- 1 week ago
Moving beyond internet satellites, SpaceX is designing a new class of spacecraft: the AI satellite. Elon Musk explains how these orbital data centers are surprisingly simpler than Starlink in some ways, yet require massive solar arrays and radiators to function. See the first draft designs and understand the technology making it possible.
SpaceX and Starlink technology could enable orbital AI computing through solar power, laser connectivity, and satellite-scale infrastructure. SpaceX's Elon Musk, Ian Dahl and Dan Huot discuss.
Credit: SpaceX
SpaceX and Starlink technology could enable orbital AI computing through solar power, laser connectivity, and satellite-scale infrastructure. SpaceX's Elon Musk, Ian Dahl and Dan Huot discuss.
Credit: SpaceX
Category
🤖
TechTranscript
00:00People probably struggle to visualize a little bit when you say like data center in space.
00:04Like we're not going to slap engines on a building and fly it up there.
00:08Like these actually look like pretty different.
00:10And so kind of walk through how you take something that's in a giant building on the ground
00:15and turn it into something that's functional in space.
00:18Yeah, I think it's pretty interesting.
00:21A lot of people don't actually know what the inside of a data center even looks like, right?
00:24Yeah.
00:25And it's some like mythical place where the internet's in the cloud or something.
00:29Yeah, some people envision wires, some people envision boxes.
00:32But like effectively it comes down to a set number of chips
00:37and the things that we need to launch into space are actually quite small when we look at it.
00:42The more challenging part is figuring out how do you get the power for it.
00:46And that's where a lot of what we've worked on for existing like Starwing technology,
00:52the solar arrays are what we want to utilize that expertise
00:58to be able to build a satellite that can actually launch the critical components
01:01of the data center into space itself.
01:05We like to look at this and say like, what is the actual engineering problem here?
01:10And it's really a combination of delivering power
01:14and then taking the waste heat and energy away
01:17and sending it into the vacuum of space as you mentioned.
01:21The AI satellite is actually much simpler than a Starlink satellite.
01:27A Starlink satellite has gigantic phased ray antennas.
01:32It's got parabolic antennas.
01:36It's got a lot of laser links.
01:42It's much more complicated than an AI satellite.
01:45An AI satellite is essentially a lot of solar cells, a radiator,
01:51and you still need some laser links,
01:53but you don't have all of the super complex antennas that you have on a Starlink satellite.
01:59So, I mean, given the two, the easier one to design for is the AI satellite.
02:05It's just a little bit bigger.
02:07It's bigger.
02:08Just makes stuff bigger, yeah.
02:10I was like, so we've got, this is our AI one,
02:14if you guys want to walk us through.
02:16Yeah, so the first thing that we're really looking at here is like,
02:20first you've got to make something compelling, right?
02:22And we thought that the right place to start is around the 150 kilowatt peak power level.
02:29But as we look at the workloads with our experience with XAI,
02:33we get to actually see that we can also support about 120 kilowatts of average compute.
02:39There's a difference.
02:40What we're showing here is kind of a draft version of the,
02:45version one of the SpaceX AI satellite, the AI one, I guess you could call it.
02:50And it seems like a reasonable place to start is 150 kilowatts peak power,
02:56120 kilowatts sustained power.
02:58And to give you a sense of what does that actually look like
03:01in terms of the size of the radiators, size of the solar panels,
03:05the assumptions here are 250 watts per square meter for the solar array
03:10and about 1,400 watts per square meter for the radiators.
03:15So the radiators, these are double-sided, radiators are radiating both sides.
03:19They're oriented knife edge to the sun.
03:24And it's 1,400 watts per square meter is a very achievable goal.
03:29Over time, we think we can probably do above 250 watts per square meter
03:33and above 1,400 watts per square meter for the solar panels and radiators, respectively.
03:40But this gives you like a, this is pretty much what the satellite is going to look like.
03:45It's a lot of solar panels, radiator, and then everything else is pretty small by comparison.
03:50And these are like evolutions of things that we have actually already launched
03:54in our Starlink constellation to date.
03:57Yeah.
03:58That's really, I think, the cool part to me is that we're looking at solar technology
04:03that we already are going to use on the V3 Starlink vehicle.
04:07So I'm like really excited to then just take those and make it bigger.
04:13Yeah.
04:14Part of what we want to convey here is that there's not some magic that's necessary
04:20that doesn't exist for the AI satellites.
04:23As Ian said, this is, a lot of this is technology we've already made for the Starlink V3 satellites.
04:31So it's, it's, we basically, we don't think this is a super hard problem compared to things we already do.
04:38There would also be probably something on the order of a terabit of connectivity,
04:42of laser link connectivity from the satellite.
04:46The 150 kilowatt peak power level is, roughly matches what, say, an NVIDIA GV300 rack would do.
04:55So if you've got a GV300 with 72 GPUs, its peak power, I think it's around 140 kilowatts.
05:02But it's rarely, it's almost impossible to get it to be at that peak power.
05:07A more reasonable operating envelope would be around 120 kilowatts average power.
05:13But, but it can peak up to 150.
05:16So that's, it's basically, think of it as a rack of compute in space.
05:21And then you can connect the, these, these racks of compute to either each other by the laser links
05:29or directly to the Starlink constellations.
05:32So you can close the link with the Starlink constellation.
05:35And then Starlink can then send that data to the ground using the existing KA and KU antennas on the,
05:45on the vehicle.
05:47It also has laser to, laser links to the ground as well.
05:50So, and this, this would not be at a particularly high latency.
05:55You know, we're, we're talking about, you know, maybe being around six to 800 kilometers above the earth.
06:04And light travels 300 kilometers per millisecond.
06:08So that's, it's about, you know, three milliseconds away.
06:12It's not, not very far.
06:14Won't worry about that too much.
06:16It's not, sometimes people want to think there's going to be some, like, high latency.
06:19I'm like, yeah.
06:20No, speed of, light moves pretty fast.
06:22Light moves pretty fast.
06:23It's a tall one.
06:25Yeah.
06:26Yeah.
06:26I think the cool thing also is the, the radiators themselves are about the same size as the existing solar
06:32rays for the V3 vehicle.
06:35Kind of, kind of in that, that realm where we're flying today.
06:38Yeah.
06:39So, I mean, they got, they got about a 70 meter wingspan.
06:41So these are fairly large.
06:43We're talking about building a lot of them and putting them up there.
06:48But you like to say, like, space is in the name.
06:51Like, there's, there's a lot of space up there.
06:53And so even when you're talking thousands or even, you know, up to a million satellites.
06:59Yeah.
06:59You got plenty of room to move around up there.
07:01Yeah.
07:01Space is really big.
07:03So it's not like, it's not like space is going to get crowded.
07:07Space is enormous.
07:09Like, if you zoom in close to the satellite, it looks big.
07:11But if you actually look at it relative to the Earth, these satellites are so tiny, you can't even see
07:18them.
07:20So they're very, very tiny compared to Earth.
07:23And I mean, we have 10, about 10,000 starlings in orbit right now.
07:28We've got a pretty good idea of how to operate just really large constellations and do it safely now, right?
07:36We are the only operator that has any experience of that scale.
07:40It's, it's a great thing that, you know, we have this background.
07:43So we know how tightly we can pack the satellites and, and, and fly them safely.
07:47That's, that's a, a number one goal when, when we look at the constellation.
07:51We're going to be building a lot of satellites and we're going to be building them here in Bastrop, right?
07:57So we've, we've got this, which, so we're in that building kind of in the middle, which.
08:02Yeah, we're sitting in that building right now.
08:04This is my first time here.
08:05The building is massive.
08:07Like you, you come around the corner, you see it through the trees and you're like, oh wow.
08:11But we're about to kind of put this building to shame, aren't we?
08:15Uh, yes, we're going to, in fact, we already have the solar manufacturing facility.
08:22It's under construction already.
08:24And, uh, and then we will be building out the AI sat production building soon.
08:30Um, and, uh, yeah, so we expect to have the, the, the, the AI sat, the AI sat production, the
08:37solar production, um, and, uh, all of that operating at, uh, some reasonable volume.
08:44By the end of next year.
08:45So if anybody wants to work on a AI satellites, this is kind of going to become the hub of
08:51that.
08:51We're also, so, I mean, like right behind us, the machines are humming.
08:54We're still making all of our user terminals for Starlink here.
08:58That's not going anywhere.
09:00In fact, we're turning on new production lines for new units, right?
09:04Uh, yes.
09:04Um, in fact, these are the new Starlink terminals, uh, which we've made in much higher volume than, than the
09:11current, uh, terminals.
09:12Um, you know, ultimately we think there's probably going to be a few hundred million Starlink terminals out there.
09:19And then our, the Starlink direct-to-cell constellation will, um, connect directly to people's cell phones and enable, uh,
09:27high bandwidth communication directly from your phone to space.
09:30All right.
09:31We're, we're two limiting factors down.
09:33We've got mass to orbit.
09:34We've got putting solar in a few third ones, chips.
09:38Yes, um, so, at least in the beginning, we can obviously launch the, the chips that are already being made.
09:46Um, so, our current reference design is for NVIDIA, uh, Rubin chips, or, it could be either GB300 or, or
09:56Rubin chips.
09:57Um, and, uh, we'll also have a reference design for TPUs, and, and essentially you can put up, put any,
10:05any existing chips into, into orbit.
10:08Um, but the, the current industry, uh, seems to be, uh, it seems like it's going to, I don't know,
10:18get to maybe around a hundred gigawatts a year of, of AI compute.
10:23But it, that, that doesn't answer the question of, well, how do you get to a terawatt?
10:28Uh, that's why you need, uh, the TeraFab.
10:32Oh, he's looking a step bigger.
10:34Yeah.
10:34Yeah, in order to get to the next order of magnitude, uh, you need, uh, a gigantic ship factory.
10:41Uh, and to give you a sense of scale here, uh, we expect that the TeraFab is going to be
10:46around a hundred million square feet, uh, which is ten times the size of the, uh, Tesla Gigafactory Texas.
10:57And what, aside from just, you know, I'm going to need Starship point to point to get from one end
11:03to the other, aside from just the size, what's going to make this unique, different from any other chip building
11:09operation on the planet?
11:11Well, I think over time, there's going to be a lot of technology evolution with the TeraFab, but fundamentally it's
11:17about scale.
11:17So even if there were no, uh, fundamental technology breakthroughs, uh, and, uh, you simply, you could simply scale, uh,
11:27the existing chip making technology, uh, with a lot of difficulty, uh, to a terawatt of chip output per year.
11:35Um, that's, if you look at it just from the logic die standpoint, that's, uh, that's, uh, that's equivalent, that's
11:42like having a billion chips per year with a kilowatt per radical.
11:47So it's a billion full radical equivalent chips, uh, each doing a kilowatt, and then you're going to need a
11:53lot of memory to go with that.
11:56A lot of people today even think orbital data centers were like a decade away.
12:01Yeah, I think we want to try to give people a sense of, of the timeframe, uh, we, at least
12:08the timeframe we're aiming for.
12:10I mean, you know, people should take this with a grain of salt to some degree because this is, this
12:14is just our best guess.
12:15So this is not a, this is not a promise of what we'll do.
12:18This is what we, what, what we are going to try to do and think we probably can do, um,
12:24which is to get to roughly an annualized rate of a gigawatt per year by the end of next year.
12:30Uh, in terms of space, uh, AI compute, um, and then aspirationally scale that by an order of magnitude per
12:38year.
12:39So in two and a half years, hitting an annualized rate of 10 gigawatts a year to space and three
12:44and a half years, maybe a hundred gigawatts.
12:47And then depending upon what progress, uh, there is in chip making in the rest of the world and with
12:54the tariff fab, uh, going beyond that to scale to a, a terawatt per year, which is a thousand gigawatts,
13:00which is, that, that's twice the electric, the current electricity consumption of the United States.
13:05I think there will be an appetite for that, but we'll see.
13:09It's a lot of satellites.
13:10I don't know what I'm just going to think about, but, uh, maybe do a lot of simulations or something.
13:16So after we've, you know, working through all the limiting factors, we've kind of topped out what we can do
13:23on earth.
13:24What is the next step to, again, try and actually notch maybe some percentage points towards becoming Kardashev level two?
13:34Why stop there?
13:36Why think small because a terawatt actually is very small.
13:40I don't want you to think small.
13:40Let's not think small.
13:42Um, so there is, in order to get to another three orders of magnitude to thousand X from a terawatt
13:50per year, the, the only way that we can really say, see that you can achieve that is on the
13:55moon with a mass driver, essentially where you do local production of, uh, photovoltaics.
14:03And so, uh, and radiators on the moon, um, maybe you bring the chips from earth or you could conceivably,
14:09uh, make the chips on, on the moon.
14:12Um, and, but you need, you need most of the mass, uh, to be made on the moon.
14:17So you don't have to transport it to the moon from earth.
14:19And, and then because the moon has no atmosphere and only one sixth earth's gravity, you can act, you can
14:26get, you can accelerate the AI satellites into deep space without a rocket.
14:31So you can basically shoot them into space using, um, an electromagnetic gun, like a, like a rail gun type.
14:39I mean, just, it's basically a linear electric motor is the way to think about it.
14:43So, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh,
14:43uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh,
14:43uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh,
14:43uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh,
14:43uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh,
14:43uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh
Comments