SpaceX plans to fill orbit with AI data centers — and astronomers say the night sky is the bill

Space 4 min 80 sources

SpaceX confirmed it will launch the first orbital AI computing satellites in 2027, part of a filing for up to a million spacecraft. Astronomers warn the swarm could brighten the night sky to half-moon glare — a cost they never agreed to pay.

Key takeaways

SpaceX confirmed it will launch the first orbital AI data-center satellites in 2027, part of an FCC filing for up to a million spacecraft, with rivals racing to build their own.
Astronomers warn the swarms could brighten the night sky toward half-moon glare and carve permanent bright lanes across orbit, crippling ground-based telescopes.
The people who study the sky — and everyone who looks up at it — have no seat at the table where the launches, filings, and funding are decided.

The biggest space story this week wasn’t a launch or a landing. It was a plan to move the internet’s compute into orbit — and a warning, from astronomers, about who pays for it.

SpaceX confirms orbital AI data centers for 2027

On June 12, as SpaceX shares began trading on the Nasdaq, company president Gwynne Shotwell confirmed the first orbital AI data-center satellites will launch “late next year” — 2027 [23]. Before that, SpaceX will put computing payloads on some existing Starlink broadband and mobile satellites as “canary sats” to test the technology [23].

These aren’t ordinary satellites. SpaceX calls the full version AI1. Each one is 70 metres long and 20 metres tall with its solar arrays and radiators deployed, generating up to 150 kilowatts of peak power — roughly the draw of one Nvidia GB300 AI server rack on the ground [23][53]. The idea is to run AI computing in orbit, powered by sunlight and cooled by radiating waste heat into space, to dodge the power and cooling limits that ground-based data centers increasingly hit [53].

SpaceX is downplaying the difficulty. “There is not some magic that is necessary,” Elon Musk said in a company video, arguing the satellites reuse Starlink V3 technology — solar arrays, thermal management — rather than a fresh engineering leap [53]. In January, SpaceX filed with the Federal Communications Commission, the US airwaves regulator, for up to one million such satellites [23].

It’s a race, not a one-company plan

SpaceX has the largest ambitions, but it isn’t alone. A Los Angeles startup called Orbital — founded this year by an electric-scooter entrepreneur — raised $5 million on June 9 toward an in-orbit computing demo, with plans to eventually deploy more than 100,000 data-center satellites [17]. Each would carry 100 kilowatts of compute; for scale, the most powerful commercial satellites flown so far manage only 20–30 kilowatts, and the only spacecraft ever in the 100-kilowatt class is the International Space Station, assembled over decades [17]. A rival, Starcloud, has raised about $200 million for a proposed constellation of its own [17].

The pull is simple: AI demand is outrunning the electricity and cooling that terrestrial data centers can get. Orbit offers unlimited sunlight and a cold vacuum to dump heat into. The capital is lining up behind that bet — one industry op-ed this week estimated up to a trillion dollars is about to chase space [49].

The night sky is the unpriced cost

Here is the part that doesn’t appear on any balance sheet. Astronomers say these swarms will wreck their ability to see [23].

This isn’t new — they’ve fought it since the first Starlink launches in 2019, when bright satellites began streaking through telescope images and radio transmissions began bleeding into radio astronomy [23]. They worked with SpaceX to dim the satellites, with partial success. But Tony Tyson, chief scientist at the Vera C. Rubin Observatory — a new telescope built to scan the whole sky for changing events — said at a June 4 National Academies meeting that the problem persists [23]. Current Starlink satellites shine at fifth magnitude, brighter than the seventh-magnitude limit astronomers recommend; the larger V3 satellites will be brighter still [23]. (Magnitude is the brightness scale astronomers use — lower numbers mean brighter, and each step is about 2.5 times the light.)

Orbital data centers make it worse. The satellites are “extremely bright” in the low parking orbits where they’re checked out after launch, Tyson said [23]. Constant launches to build and replenish the fleet will carve permanent “bright lanes” across low orbit that “ground-based folks cannot avoid” [23]. And the scale he warned of is hard to picture: “At a million satellites, the sky brightness would be similar to the glare of a half-moon, which would preclude most of the science programs that our users want to do” [23]. Even in their high working orbits, the satellites flash glints as bright as Venus — enough to ruin the search for supernovas and other sudden events [23].

The astronomers helped, asked, and lost ground anyway. The deciding table — the FCC filing, the IPO pitch, the factory in Bastrop, Texas — never had a seat for the people who study the sky, or for anyone who simply looks up at it [23].

Elsewhere this week

NASA named the crew for Artemis III, its planned return of astronauts to the Moon’s surface, now targeted for 2027 — though this first crewed test will rendezvous with the SpaceX and Blue Origin landers in Earth orbit, not land [7][35]. China’s Long March 5 lofted a classified satellite, and Japan’s H3 rocket returned to flight after an earlier failure [47][66]. And the James Webb Space Telescope, NASA’s big infrared observatory, spotted a “galaxy-killing” wind — gas being blown out of an early galaxy fast enough to starve its star formation, a possible clue to why some galaxies died young [9].

02 · Lesson · why it matters

The bill always lands on whoever wasn't at the table

The people who decide a thing and the people who pay for it are often not the same — and the gap between them is where most damage hides.

A deal between two parties, and a third one who pays

This week SpaceX confirmed it will start launching AI computing satellites in 2027 — the first of a fleet it has asked permission to grow to a million spacecraft. The pitch is clean. AI needs power and cooling; the ground is running short of both; orbit has endless sunlight and a cold vacuum to dump heat into. So the company files with the regulator, raises the money, builds the factory.

Two parties are in that room: the company that wants to build, and the regulator that says yes or no. The deal works for both. And then there’s a third party — the one who pays, and who was never asked.

The thing nobody owns is the thing nobody defends

The bill, in this case, is the night sky.

Astronomers say a million bright satellites would brighten the whole sky to something like the glare of a half-moon — enough to blind most ground telescopes. The satellites flash glints as bright as Venus and carve permanent bright lanes across orbit that, in one scientist’s words, “ground-based folks cannot avoid.” The new telescope built to watch the entire sky for sudden events would be watching through a haze that the science it was built for can’t see past.

Here is the trap. The sky belongs to everyone, which means it belongs to no one in particular. There’s no owner to send the invoice to, no one whose private loss forces them to fight. A cost spread thin across all of humanity is a cost that no single person is paid to defend. So it doesn’t get defended. It just quietly gets spent.

They asked, helped, and lost ground anyway

The astronomers aren’t naïve about this. They’ve been at it since 2019, when the first Starlink satellites started streaking through their images. They worked with SpaceX, got the satellites dimmed, won real concessions. And they still lost ground — the satellites are brighter than the agreed limit, the next ones brighter still, and now a whole new fleet arrives that makes all of it worse.

This is the quiet part. Being cooperative, reasonable, and right is not the same as having power. The astronomers had the better argument and the worse seat. The decisions that mattered — the filing, the funding, the factory — happened in rooms where their argument wasn’t a vote. You can be completely correct and still lose, if the table where the choice gets made doesn’t have a chair for you.

The pattern, once you see it

This isn’t really about telescopes. It’s the shape of almost every cost that travels somewhere it shouldn’t.

A factory’s smoke is free to the factory and expensive to the town downwind. A loud bar is cheap for the bar and costly for the neighbours who can’t sleep. A fishing fleet’s last good year empties an ocean that the next fleet inherits empty. In each case two parties strike a deal that suits them both, and a third — the town, the neighbours, the next fleet, the sky — pays a bill it never agreed to, because no one owns the thing being spent.

Economists have a dry word for this: an externality, a cost pushed onto someone outside the deal. The plain version is older and truer. Watch who’s in the room when something gets decided. The ones who aren’t are usually the ones who’ll pay.

You are downwind of more than you think

It’s easy to read this as a story about other people — companies, regulators, scientists. It isn’t.

You are the third party in more deals than you will ever count. The terms of your phone, your loan, your food, the air over your street, the sky over your head — most of it was settled in rooms you weren’t in, by people balancing their own costs against your invisible ones. Some of those deals are fine. Some are quietly spending something of yours. You mostly can’t tell which, and you almost never find out until the bill arrives.

That’s not a reason for outrage; it’s a reason for humility. The next time a deal looks like a clean win for the two people making it, the honest question isn’t is this good? It’s who isn’t here — and what are they paying so this can look free? You won’t always be able to answer. But knowing the chair is empty, and that you might be sitting in one somewhere yourself, is closer to seeing the whole than assuming the only people in the story are the ones doing the talking.

03 · Lab · your turn

The Empty Chair

Rehearse approving a clean two-party deal, then feel the cost that falls on the party who had no seat at the table.