Elon Musk explains how SpaceX could build AI data centers in space

[00:00:00] Speaker 1: People probably struggle to visualize a little bit when you say like data center in space. Like we're not going to slap engines on a building and fly it up there. Like these actually look like pretty different. And so kind of walk through how you take something that's in a giant building on the ground and turn it into something that's functional in space. [00:00:18] Speaker 2: Yeah, I think it's pretty interesting. A lot of people don't actually know what the inside of a data center even looks like, right? Yeah. And it's some like mythical place where the internet's in the cloud or something. Yeah, some people envision wires, some people envision boxes. But like effectively it comes down to a set number of chips and the things that we need to launch into space are actually quite small when we look at it. The more challenging part is figuring out how do you get the power for it. And that's where a lot of what we've worked on for existing like Starwing technology, the solar arrays are what we want to utilize that expertise to be able to build a satellite that can actually launch the critical components of the data center into space itself. We like to look at this and say like what is the actual engineering problem here and it's really a combination of delivering power delivering power and then taking the waste heat and energy away and sending it into the vacuum of space as you mentioned. [00:01:19] Speaker 3: The AI satellite is actually much simpler than a Starlink satellite. A Starlink satellite has gigantic phased ray antennas. It's got parabolic antennas. It's got a lot of laser links. It's much more complicated than an AI satellite. An AI satellite is essentially a lot of solar cells, a radiator, and you still need some laser links, but you don't have all of the super complex antennas that you have on a Starlink satellite. So, I mean, given the two, the easier one to design for is the AI satellite. [00:02:05] Speaker 1: Yeah. [00:02:06] Speaker 3: It's just a little bit bigger. [00:02:07] Speaker 1: It's bigger. Just makes stuff bigger. Yeah. I was like, so we've got, this is our AI one if you guys want to walk us through. [00:02:16] Speaker 2: Yeah, so the first thing that we're really looking at here is like, first you've got to make something compelling, right? And we thought that the right place to start is around the 150 kilowatt, like, peak power level. But as we look at the workloads with our experience with XAI, we get to actually see that we can also support about 120 kilowatts of average compute. There's a difference. [00:02:39] Speaker 3: What we're showing here is kind of a draft version of the, version one of the SpaceX AI satellite, an AI one, I guess you could call it. And it seems like a reasonable place to start is 150 kilowatts peak power, 120 kilowatts sustained power. And to give you a sense of what does that actually look like in terms of the size of the radiators, size of the solar panels, the assumptions here are 250 watts per square meter for the solar array and about 1400 watts per square meter for the radiators. So the radiators, these are double-sided, radiators are radiating both sides. They're oriented knife edge to the sun and it's 1400 watts per square meter is a very achievable goal. Over time, we think we can probably do above 250 watts per square meter and above 1400 watts per square meter for the solar panels and radiators respectively. But this gives you like a, this is pretty much what the satellite is going to look like. It's a lot of solar panels, radiator, and then everything else is pretty small, like embarrassing. [00:03:50] Speaker 2: And these are like evolutions of things that we have actually already launched in our Starlink constellation to date. That's really, I think, the cool part to me is that we're looking at solar technology that we already are going to use on the V3 Starlink vehicle. So I'm like really excited to then just take those and make it bigger. [00:04:13] Speaker 3: Yeah, part of what we want to convey here is that there's not some magic that's necessary that doesn't exist for the AI satellites. As Ian said, this is, a lot of this is technology we've already made for the Starlink V3 satellites. So it's, it's, we basically, we don't think this is a super hard problem compared to things we already do. There would also be probably something on the order of a terabit of connectivity, of laser link connectivity from the, from the satellite. The 150 kilowatt peak power level is roughly matches what say an NVIDIA GV300 rack would do. So if you've got a GV300 with 72 GPUs, it's peak power, I think it's around 140 kilowatts. But it's rarely, it's almost impossible to get it to, to be at that peak power. A more reasonable operating envelope would be around 120 kilowatts average power. But, but it can peak up to 150. So that's, it's basically, think of it as a, a rack of compute in space. And then you can connect the, these, these racks of compute to, uh, either each other by the laser links, um, or directly to the Starlink constellations. So you can close the link, uh, with the Starlink constellation, and then Starlink can then, um, uh, send that data to the ground, uh, using the existing KA and KU, uh, antennas on the, on the vehicle. Um, it also has laser, laser links to the ground as well. So, uh, and this, this would not be at a particularly high latency. You know, we're, we're talking about, you know, maybe being around six to 800 kilometers, uh, above the earth, uh, and light travels 300 kilometers per millisecond. So that's, uh, it's about, you know, three milliseconds away. It's not, not very far. [00:06:14] Speaker 1: Won't worry about that too much. [00:06:16] Speaker 3: It's not, sometimes people want to think there's going to be some, like, high latency. I'm like, yeah, it's no speed of light moves pretty fast. Light moves pretty fast. It's a tall one. [00:06:25] Speaker 2: Yeah. Yeah. I think the cool thing also is the, uh, the radiators themselves are about the same size as the existing, uh, solar rays for the V3 vehicle. Um, kind of, kind of in that, that realm where we're flying today. [00:06:38] Speaker 1: Yeah. So, I mean, they got, they got about a 70 meter wingspan. So these are fairly large and we're talking about building a lot of them and putting them up there. But you like to say like space is in the name, like there's, there's a lot of space up there. And so even when you're talking thousands or even, you know, up to a million satellites. Yeah. You got plenty of room to move around up there. [00:07:01] Speaker 3: Yeah. Space is really big. So it's not like, it's not like space is going to get crowded. Uh, space is, is enormous. Like if you zoom in close to the satellite, it looks big. But if you actually look at it relative, relative to the earth, the satellites are so tiny. You can, you can't even see them. So. They're, they're very, very tiny compared to earth. [00:07:24] Speaker 1: And I mean, we have 10, about 10,000 starlings in orbit right now. Yeah. We've got a pretty good idea of how to operate just really large constellations and do it safely now. Right. [00:07:36] Speaker 2: We are the only operator that has any experience of that scale. Uh, it's, it's a great thing that we, you know, we have this background, so we know how tightly we can pack the satellites and, and, and fly them safely. [00:07:47] Speaker 1: That's, that's a, that's a number one goal when, when we look at the constellation, we're going to be building a lot of satellites and we're going to be building them here in Bastrop. Right. So we've, we've got this, which, so we're, we're in that building kind of in the middle, which we're sitting in that building right now. This is my first time here. The building is massive. Like you, you come around the corner, you see it through the trees and you're like, oh, wow. But we're about to kind of put this building to shame, aren't we? [00:08:15] Speaker 3: Uh, yes. We're going to, in fact, we already have the solar manufacturing facility. It's under construction already. And, uh, and then we will be building out the AI SAP production building soon. Um, and, uh, yeah, so we expect to have the, the, the, the AI SAP production, the solar production, um, and, uh, all of that operating at, uh, some reasonable volume by the end of next year. [00:08:45] Speaker 1: So if anybody wants to work on AI satellites, this is kind of going to become the hub of that. We're also, so I mean, like right behind us, the machines are humming, we're still making all of our user terminals for Starlink here. That's not going anywhere. [00:08:59] Speaker ?: Yeah. [00:08:59] Speaker 1: That's not going anywhere. In fact, we're turning on new production lines for new units, right? [00:09:03] Speaker 3: Uh, yes. Um, in fact, these are the new Starlink terminals, uh, which we made in much higher volume than, than the current, uh, terminals. Um, you know, ultimately we think there's probably going to be a few hundred million Starlink terminals out there. And then our, the Starlink direct to cell constellation will, um, connect directly to people's cell phones and enable a high bandwidth communication directly from your phone to space. [00:09:30] Speaker 1: Yes. All right. We're, we're two limiting factors down. We've got mass to orbit. Mm-hmm. We've got putting solar and the few third ones chips. [00:09:38] Speaker 3: Yes. Um, so at least in the, in the beginning, we can obviously launch the, the chips that are already being made. Um, so our current reference design is for NVIDIA, uh, Rubin chips, or it could be either GB300 or, or Rubin chips. Um, um, and, uh, we'll also have a reference design for TPUs and, and essentially you can put up, put any, any existing chips into, into orbit. Um, but the, the current industry, uh, seems to be, uh, it seems like it's going to, I don't know, get to maybe around a hundred gigawatts a year of, of AI compute, but it, that, that doesn't answer the question of, well, how do you get to a terawatt? That's why you need, uh, the tariff app. Oh, he's looking a step bigger. Yeah. Yeah. Yeah. In order to get to the next order of magnitude, uh, you need, uh, a gigantic ship factory. Uh, and to give you a sense of scale here, uh, we expect that the tariff app is going to be around a hundred million square feet. Uh, which is 10 times the size of the, uh, a Tesla gig factory, Texas. [00:10:55] Speaker 1: And what aside from just, you know, I'm going to need starship point to point to get from one end to the other, aside from just the size, what's going to make this unique, different from any other chip building operation on the planet? [00:11:10] Speaker 3: Well, I think over time there's going to be a lot of technology evolution with the tariff app, but fundamentally it's about scale. So even if there were no, uh, fundamental technology breakthroughs, uh, and, uh, you simply, you could simply scale, uh, the existing chip making technology, uh, with a lot of difficulty, uh, to a terawatt of chip output per year. Um, that's if you look at it, just from the logic die standpoint, that's, uh, that's a public, that's like having a billion chips per year with a kilowatt per radical. So it's a billion full radical equivalent chips, uh, each during a kilowatt. And then you're going to need a lot of memory to go with that. [00:11:55] Speaker 1: A lot of people today even think orbital data centers were like a decade away. [00:12:00] Speaker 3: Yeah, I think we want to try to give people a sense of, of the timeframe. Uh, we, at least the timeframe we're aiming for. I mean, you know, people should take this with a grain of salt to some degree because this is, this is just our best guess. So this is not a, this is not a promise of what we'll do. This is what we, what we are going to try to do and think we probably can do, um, which is to get to roughly an annualized rate of a gigawatt per year by the end of next year. Uh, in terms of space, uh, AI compute, um, and then aspirationally scale that by an order of magnitude per year. So in two and a half years, hitting an annualized rate of 10 gigawatts a year to space and three and a half years, maybe a hundred gigawatts. And then depending upon what progress, uh, there is in chip making in the rest of the world and with the tariff fab, uh, going beyond that to scale to a, a terawatt per year, which is a thousand gigawatts. Right. Which is, that, that's twice the election, the current electricity consumption of the United States. Yeah. I think there will be an appetite for that, but we'll see. It's a lot of satellites. I don't know what I was going to think about, but, uh, maybe do a lot of simulations or something. Yeah. [00:13:15] Speaker 1: So after we've, you know, working through all the limiting factors, we've kind of topped out what we can do on earth. What is the next step to, again, try and actually notch maybe some percentage points towards becoming Kardashev level two? [00:13:34] Speaker 3: Why stop there? Yeah. Why think small? Because a terawatt actually is very small. I don't want you to think small. Let's not think small. Um, so there is, in order to get to another three orders of magnitude to a thousand X from a terawatt per year. Um, the, the only way that we can really say, see that you can achieve that is on the moon with, uh, a mass driver. Essentially where you do local production of, uh, photovoltaics and solar and radiators on the moon. Um, maybe you bring the chips from earth or you could conceivably, uh, make the chips on, on the moon. Um, and, but you need, you need most of the mass, uh, to be made on the moon. So you don't have to transport it to the moon from earth. And, and then because the moon has no atmosphere and only one sixth earth's gravity, you can act, you can get, you can accelerate the AI satellites into deep space without a rocket. So you can basically shoot them into space using, um, an electromagnetic gun, like a, like a rail gun type. I mean, just, it's basically a linear electric motor as a way to think about it. That's a way to think about it.