Prática de Shadowing: The Biggest Problem No One Is Talking About - Aprenda a falar inglês com vídeo

This is Earth today, around 15,000 satellites in orbit, and two-thirds of them belong to a single company, SpaceX.

From the ground on a really dark night, you can maybe catch two or three of them drifting past.

Now, this is what your night sky might look like by 2050, because in January, SpaceX asked for permission to launch up to a million brand new satellites. And it's not for internet, it's to run AI. And they're not the only ones.

Blue Origin, Amazon, China and a handful of others have all filed for their own plans.

If you add them all up, there are over 1.7 million spacecraft waiting for a signature to go up. If even a fraction of this happens, you'll look up at night and see more machines crossing your sky than actual stars.

The Milky Way will become a distant memory.

And if two of them ever clip each other up there, it could set off a chain reaction that traps every one of us down here on Earth for good.

Okay, this all sounds pretty scary, so why are they actually doing it? It's really because AI is hungry.

This is an AI data centre named Colossus, just outside of Memphis, Tennessee.

Inside of it, you would find around 200,000 of the most powerful computer chips on the planet, all organised in racks like this.

Each rack holds about 15,000 times more computing power than your laptop and draws as much electricity as roughly 70 family homes.

XAI, Elon Musk's AI company, which is also now a part of SpaceX, built the whole thing in just 122 days.

And the grid wasn't ready.

They needed enough power for 250,000 family homes, and that just wasn't possible.

So their solution was buy a power plant from overseas, ship in 35 gas turbines, and run it themselves.

Most of them didn't even have permits.

Colossus exists to train artificial intelligence, and it needs immense amounts of power, more than the grid has to offer.

And it's not the only one.

Every time you ask an AI chatbot to write an email or make an image, somewhere in a building like Colossus, a machine like this lights up and pulls electricity out of the grid.

Add all of them together and the data centres that are powering this AI revolution burn through about 415 terawatt hours of electricity in 2024 24 alone. That's more than the entire United Kingdom used in the same year.

About 1.5% of all of the electricity used on Earth. And it's climbing really fast.

By 2030, that number is expected to more than double, and the grid simply won't keep up.

You can build a data centre like this in about a year, but a new power plant takes up to five years to build, and the transmission lines to actually carry that power across the country, up to a decade.

For companies that are moving as fast as these AI companies are, that's like an eternity.

In Ireland, around Dublin, it got so bad that the grid operators just stopped letting new data centres connect.

A flat out ban because they were simply scared of blackouts. And it's not just energy.

There's another huge problem.

Heat. All of those computers run extremely hot, and to stop them from cooking themselves you have to cool them all down.

down. And the cheapest way to do that is with water.

A single big data centre can drink up to about 5 million gallons of water per day.

That's enough for tens of thousands of people.

The utility of AI and the rate at which it's improving is undeniable, but honestly it's also kind of scary. A video about that is coming soon.

This is the wall that the AI industry is driving straight into. Demand keeps climbing but there's no way to keep up. And so a few people looked at that problem and thought, if there's no way we can keep up from down here on the ground, why don't we go up into space?

In space, the sun never sets, there's no clouds or atmosphere to block the sunlight, no neighbours filing complaints, no land permits, just unlimited sunlight 24 hours a day waiting to be collected.

it. And when you put it like that, the idea does sound appealing. Why fight over power lines and pump entire rivers through data centres down here when you could just take the computers themselves into space? And so earlier this year SpaceX filed a real plan to do exactly that. Jeff Bezos is on record saying that we will be building gigawatt AI clusters in space over the next couple decades, because up there the sun is always shining.

And this fixes the heat problem too, right?

I mean space is supposedly really cold.

Out there, it's like minus 270 degrees Celsius.

That's nearly as cold as you can physically get.

Surely you can just park your computers up there and they'll run nice and cool forever.

It sounds great, but as we're about to find out, space isn't cool like a fridge in the way you'd think, and it might actually make keeping these computers cool even harder.

On January 30th, 2026, SpaceX formally filed for up to a million satellites that would work as solar-powered data centres in orbit. This is the future that they're imagining.

A few years from now, you wake up to footage of a SpaceX Starship climbing out into Earth's orbit.

By now, this is commonplace.

Starship has been a fully operational rocket since 2027, and was even used in NASA's Artemis III mission to prepare for landing on the moon. But this one is different.

It isn't carrying people, or starlinks, or even fuel. This one is carrying a single satellite. Just one.

It reaches orbit, deploys, and as Starship starts to pull away, the thing begins to unfold.

And keeps unfolding, and unfolding, until it's stretched about 70 metres from tip to tip, roughly the length of a Boeing 747. That whole thing is a computer, a data centre flying 600 kilometres above your head, running on pure, unfiltered sunlight.

This is SpaceX's AI-1 satellite, and it isn't science fiction, this is the real future that SpaceX has actually filed for.

These are the specs that Elon Musk stood on stage and presented.

So let's just take them at face value and see what that future would actually look like.

The AI-1 is 70 metres wide.

These are the solar panels.

In the middle you'd find a computer.

120 kilowatts of AI chips.

That's about one rack of the servers from the colossus data centre that we just talked about.

And this is the radiator.

It's built to cool the whole thing down.

Their plan is to put a million of these in orbit in shells between 500 and 2,000 kilometres above the surface.

All of them will be connected, talking to each other through laser links and completely powered by sunlight. But this thing is huge.

Here's a normal Starlink we have today next to the AI-1. The AI-1 weighs nearly two tons, about two and a half times more than the Starlink.

So a million of them would weigh around two million tons, all of it needing to be flown out into orbit. In the entire history of space flight from Sputnik in 1957 all the way right up to today, everything we've ever launched, every rocket, every probe, every satellite, every piece of every station, all of it adds up to about 20,000 tonnes.

That's just 1% of all of the mass that SpaceX would need to launch for this constellation.

And that's only to launch them once.

These satellites will wear out, radiation chews through their chips in about 5 years, so to keep the constellation alive you're basically launching replacements forever, something like 4,000 starship loads every single year.

That's 11 launches a day, one every few hours, forever. And there's one particular sentence in their proposal that really got me curious.

SpaceX wrote that launching this constellation is, and I quote, "...a first step towards becoming a Kardashev Type II civilisation, one that can harness the Sun's full power. Now the Kardashev scale is a real idea.

It's from a Soviet astronomer back in 1964.

It ranks civilisations by how much energy they can command. A type 1 controls all of the energy of its home planet. A type 2 controls the energy output of its star, every last photon. And a type 3, the whole galaxy.

It's the scale that we use to picture what a truly advanced civilisation would really look like.

Right now humanity sits at about a 0.73 on the scale. We haven't even mastered our own planet's energy yet.

But this, SpaceX says, is the first step. And if they did get all of these million satellites up there, they would end up with about 100 gigawatts of AI compute in orbit.

Around twice what all of the AI data centres on Earth can do right now. Sounds pretty promising, but let's talk about the big problem we haven't addressed yet.

Heat. Did you know that data breaches jumped by 211% between 2023 and 2024? And every one of those breaches is another chance for your personal information to be exposed online.

Your name, your address, your habits, all making you an easy target for identity theft and scams.

And that's exactly where today's sponsor comes in.

Incogni hunts down your personal data and gets it removed, scouring online directories and commercial databases to clear you off them. Their fully automated process strips your information from hundreds of data brokers without you ever having to lift a finger. But here's the part that I really like.

Incogni has a custom removals feature so you can go deeper yourself. Just point them at specific websites that you've found exposing your details, submit them and their team will work to get your data taken down from almost anywhere. And this genuinely matters.

Some health insurers can actually raise your rates based on your online activity, right down to the data pulled from health apps. Custom removals helps you shut that door.

So if you want to take your personal data back, head to Incogni and use my code ASTROKOBI at checkout to get 60% off any annual plan.

Just click the link at the top of the description to see all of their options, and a huge thanks to Incogni for supporting the channel.

Everyone assumes that because space is freezing cold it's the perfect place to put a data centre, but space isn't cold in the same way that your freezer is.

Think about how everything you own gets rid of heat.

The fan in your laptop blows the hot air away, that's convection. If you rest a hot pan on the counter, the heat drains into the counter, cooling the pan, that's conduction.

Both of these need some kind of matter, air, metal, molecules, something physical to carry the heat from the hot place to the cold place.

In the vacuum of space, there is no stuff, no air to blow, no cold countertop to touch.

A computer in orbit can only lose its heat one way, by glowing.

Every warm object glows, I mean you're glowing right now, giving off infrared heat.

If you've ever seen footage from a thermal camera, that's exactly what you're looking at.

In space, that faint glow is the only exit that heat has. Take the ISS, every computer, every light, every life support system that generates heat, and if you add all of it up, it's about 70 kilowatts that has to be dumped.

That's still less than a single one of Nvidia's top AI server racks, and just to get rid of that, the station needs those huge white wings that you always see in the photos.

156 square metres of radiators, that's about 7 tonnes of material.

Now let's look at SpaceX's AI-1.

It's got 120kW of computing power running inside of it, and that means 120kW of heat to get rid of.

So look at the design.

You'd assume that these two huge wings would be the radiators, but they're not. Those are the solar panels, around 600 square metres of them, soaking up the sunlight to power the computer. The cooling is this thin strip running down the middle.

It's a single radiator, 110 square metres of panel.

But importantly, it glows from both sides, front and back, so it's really 220 square metres of surface, all throwing off heat into space. And the way it works is closer to your car than you'd probably think.

Liquid coolant gets pumped right past the chips which soaks up the heat and then carries it out to the fins.

The panel then spreads that heat across its whole surface and glows it away as infrared, and the cooled liquid loops back to do it all again.

So is 220 square metres… actually enough? If you run it at a comfortable room temperature, say 25 degrees, a real radiator glows away about 400 watts per square metre at that temperature, so the whole panel sheds about 90 kilowatts.

And that's a problem because we need 120.

At room temperature you just can't keep up.

But it doesn't have to run at room temperature.

It can get hot. And running hot changes everything. As a surface heats up the heat it glows away doesn't just climb steadily.

It climbs with the fourth power of the temperature.

A little bit hotter and a lot a bit more heat gets thrown off. Warm that panel to 48 degrees Celsius and it now handles the full 120 kilowatts. 48 degrees is perfectly comfortable for modern silicon chips.

Push it to 61 and it copes with the AI-1's 150 kilowatt peak.

And that's the whole trick that they're having to rely on.

By running the system hot, you dump far more heat through radiation. The most inefficient way to move heat there is without having to build a much bigger radiator. SpaceX is even designing a special chip that runs extremely hot on purpose, so they can pack more computing into each satellite without having to make it bigger and harder to launch.

Okay, so does any of this actually work?

Well, in theory, yes.

If SpaceX could build the satellite that they've announced, there's no law of physics that says it would overheat and die. A startup called StarCloud has already flown a real Nvidia H100 in orbit and trained a small AI model up there. China's already got clusters of AI satellites talking to each other with lasers.

The physics of cooling is hard, but it is understood, and people are running early versions of this right now. But there are two even deeper problems. The first is whether this thing would even last. Space is completely soaked in radiation, and radiation is poison for cutting edge AI chips. A single cosmic ray can flip a transistor, corrupt a calculation, or just slowly destroy the silicon.

You can build chips that are specifically hardened for space, but those are years behind the ones that are already in your phone. Radiation will wear these computers down in about five years and you can't exactly send someone up to swap a graphics card.

So every few years you'd have to launch your entire AI data centre all over again.

And the second big problem is the one these AI companies know lots about, money.

Ultimately if it's just too expensive none of this happens. Google even ran the numbers for orbital compute and they found that for data centres in space to cost about the same ones as the ones here on the ground, you'd have to get a kilogram of hardware up there for about $200. Today at best, that number is about $2700 on a Falcon 9. For this to work, they need it 13 times cheaper than it is right now. And all of that hinges on one rocket, Starship. It's massive, and if it hits expectations, it drops launch costs to below $100 a kilogram.

That's cheap enough to make this whole idea work.

But as I'm making this video, Starship has still never reached orbit, and I have no doubt that it will eventually, but when is anybody's guess.

Okay, so say they crack Starship and actually start building these AI-1 satellites.

What does that future actually look like?

As we've seen, getting them up there is years of near constant launching all on its own.

You never really stop, because radiation is always killing the oldest ones.

Eleven launches a day, forever, just to stand still. But once they're finally all up, this is what it will look like.

Each satellite ends up with about 600 square kilometres of sky all to itself.

If you picture it, that's one satellite, and then nothing but empty black for about 24 kilometres, the length of Manhattan, and then the next one. That doesn't sound too bad, right?

A scattering of these lonely little machines, each in its own Manhattan-sized box of emptiness.

Maybe it won't be a catastrophe after all.

But nothing up there sits still.

Every one of these huge spacecraft is moving at 27,000km an hour, and they're not all going the same way.

At that speed, a single fleck of paint carries the energy of a rifle bullet, a stray bolt hits like a hand grenade, and when two whole satellites cross paths, they meet at about 11 kilometres per second.

Faster than you could blink and they would have passed completely cleanly through each other and both would be gone.

Right now, Starlinks avoid this by flying themselves.

Each one gets a constant stream of data on everything around it, and if something is coming, it fires its thrusters and slides out of the way.

No human involved at all.

And the scale of this today is already hard to believe.

In the first six months of 2025 alone, Starlink had to swerve more than 140,000 times.

That's about 800 dodges every single day.

Across all of the constellations up there, two satellites now pass close enough to trip an alarm roughly every 22 seconds. And that's just the traffic today, with only 15,000 satellites.

Not a million. Not the 1.7 million that have actually been filed for.

war. This is one of the greatest dangers of pushing into orbit this hard. It's called Kessler Syndrome.

And the reason it scares me is because it's just so simple. One collision doesn't just destroy two satellites, it smashes them into thousands of fragments and every fragment becomes its own bullet, on its own unpredictable orbit, hunting for the next satellite, which then shatters into a thousand more. It's a runaway cascade effect until the entire band of orbit becomes a shell of lethal shrapnel that no rocket can fly through for decades, and we become stuck on Earth. Right now, the satellites can dodge, and that's the only thing that's saving us.

But what if they couldn't?

A powerful enough solar storm could scramble the electronics across the entire fleet all at once.

They'd lose the ability to steer.

A million machines suddenly just drifting. If every satellite lost the ability to navigate tomorrow, tomorrow, how long until a catastrophic collision would occur?

Back in 2018, before the mega constellations, the answer was about 120 days.

That's just 4 months of breathing room.

By 2025, that clock had fallen to under 3 days. In 7 years, we went from months of safety to 72 hours.

There's one thing that's on our side though, and it's height.

Down low, around 550km, there's still the faintest trace of atmosphere.

atmosphere, just enough drag to pull a dead satellite back down and burn it up within about 5 years.

Down there, the sky kind of cleans itself.

But SpaceX isn't keeping them all that low.

The constellations climb all the way up to 2000km, and up there there's basically no drag at all, a wreck at that altitude doesn't come down in 5 years, it stays for centuries, some of it for thousands of years. And remember, this isn't just SpaceX, companies and countries from all over the world will be building their own constellations, all forced to share the same orbit, all having to cooperate to avoid this terrifying future.

This is what happens if all of the plans go ahead.

It already feels dystopian and we haven't even reached the scariest part yet.

Now I want to show you the saddest part of all of this. what happens to the night sky down here on Earth?

An astronomer named Samantha Lawler ran a simulation of the sky as each of these constellations comes online.

At 65,000 satellites, which is a tiny fraction of what is proposed, one in every 15 points of light that you can see at night is no longer a star. It's a machine.

So what happens when you scale it to a million?

There are more satellites in the sky than stars for most of the night, night, most of the year, everywhere on Earth.

Under perfectly dark skies, your eyes can take in around four and a half thousand stars at once.

A million satellites would mean more than four and a half thousand man-made points of light sliding across the sky every time you looked up.

And it somehow still gets worse.

Right now, if you wanted to escape the light polluted skies, you would just leave the city and drive somewhere dark.

But in this future, nature, the sky itself glows.

There's nowhere left to drive to.

Just last month, the European Southern Observatory modelled how much these satellites will increase sky glow.

With 60,000 satellites, all faint enough to meet the IAU's brightness guidelines, the sky brightens by about a 10,000th of a percent.

You wouldn't even notice it.

The real problem is the mirrors.

A company called Reflect Orbital wants to put giant reflectors in orbit that can beam sunlight to any spot on Earth.

Literally sunlight in demand whenever you want it.

5,000 of these satellites by 2030 would brighten the night sky by 20 to 30 percent.

Scale that up to the 50,000 they want by 2035, it's 200 to 300 percent brighter.

That's not even a dark sky anymore, that's the light pollution of a suburb everywhere on Earth, every single night. ESO's recommendation is very simple. Keep the total number of satellites under 100,000 and keep every one of them fainter than a magnitude 7.

We're nowhere near that.

There are already proposals for 1.7 million new satellites. The old reassurance was that satellites are only visible because they catch sunlight and reflect it. So once you've turned into the deep dark of the night, sitting in Earth's own shadow, they should go dark too.

And that's mostly Seriously true today, but only because today's satellites orbit down low.

These AI data centres sitting up at 2000km high will stay bathed in sunlight almost around the clock, which is the whole point. They want to maximise their time in the sun to keep the computers running.

The AI-1 has 600 square metres of solar panels, that's 5 or 6 times more reflective area than the Starlinks that are flying over your head right now. And there's no mention anywhere in the filing of any sort of dimming, coating, or mirror film to bring the brightness down.

So depending on how high it's flying, each one would sit somewhere between one of the brightest stars in the sky and a dot brighter than Jupiter.

And do I need to remind you, SpaceX wants a million of them. There might never be a moment again from anywhere on Earth with a clear sky, without a satellite trailing across your view.

Well that's pretty depressing.

pressing. So who actually gets to say yes to this?

Who gets to decide what the night sky will look like for everyone else? You'd think it would be some huge debate at the UN, some treaty, or some global vote between all the countries on Earth, but it isn't.

It takes just one American agency, the FCC. And the FCC was set up in 1934 to stop American radio stations from broadcasting over each other.

That is literally it.

If you want to run a satellite, you need to be able to talk to it and for it to talk back, which means using the radio spectrum. So when SpaceX files for a million satellites, the FCC checks essentially just one thing.

Does the radio signal interfere with anyone else's?

They don't check how bright they'll be, they don't check space junk, they don't check the ozone damage from de-orbiting them, just radio. Samantha Lawler, who's that same astronomer who ran the simulations from earlier, put it better than I can.

It's wild that one little company in California, with permission from one agency that looks after radio transmissions, can change the sky for everyone in the entire world. The year is 2050.

You take your child outside to show them the Milky Way, and to tell them how you used to love stargazing, way back in 2026.

You look up, and the sky is alive.

Not a few satellites drifting past like today.

Thousands of them, moving so fast in so many overlapping directions, it's hard to even track, the sky is dizzying. Your child doesn't even really believe you when you say that you used to see the bright arm of the Milky Way stretched out overhead, they think the old pictures are AI generated.

For the first time in human history, there is nowhere on earth that you could stand and see a truly natural night sky. But all of those satellites overhead aren't decorations, they run the GPS in your car, car, the forecast on your phone, the internet the world lives on, the systems your bank and your hospital depend on, and it's all balanced on a knife's edge.

If one major collision happens, we lose it all, and we're stuck down here on Earth, completely helpless.

At this scale, a satellite falls back to Earth every few minutes, tearing at the ozone layer a little bit more each year.

And when your child asks you, why did they do all of this, you'd have to answer honestly, to train AI.

And everything I've just shown you is the tame version, the one where things go roughly to plan.

And we haven't even touched on the biggest question of all, whether reshaping the one sky that every human on Earth shares just to train AI is something we should be doing at all.

But this isn't guaranteed.

Not yet. We have a choice.

The technology to make these satellites nearly invisible already exists. We've been building it for years now. The difference between a sky that we can keep and a sky that we lose isn't physics.

It's whether someone decides the space above our head is something that's worth protecting before it's too late.

The stars aren't disappearing because they're going anywhere.

They're disappearing because we're building a wall between us and them. This is a future that we are choosing, which means right now it's still one that we can choose not to.

I've put resources down below if you want to learn how you can help to protect this for the generations coming after us. See you in the next one.