What Would SpaceX's Space Datacenter Plans Look Like? [11:31]

YouTube transcript reformatted at textformatter.ai follows.

[SIDEBAR] Last year I made a video on Lumen orbit's proposal to build data centers in space, now all the big players in spaceflight and AI are seriously looking at this. And while there's economic arguments that may be realized at some point in the future, there are other forces that may drive this.
Why Everyone Is Talking About Data Centers In Space | 28:10
Scott Manley | 1.83M subscribers | 339,743 views | December 21, 2025

What Would SpaceX's Space Datacenter Plans Look Like? - Transcript
Hello, it's Scott Manley here. This is a model of what we think SpaceX's data center mega constellation may in fact look like. And I say mega constellation when we've been calling Starlink a mega constellation, but this actual FCC application or whatever suggests that they may in fact have a million satellites as opposed to Starlink, which is merely thousands or tens of thousands.

So this is all rendered in Universe Sandbox, which is a game that you can buy on Steam. It's basically a physics simulator that lets you set up planets, lets you occasionally drop stuff into them, and lets you maybe cut them in half with a laser. Here's me just dropping a moon into the satellite constellation to see what kind of havoc it wreaks as it orbits through the sun-synchronous halo.

These objects are relatively small, 500 kg, roughly consistent with the Starlink V2 minis, but they still leave a flash as they slam into the surface at several kilometers/s. But of course, I'm doing this because about a week ago, SpaceX announced that it had acquired some company called X AI. XAI also owns something that used to be known as Twitter, which is now known as X. There was a press release which appears to have been authored by Elon Musk, at least certainly in his voice, talking about how they will use this to launch large data centers into orbit. They're going to build out humanity's compute in space. This is the way that he sees the world shall move forwards towards a cardv scale type simulation. That is just basically how much of the sun we can harness.

And no doubt there are many SpaceX fans out there who want to live in this future where we have access to huge amounts of compute powered by near limitless solar energy. On the other hand, there are no doubt a whole bunch of skeptics who have pointed out that XAI has invested something like 8 billion last year and brought in $100 million in revenue, which compares rather poorly to the billions of dollars brought in by OpenAI. I am not going to spend my time figuring out whether this is another solar city-style acquisition. I just wanted to visualize the constellations.

So this was the filing that was made early in January. And very quickly, in the orbital parameter section, you'll see that it talks about requesting the ability to distribute these satellites in near circular shells at altitudes between 500 and 2,000 km and at 30° and sun-synchronous inclinations. Now the sun-synchronous part is common to a lot of the data center orbital parameters because if you put it in a sun-synchronous orbit over the poles, then if you get them in the right orientation, then these things can always be in sunlight.

So that is the group that is going over the poles here, right? But they also talk about a 30° shell, and that's what is sort of sitting as a band around the planet there. That's three shells at different altitudes. These are arranged in what are called walker shells. And that's from the guy that came up with the idea. It's essentially you have a way of distributing the satellites to cover as much of the area as possible. The ones over the poles, those are SSO bands. I like to call them those sun-synchronous halos because it kind of looks a bit like a halo.

So it sounds like there's basically two different tiers of compute. You have the continuously operating stuff that is in sun-synchronous orbit, and that will obviously talk to Starlink, which will ultimately talk to the end users that will get 100% non-stop sunlight. But then you have this 30° band which we know will go behind the planet and go into night, but while they are on the daytime side of the planet, guess what? Most people are working during daylight hours; there is greater demand for all sorts of services during daylight, and I guess the idea is that this band will be delivering those extra daylight services close to where the daylight actually is.

So now this is Universe Sandbox, and we have a model of the Earth. We can put our camera on the surface of the Earth and we could see what a hypothetical sunset looks like. So let's just set up some background stars, and you can just see these objects flying over. There's also a sun there. So this would kind of be hard to see because this is daylight. Nighttime is coming, and with nighttime we have that giant sun-synchronous halo going over the poles, and that's going to be visible from pretty much anywhere on the planet just after sunset.

However, I will caution that the visibility of these objects in this particular simulation is in no way representative of the real thing. This is just what I could do to make them actually visible with the software that I had. The actual visibility calculations would require a whole bunch of information that we simply don't have. We'd need to know the geometry of the satellites and exactly how they would maneuver in space. We can probably presume that the solar panels would be facing the sun directly, and that would mean that most of the light, the specular reflection from those panels, would be reflected back at the sun. So there would be certain regions where you'd sometimes get flares, but it would be pretty rare.

Similarly, if they use a flat style body like Starlink and decide to orient that towards the Earth, then there's going to be regions in the sky where you get these little flares that pilots report. There are understandably a lot of people who are concerned about exactly how bright this stuff will be, and they will no doubt be doing the mathematics when we actually get better ideas of what we're looking at. On the other hand, I bet there's a bunch of you who would like to see this with realistic satellite sizes. And so, here we go. I know you can't see any of the satellites because they are very small compared to the size of the Earth, which as we know is pretty big. Although the Earth is still pretty small compared to space, a better way to understand how big the satellites are is to use Google Earth.

Now, if we work on the assumption that they're using the V3 size of satellites that will be launched by Starship, that's those ones on the bottom. How big are they? Well, the estimated span is about 50 m. So if we go into Google Earth and find an object which is about 50 m long, which could conveniently be one of these propellant storage tanks near the launch site of Starship, that gives you an idea of how big that would be on the ground. And so now as we zoom out, this very quickly becomes smaller than a single pixel. But you know, stars are smaller than a single pixel. It really depends on the viewer being in the right place to see the reflected sunlight from the panels. And that is a study that I'm basically not equipped to do right now.

Instead, I'd like to talk about how I coerced Universe Sandbox to display this giant constellation. Now, Universe Sandbox has a lot of tools that let you build out things. So, you can build out planets with rings. You can have entire solar systems smashing together. It can be a lot of fun. However, building something like this required a little more hacking. So what I did is I started out by looking at where the save files were and found they were in this uubbox format. What is that? Well, if you open that up, it turns out that it is a zip file, right? So we can go into test simulation. Please ignore. And you've got like this info simulation blah blah blah. The main part is simulation. If you open that up, it is a JSON file.

So if you unzip this and open up the file, you find essentially a bunch of JSON code in here. For example, in the bodies or entities section, you can find Earth, and if you look a bit further down, you can find another object, test satellite, please ignore. What my plan was was to basically copy this structure and create thousands of these for all the satellites in the scene. So I created some Python code which would do this walker shell construction. It would take a bunch of parameters, it would emit a bunch of parameters for the satellites, and then this would basically copy the entities in place and generate a new JSON file and dump that out. And if you zip that up and drag it back into the game, you can see what you have rendered.

And one of the hard parts was actually being able to see these objects. As I said, you know, they're obviously very small, and it didn't actually want to render these things. So what I had to do was cheat. I had to go in and tweak their parameters and make sure that they were so hot that they were emitting light like miniature stars. Obviously, these are data centers that have solved the cooling problem by emitting it as light. I also set it up so that I could view real data. It would take the existing TLE data for the Starlink constellation that I got from like a Celestra and just turn that into a Universe Sandbox file. I've shared this on the workshop if you want to mess around with it.

I also have another workshop file which has all of the satellites that are in the Celestra list, and these are of course available on the workshop for Universe Sandbox if you want to download them and mess around with them and smash stuff into Earth and see what happens. The Python code I'll probably share on my Patreon. It's kind of janky and hard to maintain, but for those of you who really want it, I will definitely share it.

So anyway, this is obviously incredibly ambitious and has been stated as something like 3 years away in Elon time. I doubt that will be ready by this point, and it will obviously be contingent on AI not being some bubble, somebody not inventing a faster and cheaper, lighter weight way of running these models. There are so many things that could render this pie in the sky, so to speak. But a more important thing was on Sunday afternoon when Elon Musk essentially said that SpaceX has already shifted focus to building a self-growing city on the moon. Mars is now very much on the back burner, and honestly, this is what a lot of people have been saying because when you can only launch payloads every 26 months, you end up twiddling your thumbs.

There's a lot of problems that can be solved and tested and explored from the moon, and building out a giant data cluster or whatever like this helps if you can say build solar panels on the moon and then launch them to space with a mass driver or something. This is of course what Jeff Bezos at Blue Origin has been saying for a long time. And soon after he was tweeting some images of them being the tortoise versus the hare. And of course, just over a week ago, Blue Origin surprised everyone by pausing New Shepard flight, saying that they wanted to focus on the moon.

This whole focusing on the moon thing does feel like they've had some talks in the back rooms with NASA and various other people in the administration saying that we would really like America to be first on the moon. So yeah, if you're the kind of person that has an Occupy Mars or a nuke Mars shirt, you probably now need to have a moon first shirt. I'm Scott Manley. Fly safe.

As someone who has been in quite a few datacenters during my work career. I wonder the same thing, what would a datacenter in space look like.

what happens when something fails in the datacenter, will a manned mission be required to repair something.

The entire concept sound fascinating, I just wonder about the day-to-day activities that happen in datacenters here on earth versus how they would be handled by datacenters in space.

Bkmk

Now expand that idea to the solar system arranged in a tight pattern, and you have Tabby’s Star and 14 other stars around us. Or that’s one possibility for the anomalous well defined shading across Tabby & the 14 star cluster out of 4,000 surveyed nearby stars.

https://arxiv.org/pdf/2111.01208.pdf

The Deep State AI algorithm will rule the world and complete the new world order. Freedom will be banned and the human race forever slaved.

I would assume COOLING (which is a real problem here on Earth), wouldn’t be a problem in space. You could wick that heat off into the void and it wouldn’t care in the slightest.

I guess “radiate” is probably the better term.