 2025 and has already demonstrated in-orbit AI model deployment and inter-satellite networking. The full plan calls for 2,800 satellites delivering 1 quintillion operations per second — the most ambitious national-scale orbital compute program outside the US, and already partially operational. Named after Liu Cixin’s sci-fi novel, this is China’s direct answer to US orbital AI dominance.

Axiom Space (🇺🇸) is taking a unique approach — deploying an Orbital Data Center (ODC) Node on the ISS in partnership with Spacebilt, Skyloom (optical comms), Phison, and Microchip Technology. It provides AI/ML compute, cloud storage, and processing accessible to satellites in LEO and astronauts onboard. The only station-hosted compute node, serving a dual defense/commercial role — and a proof point that orbital compute doesn’t have to mean free-flying satellites.

The Future: What Comes Next

On-orbit AI is in its early innings. Here’s what the next decade looks like:

Autonomous satellite-to-satellite coordination. Constellations will develop peer-to-peer AI networks — satellites talking directly to each other via inter-satellite laser links, sharing workloads, distributing inference, and coordinating maneuvers without ground involvement.

On-orbit model training. Today’s satellites run inference on pre-trained models. Tomorrow’s will retrain and fine-tune models using new data collected in orbit — a continuously improving AI that gets smarter with every pass. Pixxel has explicitly stated this as a goal.

AI-guided in-orbit servicing. Robotic satellites will use AI vision and manipulation to autonomously rendezvous, dock, and service other satellites — refuelling them, replacing failed components, and upgrading obsolete hardware. Northrop Grumman’s MEV-1 already demonstrated docking; the next step is AI-autonomous servicing at scale.

Climate and methane monitoring. Real-time on-orbit AI for carbon/methane emission quantification from oil fields, pipelines, and industrial sites. Every barrel equivalent of leaked methane can be detected, attributed, and reported automatically — enabling real carbon accountability at a global scale.

Precision agriculture from orbit. AI-driven crop stress detection, yield prediction, irrigation guidance — downlinked not as images but as actionable farm management recommendations, generated onboard.

IoT edge intelligence. Even small LEO IoT satellites — not just large Earth-observation platforms — can run lightweight AI. A 6U CubeSat collecting telemetry from remote sensors can detect anomalies onboard and only alert the ground when something unusual happens, rather than streaming every data point.

Summary

The satellite is no longer just a sensor pointed at Earth — it’s becoming an intelligent platform that thinks, decides, and acts in orbit.

The progression is clear: from simple cloud filtering on a 3U CubeSat drawing 10 watts, to LLM inference on a 15 kW orbital data center. From detecting deforestation to predicting missile trajectories. From managing collision avoidance to tracking asteroids that could threaten Earth.

The key insight: the bottleneck in space is no longer getting hardware to orbit — launch costs have collapsed. The bottleneck is now intelligence on orbit. The satellites that win the next decade won’t be the ones with the best sensors. They’ll be the ones with the best AI.

Hardware on orbit × Intelligence on orbit = the commercial space economy.

What Is a Satellite?

A satellite is any object that orbits another object due to gravity. The Moon is Earth’s natural satellite — it has been circling our planet for about 4.5 billion years. But when most people say “satellite,” they mean artificial satellites: human-made machines we deliberately launch into orbit to do useful work.

That work ranges from relaying TV signals and phone calls, to predicting hurricanes, to guiding your car’s GPS — and increasingly, to connecting remote sensors in places no cell tower can reach.

The First Satellite: Sputnik 1

On October 4, 1957, the Soviet Union launched Sputnik 1 — the first artificial satellite in history. It was a polished metal sphere just 58 cm across, weighing 83.6 kg, with four whip antennas trailing behind it. It transmitted a simple radio beep.

That beep changed the world.

Sputnik completed one orbit every 96 minutes at altitudes between 215 and 939 km. It stayed in orbit for three months, then burned up re-entering the atmosphere in January 1958. Its only “payload” was a radio transmitter — but its political and scientific impact was enormous, triggering the Space Race between the US and USSR.

Just four months later, on January 31, 1958, the United States launched Explorer 1, its first satellite — and it immediately made a scientific discovery: the Van Allen radiation belts surrounding Earth.

A Brief History of Satellites

The evolution follows a clear arc: from one-off national prestige projects → scientific instruments → commercial infrastructure → today’s software-defined, rapidly-replaceable commercial spacecraft.

How Many Satellites Are Up There Right Now?

As of early 2026:

• ~10,000+ active satellites in orbit

• ~50,000+ tracked objects total (including rocket bodies and debris)

• SpaceX Starlink alone: ~7,000+ satellites — one company now operates the majority of active spacecraft in history

• Planet Labs: ~200+ Earth-imaging Doves — the largest Earth observation constellation

• New satellites are being launched at a rate of hundreds per month

This is a dramatic inflection point. From 1957 to 2019 (~62 years), humanity launched roughly 8,000 objects total. In the 2020s alone, we’re on pace to launch that many again — in under a decade.

Key Systems Inside Every Satellite

No matter the size — from a 1U CubeSat to the Hubble — every satellite needs the same core subsystems to survive and operate in space.

1. Power (EPS — Electrical Power System)

The primary source is almost always solar panels (photovoltaic cells) that convert sunlight into electricity. Batteries store energy for when the satellite passes through Earth’s shadow. Some deep-space or nuclear-powered satellites use RTGs (Radioisotope Thermoelectric Generators) — like Voyager or Cassini.

2. Structure (Bus)

The physical frame holding everything together. Must survive launch vibrations (~20g), thermal cycling (from -150°C to +150°C every 90 minutes), and radiation. Usually aluminum, titanium, or composite.

3. Attitude Determination and Control (ADCS)

The system that points the satellite correctly. Uses a combination of:

• Reaction wheels (spinning gyroscopes that rotate the satellite without fuel)

• Magnetorquers (electromagnetic coils interacting with Earth’s magnetic field)

• Thrusters (for larger corrections)

• Star trackers, sun sensors, GPS (for knowing where you’re pointing)

4. Communications (Comms)

Radio antennas and transceivers for sending data to Earth (downlink) and receiving commands (uplink). Frequency bands vary: UHF/VHF for small sats, S/X/Ka-band for high-data-rate systems. Ground stations on Earth complete the link.

5. On-Board Computer (OBC)

The satellite’s “brain” — runs software, manages subsystems, processes commands, handles fault detection and autonomous recovery.

6. Thermal Control

Space has no air, so heat can’t convect — it only radiates. Satellites use passive thermal control (insulating blankets, gold foil, black paint) and sometimes active heaters or heat pipes to stay within operating temperature ranges.

7. Propulsion

Not all satellites have it. When present, options include:

• Cold gas thrusters (simplest, low thrust)

• Monopropellant / bipropellant chemical (medium thrust)

• Ion / Hall-effect electric propulsion (very efficient, low thrust, ideal for LEO station-keeping and constellation management)

• Plasma thrusters (emerging for CubeSats)

8. Payload

The actual mission instrument: a camera, a radio transponder, a scientific sensor, a GPS receiver, an AI processor — whatever the satellite was built to do. The bus exists to keep the payload alive and pointed correctly.

Types of Satellites by Size

One of the most exciting developments in the last 20 years is the miniaturization of satellites. What once required a school-bus-sized machine can now be done with something you can hold in your hand. Here’s the full spectrum:

FemtoSat (< 0.1 kg)

The smallest class. Experimental chips or wafer-scale satellites — some literally the size of a cracker. Still mostly research prototypes. Examples: KickSat Sprites (5g each, 3.5 × 3.5 cm). Capability is extremely limited but the concept is proving out — thousands could theoretically be deployed like a cloud.

PicoSat (0.1 – 1 kg) — including PocketQube

Slightly bigger. The PocketQube standard (5 cm × 5 cm × 5 cm cubes, labeled P: 1P, 2P, 3P) is the dominant picosat form factor. A 1P weighs roughly 100–200g. Useful for basic radio experiments, amateur astronomy, and technology demonstrations. Launch costs can be under $10K.

NanoSat (1 – 10 kg) — including CubeSat

The CubeSat revolution. The CubeSat standard (defined in 1999) uses a base unit of 1U = 10 × 10 × 10 cm (about 1 kg). Satellites come in 1U, 2U, 3U, 6U, 12U configurations. A 3U CubeSat (roughly the size of a large water bottle) can carry a useful camera, radio, or sensor.

CubeSats democratized space. Universities, startups, and national space agencies in developing countries can now afford to build and launch real spacecraft. Companies like Planet Labs built entire Earth-observation businesses on 3U “Dove” CubeSats.

Why CubeSats matter:

• Standardized = cheaper to build, faster to design

• Fits on rideshare launches with many others

• 3U launch cost: ~$30K–$100K depending on orbit

• 6-month to 2-year development cycles vs. 5–10 years for traditional satellites

MicroSat (10 – 100 kg)

Bigger than CubeSats but still small by traditional standards. Often custom form factors. Can carry higher-resolution cameras, more powerful radios, or specialized payloads. Companies like Spire Global, HawkEye 360, and early Planet SkySat satellites fall here. Typical launch cost: $500K–$2M.

MiniSat / SmallSat (100 – 500 kg)

The “small satellite” category as commonly used in industry refers to this range. Can carry serious imaging payloads, radar systems (SAR), or communication transponders. Planet SkySat, some OneWeb satellites, and many commercial Earth observation platforms live here.

Medium Satellite (500 kg – 1,000 kg)

Traditional workhorse class. GPS Block II satellites (~845 kg), many weather satellites, medium communication relay satellites.

Large / Heavy Satellite (1,000 kg+)

Traditional GEO communication satellites, military reconnaissance birds, and scientific spacecraft like Hubble (~11,000 kg) and the James Webb Space Telescope (~6,500 kg) live here. Cost: $300M–$1B+. Build time: 5–10 years. These are being replaced in many use cases by constellations of smaller satellites.

Types of Satellites by Purpose

Beyond size, satellites are classified by what they do:

Communication Satellites (ComSats)

Relay signals — TV, phone, internet, maritime/aviation communications. Two main orbits:

• GEO (Geostationary, 35,786 km): Appears stationary above one point on Earth. Three satellites can cover the whole planet. High latency (~600ms round-trip). Traditional TV/telecom.

• LEO constellations (500–1,200 km): Low latency (~20–40ms). Starlink, OneWeb, Amazon Kuiper. The new paradigm.

Earth Observation (EO) Satellites

Image Earth’s surface for agriculture, urban planning, military intelligence, disaster response, climate monitoring. Resolution has improved from 10-meter (Landsat in the 1970s) to 30-centimeter commercial imagery (Maxar today). Planet Labs images the entire Earth’s landmass every day with their Dove fleet.

Navigation / PNT Satellites

GPS (US), GLONASS (Russia), Galileo (EU), BeiDou (China), NavIC (India), QZSS (Japan). These constellations provide positioning, navigation, and timing signals that every smartphone, autonomous vehicle, aircraft, and ship depends on. Canada is the only G7 nation without a sovereign PNT system — an opening for new players.

Weather / Meteorological Satellites

GOES (US geostationary weather), NOAA POES (polar-orbiting), Meteosat (European) — these are the satellites behind every weather forecast you’ve ever seen. They measure cloud cover, sea surface temperature, atmospheric moisture, storm intensity.

Scientific / Research Satellites

Hubble, James Webb, Chandra (X-ray), Fermi (gamma-ray), GRACE (measuring Earth’s gravitational field), DSCOVR (solar weather). One-of-a-kind missions designed for pure science.

IoT / Machine-to-Machine (M2M) Satellites

The newest and fastest-growing category. Small LEO satellites specifically designed to collect tiny data packets from sensors in remote locations — mines, pipelines, ships, wildlife trackers, agricultural sensors. Myriota, Kineis, Astrocast, Lacuna Space — and new entrants like Player One Space — are building constellations specifically for this.

AIS / Maritime Domain Awareness

AIS (Automatic Identification System) transponders are mandatory on large ships. Space-based AIS receivers on LEO satellites collect ship position data globally, enabling maritime tracking even in the middle of the Pacific. exactEarth (acquired by Spire), Orbcomm, and Spire Global sell this data to coast guards, insurers, and logistics companies.

ADS-B / Aviation Surveillance

Similar to AIS but for aircraft. Ground-based ADS-B has coverage gaps over oceans. Aireon (on Iridium satellites) provides global aircraft tracking from space. Critical for oceanic flight management.

Military / Reconnaissance Satellites

Early-warning missile detection (DSP/SBIRS), signals intelligence (SIGINT), imagery intelligence (IMINT), and communications for military operations. Often classified. The most capable imaging satellites can reportedly resolve objects under 10 cm from 400+ km altitude.

Technology Demonstration Satellites

Small satellites launched specifically to test new technologies before committing them to an expensive operational mission. CubeSats are ideal for this — fail cheaply, iterate fast.

Orbits: Where Satellites Live

Orbit altitude determines what a satellite can do, how often it passes overhead, and how long signals take to travel.

The CubeSat Standard — A Closer Look

Because CubeSats represent such a paradigm shift, they deserve their own section.

The standard:

• 1U = 10 × 10 × 10 cm, ~1 kg

• 3U = 10 × 10 × 30 cm, ~3 kg (most common)

• 6U = 10 × 20 × 30 cm, ~6 kg (current commercial sweet spot)

• 12U = 20 × 20 × 30 cm, ~12 kg (large CubeSat)

• 27U = 30 × 30 × 30 cm, ~24 kg (emerging)

Why the standard matters: Standardization means off-the-shelf radios, batteries, solar panels, OBCs, ADCS modules, and deployers are all interoperable. You can buy a CubeSat kit like a PC — choose your components, integrate, test, launch. This dropped development costs from $100M+ (traditional satellite) to $50K–$500K (6U CubeSat).

Real-world CubeSat examples:

• GOES-16 Compact Comms Demo — 6U, NASA

• Planet Labs Dove — 3U, commercial Earth imaging

• Spire LEMUR-2 — 3U, weather/AIS/ADS-B

• LightSail 2 — 3U, solar sail propulsion demo

• CAPSTONE — 12U, lunar pathfinder for the Gateway station

The Future: What Comes Next

The satellite industry is in the middle of a transformation that looks a lot like what happened to computing in the 1980s — from mainframes to PCs.

Key trends:

• Megaconstellations — Starlink, OneWeb, Amazon Kuiper, and China’s SatNet are deploying thousands of satellites to provide global broadband. By 2030, there could be 100,000+ active satellites.

• Software-defined satellites — Satellites that can be reprogrammed in orbit to change their mission, frequency bands, or beam patterns. The satellite becomes a flexible radio/compute platform, not a fixed instrument.

• On-orbit AI — Processing data on the satellite before downlinking (edge AI). Dramatically reduces the data volume that needs to be transmitted. Instead of sending raw imagery, send “I found a wildfire at these coordinates.”

• Satellite IoT — The next frontier of connectivity. LEO satellites specifically designed to collect micro-data from billions of sensors in places no cellular network reaches. Growing at high-teens CAGR.

• In-orbit servicing — Robots that can refuel, repair, or upgrade satellites in orbit. Northrop Grumman’s MEV-1 has already successfully docked with and extended the life of an aging GEO satellite.

• Debris management — With ~50,000+ tracked objects, orbital debris is a real problem. Active debris removal, better deorbit systems on new satellites, and international regulations are all developing rapidly.

Summary

Satellites went from a 58cm metal sphere beeping over the Soviet Union in 1957 to 10,000+ active machines orbiting Earth today, doing everything from guiding your phone’s GPS to imaging every acre of farmland daily to connecting sensors in Arctic shipping lanes.

The key insight: satellites are no longer special. A university team can build and launch one. A startup can operate a constellation. A remote sensor in the Sahara or on the deck of a ship 2,000 km from the nearest coast can now transmit data reliably, affordably, and regularly.

We are in the early innings of a world where everything connected to the internet isn’t connected via a tower on the ground — but via a satellite overhead.