Why does Elon Musk want to put data centres in space? | The Economist

Some people in the industry are saying. that the best place to build data. centers in the very near future is going. to be in orbit. And that is why Elon. Musk has just merged his rocket company. SpaceX with his AI lab XAI because that. gives him and so far only him both. pieces of the puzzle. But he's far from. the only person pursuing this idea. And. as it becomes more expensive to build. data centers on Earth and more difficult. to get enough energy to power them, the. idea is gaining ground. So I suppose the.

starting point here, Alex, and I think. the reason people have been paying more. attention to this in the past few weeks. is the merger that Elon Musk has done of. basically merging XAI into into SpaceX. AI and rockets. Why do they go together? What is he thinking? >> Yeah. Well, I mean, obviously there are. a lot of reasons for Musk to do this. Partially just consolidating his empire. partially uh SpaceX throws off cash. because it has actual subscribers who. pay money for Starlink and XAI eats up. cash because it's got infinite training.

costs. So even before you get to the. flashy top line, I think there's some. financial engineering that makes this. make sense. But the flashy top line is. really flashy, right? Elon Musk has. applied to the FCC to launch a million. satellites to train AI models in space. This is something that I thought was. ridiculous. Well, we thought 40,000. satellites for Starling. I mean, that. sounded It was like originally 8,000 or. something, and now it's going to be Oh, no, it's going to be 40 or 50. So, now. he's going to a million. Um, and I think. we've actually got a a video of what.

this might look like. So, uh, so what's. the basic idea here? What what are we. seeing? This is by Scott Manley, who is. a aerospace, uh, YouTuber. 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. >> That's a weird layout, right? >> So what are we seeing there? >> So basically you've got that kind of. halo around the outside and that's in a. very special orbit called a. sun-synchronous orbit. And this is an. orbit that essentially means you're on. the line between day and night all the. time. And uh if you can put something. into that orbit, it's basically in. sunlight 24/7 or pretty much 24/7. So uh. it's a really really good place to put. satellites if you want them to be. powered by solar power all day long. Um. but then you might also want to have um.

more capacity. So then you can have. these other this other band of. satellites that goes around the middle. Um, and uh, you know, maybe that part of. the constellation lights up for people. when it's the daytime because people. mostly use AI in the day. Uh, so you're. then able to use these satellites to run. AI for them during the day. >> And so basically, you've got AI, you've. got the orbits with cheap power, but. that is quite far away from people and. not necessarily over them when they need. them. And you've got the orbits with. cheap power only sometimes, but very. close to where it uses.

>> Exactly. Exactly. So that that's why. otherwise you just end up with a halo. that gets really really far from the. Earth. And there are various various. problems with that. Anyway, this sounds. crazy enough, but it gets crazier. because um because Elon Musk has said. that, well, this is just the start uh. because you might actually want to go to. the moon. And we've got a clip of him. talking about why. >> we'll be launching with SpaceX orbital. data centers at the 100 to 200 gawatt. per year level. Not cumulative, I mean. per year. And ultimately we see a path.

to maybe launching as much as a terowatt. per year of compute from earth. But what. if you want to go beyond a mere terawatt. per year? In order to do that, you have. to go to the moon. So by having. factories on the moon building AI. satellites and having a mass driver, which is the kind of thing you really. learn about in or read about in science. fiction, but we're going to make it. real. So, a mass driver is a kind of. rail gunner and you basically build. these satellites and then you throw them.

from the moon so that they go into orbit. around the Earth, >> not to not to prejudge what we're going. to talk about, but while the space stuff. might be plausible, that's nonsense. >> Right. Right. Exactly. But but there is. an agenda here, right? Because he wants. to sell this company. Right. This is a. um and it's interesting because he's. talking about the moon having talked. about Mars for 20 years. Right. >> Previously, his explicit statement was. the moon as a distraction. don't care. about. >> now he's pivoted to the moon and surely. that's because you can make a kind of. business case whether you believe it or. not there's a kind of business case. we're going to have factories on the.

moon we're going to make all these. satellites we're going to we're going to. own AI there wasn't really a business. case we're going to Mars it was just a. kind of this would be cool um so this is. kind of about juicing the the share. price before the IPO right. >> and we're I mean we're expecting an IPO. from SpaceX we're expecting an IPO from. OpenAI we're expecting an IPO from. Anthropic like we are going to see the. the AI IPO bonanza over the next few. years, >> right? And and as we know, Sam Olman of. OpenAI and Elon Musk uh don't get on and. both of them would like to have the.

biggest startup in the world. And. currently uh SpaceX after this uh. financial maneuver is worth 1 point. valued at 1.25 trillion I believe. Um. but Sal Mland has weighed in on this too. and we've got a clip of what he thinks. of this idea. >> Honestly think the idea with the current. landscape of putting data centers in. space is ridiculous. It will make sense someday. But if you. just do like the very rough math of. launch costs relative to the cost of.

power we can do on Earth to say nothing. of how you're going to fix a broken GPU. in space and they do break a lot still. unfortunately. Um we are not there yet. There will come a time space is great. for a lot of things. Orbital data. centers are not something that's going. to matter at scale this decade. And so. that's the negative view, but it's not. just Musk, right? This is the thing. that's interesting to me. I am quite. inclined to dismiss Musk's lights of. fancy, but you've been speaking to other. people who actually think this is.

>> Yes, a lot of people have been looking. at this. Um, last year I spoke to uh for. a piece I reported on a startup called. StarCloud, and um they have been talking. about this, I think since 2024. Um, and. they actually put a satellite into space. and I think we've got a clip of that in. November. So, they've actually got a GPU. in space and they've trained. >> cloud one separation confirmed. >> There you go. >> ISI 3 separation confirmed. >> and obviously coming from a SpaceX.

>> satellite. Exactly. Because SpaceX, you. know, dominates the launch industry. But. what StarCloud is um is doing here is. they are essentially trying to get some. real numbers on how plausible this is. How much cooling do you need? uh how. does a GPU behave in space? Do you need. to give it lots of shielding? Um etc., etc. And they have uh various you know. plans to they've talked about a 2x2. kilometer sort of a massive square thing. uh in space that has sort of shipping. containers full of GPUs that you you. park with it. I mean that's just like. you know also super ambitious. Uh but.

it's not just them and you know to to. their credit they've actually flown. something. Google is looking at this. too. So they they did a they've done a. tie up with Planet and Planet has. already quite a lot of um uh I think a. few hundred satellites in space. They. are a earth observation company but they. know how to build and launch satellites. and so uh so Google is going to fly. something with them next year. Jeff. Bezos has expressed interest in this. He's been talking about factories in. space for a long time and this is kind. of the first step to a factory in space. because an AI factory you only have to. send bits up and down. You don't have to. send atoms. So this is going to make.

sense before factories make sense. and. he says, "Yeah, will, but probably not. for another decade." And then Eric. Schmidt, the former CEO of Google, bought a rocket company last year and. everyone was like, "Why has he done. that?" called Relativity Space. And he's. now said it's it's essentially because. he wants launch capacity because he. thinks orbital data centers is going to. be a thing. And also um Sam Alman. actually looked at a um an acquisition. of a space startup called Stoke Space. last year and decided against it. So. he's clearly crunched the numbers and.

said, "Don't want to do this now." that. would look like a distraction. Uh so for. him it's a no, I don't want to do it. It's ridiculous, but maybe later. But. for Musk, it's like, okay, now um if I. put these pieces together, I've got the. most valuable startup in the world.