Disruptors and curious minds, welcome to another episode of Thinking on Paper, where we take you, the curious individual, behind the scenes of the latest emerging tech and we unpack it together. Why do we do it? Well, we all want to understand why this technology is going to be important, how the world's going to change, what it's going to mean to our families, our jobs, our own curiosity, and that's what we're here doing.
Listen, all of these episodes won't be extremely easy to listen to, but I tell you what, they're going to be informative, they're going to be fun, and they're going to get you using this thing in your head, this jelly that's sitting in your skull. You're going to be thinking a little bit differently, hopefully, about this. So Mark, I'm stoked. I'm stoked today because we're talking about something that I spent a lot of time in, traditionally, not the future look of it, but the past version of it. So I'm excited.
What did we learn about, Mark? Drum roll, data centers in space, five gigawatt data clusters with solar arrays four kilometers across, 400 miles up in the sun synchronous orbit. Now listeners might be wondering why? Why not just build them here on Earth? And before we get into the how, I want to explain the why with a few quotes from the Starcloud white paper. So that's a clue to our guest today, Starcloud. Couple of quotes to explain why. Okay, the first is a quote from our good friend of the show, Elon Musk. We have silicon shortage today, a voltage step down transformer shortage probably in about a year, and then just electricity shortage in general in about two years. And then there's another quote from Tom Mueller, employee number one at SpaceX. The amount of power to run compute by 2045 will be the base power of the planet right now. Not sure what that means. The drain on resources is so high, you need to put that compute in space and use the power of the sun. That's a really good use of space to help save the planet. And that's why data centers in space. Now, Jeremy, I know this is your specialized subject and Philip, it's his specialized subject as well. So I've bought a bell. If you two get too deep in the weeds about data center tech, I'm going to ring this. Just to reel you in, okay?
Love that. I love that. And before we get into this episode, want to remind everybody that we have our hot buttons section that Mark runs. It's going to be really fun. We had our Thinking on Paper news, we have our carryover question, and we also unpack everything backstage at the very end. So stick around. Lastly, before we introduce the guest, Mark, I'm going to tee up a quote from one of my favorite books of all time. I'm not going to nerd out on this too much, but there's a guy named Gerard O'Neill that wrote a book in 1976 called The High Frontier. It was about building colonies and manufacturing in low Earth orbit. Someone who you know, who we referenced last show, Carl Sagan, was quoted in 1976 as he was talking to the US House of Representatives on Gerard O'Neill's work. The engineering aspects of it are perfectly worked out by O'Neill's study group. It is practical. This was in 1976. We all know it's 2025 today. Without further ado, let's bring on our guest and talk about it. Philip Johnston, co-founder of Starcloud. Welcome to Thinking on Paper.
Great to be here. Thanks so much for having me, Jeremy and Mark. I like the reference to O'Neill. Do you guys know what an O'Neill ring is? It's like the, it's the best example of, or the best design for an orbital colony. Anybody that's interested should Wikipedia the O'Neill ring concept.
So basically it spins so you have artificial gravity, but it's designed in a way that would reflect sunlight.
Yes, it looks a bit like that's a bit like what I'm thinking. The thing I'm thinking is more.
Yeah, it's basically a long tube which has the sides opened out to reflect sunlight and generate power. So the whole thing is either permanently in sunlight or permanently in dark, but you have artificial gravity and it's better than just having a ring spinning basically.
So this is a timely topic. AI is getting thrown at us left, right, top, and bottom. And I think some of us who are not in the data center know or even starting to figure out, well, something's got to power all of this AI stuff. Data center demand is up. We know we're limited on power in the US, in the world. There's a finite amount of that. So you come up with this idea along with your team or kind of synthesize this idea together to actually put data centers in space. So let's simplify it. In your definition, what is a data center?
A data center, I mean, actually in really the simplest terms, can be thought of as a low cost energy provider. So you can run data centers on us, but you can run other things too. But in the simplest terms, a data center is a bunch of chips processing or storing data. I don't know if you have a better definition than that, but.
Works for me. Keep it simple. I spent about 10 years as a data center consultant, basically helping people design and put their IT equipment into these big facilities, whether they were leased or we build them or whatever, you know, that sort of thing. So it's more on the mechanical electrical side that supports it. Having this capability in space, let's talk through it. Okay, so let's say I'm a company, right? And I've got an IT compute load that powers my business. And I'm trying to figure out where to put it and why and cost benefits and that sort of thing. When you have a conversation or when you envision, I know you're still building this and you have the test, the pilot is happening in like 2025, but how do you envision talking with somebody to figure out, hey, Jeremy, what's your IT load? And then what's your redundancy requirements? And then how do we configure that, put it on a rocket and let it spin around in low Earth orbit? How does that process look to you?
Yeah, so I'll try and distinguish between the sort of quite near term and the medium to longer term. So the first few iterations of the satellite, you mentioned the first one was going up in a few months in summer this year. These are designed to offer edge compute workloads for mainly other satellites and also to some degree terrestrial customers who care about data sovereignty and would like to run a workload in space. So for those guys, the design of that satellite is quite similar to the design of most satellites in that critical systems have to be redundancy on critical systems and then over provisioning on systems which will degrade over time, particularly solar panels and chips. So yeah, for those conversations, it's basically if we can provide you this amount of compute on orbit, you know, how valuable is that to you and can we provide that at a price point that makes sense. In the longer term, the ambition is that you can think of us a bit like a CoreWeave or Lambda. So, Amar, these guys, essentially, we will be renting out GPU time and whoever the buyer of that GPU time is shouldn't really care whether it's in space or on Earth. We'll be able to say basically, we can provide you H100 time for $150 an hour or $2 an hour, whatever it's going to be. And we do the connectivity, we do the cooling, we do the power, and they don't have to think about that really.
So one of the things I kind of think about too is, we'll talk about how things get up there and where they land and what Lego blocks they end up plugging into. When you think about, the one thing I tend to think about is if you're doing processing there and it's way far away, we know we have latency between here and Chicago, between here and London, between Atlanta and London. How does the latency thing work and how do the things you process in that module come back to Earth and how is that being optimized to make sure it doesn't take too long?
Yeah, so this was basically an unsolved problem until about two years ago when the mega constellations started allowing access directly from other satellite customers. So we're going to plug into one of the mega constellations, be that Starlink, Kepler, one of the others, directly with an optical terminal, so with a space laser. Which, I mean, the data rates on these lasers is extremely high, potentially hundreds of gigabits per second per laser. And you can have multiple lasers if you want. Which is about what a current terrestrial data center... There are definitely terrestrial data centers that have an in and out pipe of about 100 gigabit a second. So it's not too different. In terms of the latency, flying in low Earth orbit is no different really from having a data center terrestrially. If you're 400 kilometers up, that's not that different from having a data center in Alaska or in Seattle or wherever. So yeah, anything that requires sub-20 millisecond latency will need to remain on Earth. But my view is that all other data centers will inevitably move to space once the launch cost hits the same point.
What might need sub-20 to remain on Earth? What kind of use case is that?
Very few use cases to be honest. So nothing that is like an internet-based use case needs that. So any kind of, you know, like this call we're having now could easily be a hundred millisecond latency. And even things which are very latency sensitive, like for example, if you want to, for example, arbitrage two exchanges between London and New York. It's very latency sensitive, but it's actually quicker to do it through space than it is to do it through fiber. It will be very soon if it isn't already much quicker to do it through space. I mean, for one thing, the speed of light is about 50% faster in a vacuum than it is through fiber. So really we're talking about, I mean, very few situations come to mind that require sub-20 milliseconds. I literally know one just because it's a project that I worked on with some buddies. It's a real-time music performance. So if you have, all of us are playing musical instruments from different spots, it requires about 10 milliseconds of latency just to keep us all in time and rhythm.
I didn't get the... Zoom experiments during Covid were so bad?
Yeah. Well, they weren't using the stuff that we were using, but I digress. Let's stay focused on data centers.
All right. So Philip, we get this, so let's, we get this box, right? That we put on, and I'm totally simplifying because the technology that you're building is actually quite brilliant. You put it on the back of a rocket. We send it up to space. There's some kind of station up there, right? That it will connect into, like talk me through the Lego blocks and if we can keep it kind of high level.
Yeah, so the modular design that we have in the white paper and that's in the concept video, I would say that's about a 10-year time frame. And the reason you might want to modularize is if you want to, the largest compute cluster you can launch individually is about a 40-megawatt data center. So that's about a hundred tons worth of compute, solar panels, and radiators. It's what you can fit on one full Starship payload bay, for example. So if you want to build a data center that's internally networked, that's larger than that. And so for example, for training, to be extremely highly internally networked, you can't have just an optical terminal between them, then you would need to have them all hooked up on the same spine basically somewhere in space. That's what that concept is. I think probably in the more immediate term we're going to just be launching these individual 40-megawatt blocks and they'll have their own solar panels and radiators attached. So you're talking about 10 football fields of solar panels and radiators for one of those 40-megawatt blocks versus like a 4-kilometer by 4-kilometer for the 5-gigawatt block. So yeah, I would say that's probably the thing to focus on in the slightly more immediate term is these units of 40 megawatts. And you can use that for all kinds of inference workloads and everything else. It's only really training very large models that you need to cluster these things together.
So 40 megawatts, friends and neighbors, is quite a bit of compute load, just to let everybody know. 40-megawatt data centers terrestrially are massive, like massive, acres upon acres upon acres. When you consider 40 megawatts, is that pure IT load, or does that include kind of the backup power stuff as well?
That includes backup power. I mean, well, actually, one of the nice things about being in a dawn-dusk sun-synchronous orbit is that we're always in the sun. And so we'll need less backup power than... We'll still have some redundancy, but we'll need less than a terrestrial data center would.
Is this the Terminator line?
Yes, the Terminator line. Exactly.
I've been waiting to say that. I've been waiting to say that. So, because the sun-synchronous orbit, I've heard of low Earth orbit, but I've never heard of sun-synchronous orbit. And is that to keep it on the terminator line? So it's always on that line between darkness and daylight. The solar energy is always there.
Yes, exactly. It's a bit, it's not super easy to explain without oranges and apples next to me, but as the Earth progresses around, you're in an orbit that maintains the same, they call it LTAN, like the same position relative to the sun, as the Earth moves around the sun, unless you, unless you're, unless you have this sun-synchronous property which you get by having, because basically the oblateness of the Earth, like it's slightly fatter on the Northern Hemisphere than it is on the bottom hemisphere, means if you're at a certain angle, about seven degrees off, then as the Earth moves around the sun, your whole orbit moves by 1/365th of an orbit every day. And so that means that you stay on the terminator as the terminator goes, as the Earth rotates around the sun. That's what that means.
Shout out for a book I know about the Terminator line, because not because I'm a T-1000, but because I read Orbital by Samantha Harvey recently. And for anyone who's listening who wants to try to imagine what it's like in orbit, check this book out because it's beautiful.
I've not seen that, I might have to buy that myself. A very short one about being on the space station for six months.
Awesome, sounds amazing. I just bought it too, Philip. I haven't gotten into it yet. So, check this out. I know you're probably thinking, all right, putting stuff in space, that seems pretty cool, pretty interesting, but doesn't it cost a pile of money to send stuff up there, right? And where we're at, like launch costs right now is what, about eight million bucks-ish to kind of put some stuff in space just for round numbers?
If you want to book your own launch, about eight million dollars. The beautiful thing very recently and in the last three, four years is these rideshare missions. Our first satellite is a 50-kilo small satellite, it's about the size of a fridge. That cost us 300 grand.
It's amazing, it's honestly amazing.
Yeah, because we're hitching a ride. Those Falcon 9 rideshare missions will launch about 20 missions at once, each of which is paying about $6,000 a kilo. And you can put pretty small stuff on there. If you want to go down even smaller than us, you can pay like 50 grand and put a CubeSat on there if you want.
I love that. And on the other side of that, we talk about the cost to get up in space, but like based on your white paper and some stuff that we've looked at, I mean, significant cost reduction as far as electricity expenses and that sort of thing from the hundreds of millions to the tens of millions. I mean, when you start talking in orders of magnitude like that, things start to seem really, really, really compelling.
Yeah, I would say it is all highly contingent on the launch cost continuing to fall. So it used to be with the shuttle about 15, 20 years ago, it was about $60,000 a kilo. Now it's come down to $6,000 a kilo. That's what we're paying for the first launch. We need the launch cost definitely under $500 a kilo. The marginal launch cost with Starship is projected to be $10 a kilo. So that gives you some sense of where we're heading. Yeah, anything under about $500 a kilo and we're good to go. But if it gets down to $10 a kilo or even if it's $50 a kilo, which is what they say they're going to charge customers, we will be an absurdly profitable company.
Sounds good. Mark, I'm going to try and get this in before you ring the bell on me. But going back to O'Neill's book in High Frontier, he talks about actually setting up manufacturing in space and using these, what do you call them, mass drivers? Like this magnetic rail system from the moon that fires things out. Is that something that is being entertained as we look down the road as an alternative to trying to get escape velocity from Earth?
Yes, I mean, actually people are doing these rail guns on Earth. There's a company called, like, it's not Slingshot, there's a company called something which is doing this rail gun from Earth. Certainly from the moon, it's like, that is almost certainly the way you would do it from the moon in the end state. You can also actually quite feasibly have space elevators hanging off the moon. So yeah, that's one way to do it. And you can have a space elevator on a track that rotates around so that the space elevator, so that it's, I don't know, anyway.
That's a callback to Quantum Supremacy Book Club where we talk about space elevators and material science and the possibilities of what quantum can do. Sounds like Total Recall, a mix of all of our science fiction, science fiction becoming reality, becoming science fiction becoming reality.
I mean, I'm of the view you could build a space elevator on Earth for about $10 billion. The problem with space elevators terrestrially is they're just going to get pummeled with stuff in LEO. So even if you have these avoidance maneuvers, stuff is going to be colliding with them at about 27,000 kilometers an hour all the time. So yeah, it just doesn't really work as well.
Like gnats on a windshield on the highway, right? Like we're talking, yeah. And LEO, low Earth orbit, yeah. If they're strong enough to be there, are they not strong enough to withstand almost all impact?
No, I mean, they're strong enough basically to, they have to have an anchor out beyond GEO, so they have to be about 30,000 kilometers long, so they're strong enough tension-wise like this, but if you shoot a gun at it, it's like shooting a gun at a taut rope, you know, like it can hold this way, but if you shoot a bullet, it's just going to rip the whole thing apart. But just back to your point on the moon and manufacturing in space. Yeah, I'm very convinced manufacturing in space is going to be a thing as well. As I say, the whole thing is contingent on low-cost launch. Once you have low-cost launch, it makes no sense to do lots of things on Earth, manufacturing, data processing, lots of things. The moon is interesting because the lunar regolith is mainly silicon, which is useful for manufacturing solar panels. And it would take about 20 times less energy to get the silicon cells from the lunar surface to LEO than it does to get silicon cells from Earth to LEO. So it does make a lot more sense to manufacture them on the lunar surface and then transport them to LEO if you want to have low latency. If you want to build big solar farms in LEO, then you probably want to manufacture them on the moon.
It's got to be pretty fun to think this big, Philip, and also start to execute on this.
It's very fun. I do like it.
It's incredible. How many people are thinking as big as you? You mentioned manufacturing on the moon. Some people might think that these are flights of fancy, but they're not. They're happening. You said these things are happening. Is it a very small group of you doing this or is there much more happening than we know?
There's a... I mean, it's growing as Starship becomes... looks like it's going to work more and more basically. There's a whole bunch of people looking at lunar, both mining of various lunar resources. I guess you guys have seen the Interlune guys, Rob Meyerson, they're trying to mine helium-3 on the moon. It's still relatively small now. It's going to be like, the capacity for Earth to put stuff in orbit is going to go up by about 100,000x in the next five to 10 years. So we're talking about the very early days of an industry. It's a bit like, it's like the first people to arrive in California back in the 1850s or whatever. That's how it feels to me anyway.
Well, keep us in mind as you run across these folks, because space is definitely, space technology is one of our threads that we explore along with quantum, along with AI, along with blockchain, a lot of different areas. I just had a question on that analogy to the gold rush. Then are you selling the shovels or are you digging for the gold? Because you mentioned that Starcloud isn't just data centers, but we kind of prepped for this that it was just data centers. So maybe you're...
I mean, it is. We've got enough to bite off with data centers to be honest. And data centers is definitely the first thing you would do. And it's also the biggest market. Like data centers in the end state is, I mean, in my view, data centers in the end state is like 99% of the economy. So like the end state of any intelligent star system is to build a Dyson Sphere hooked up to as much computers as you can physically support, which is like a Matrioshka brain is what they call it. That takes a few million years, but that is nice. Starcloud is the first company ever that is on a path to building a Dyson Sphere slash Matrioshka brain.
So for those of you guys just listening and you're not seeing us, half of my face just melted in a couple of those thoughts and words. But no, it's awesome. All right, Philip, one quick question. There was a point there I wanted to touch on. What was the start of your question? The analogy to the gold rush, are you selling the shovels or are you digging for gold?
I like to think we're digging for gold. I know it's very popular in Silicon Valley to be super keen on selling shovels. There are times when you just want to dig for gold. I like to think we're digging for gold.
Go big or go home. So all right, here's one question, and I want to actually unpack this pilot that you're getting ready to put up in the next few months, but I remember about 10 years ago when I was knee deep in data center world, solar flares was this big thing, and it was like, we got to prepare for solar flares, because when they happen, it's going to wreak havoc on all of our systems and all of this, and there were some effects for some big ones, but that being a little bit more applicable when you have these assets in space. Have you guys thought about the solar flare thing and what does that mean for you?
Yeah. So there was a huge solar flare in mid last year. I don't know if you guys saw the Northern Lights around April last year. Even Seattle, it was crazy. I think it was the biggest one in maybe 20 years at least. Not a single Starlink satellite went down and they had 5,000 in all of it. Yeah, so I mean, it's not as big of a problem as people I think think. If there's like a Carrington event in the 1850s that had this big blackout, if there's that scale, then yes, we're a bit fucked, but so is everybody else.
Everything on Earth is too, right?
Yeah, the data centers on Earth are going to have trouble with a Carrington event too, right?
Yeah. So solar flares aren't, unless there is another Carrington event, a major concern. Space junk, maintenance, cooling, this kind of matrix of challenges.
Cooling is by far the biggest challenge we are solving. I mean, people think space is cold and so cooling is easy. Because there's no atmosphere, you don't have convection or conduction, so everything needs to be dissipated with infrared, like black body radiation. So that means having very large, low cost, low mass deployable radiators, basically two sheets of metal that you're pumping coolant through. It doesn't need to be expensive, but right now they are very expensive. So that's our biggest engineering challenge is making these large, low cost, low mass deployable radiators. I would say that's actually like core technology for energy, for any high energy use case in space also, space manufacturing and everything else. The others are solvable. Space debris is not as big of a problem as people think. So the way, either you fly very low or you fly very high. So we're at around, we're flying in LEO for the first two satellites, 350 kilometers. It's like a self-cleaning orbit because everything naturally deorbits within a month because of the low levels of drag in the upper atmosphere. Or when the satellite's larger you can fly higher. So we have the problem with flying low is you need thrusters to maintain your orbit and the problem with flying high is you're in the Van Allen radiation belt. But luckily as the satellite gets larger, shielding is naturally taken up. That structure itself is naturally more shielding.
So let's unpack that a little bit. As you get higher, you have to deal with more radiation. As you get lower, you require thrust and propulsion to get you to maintain the spot there. And what happens if you don't have that? Does it just kind of back to Earth? If you're at 350km, it'll take you about a month before you burn up in the upper atmosphere. So how do you have enough fuel to maintain that? Like how does that work?
You use ion thrusters, so we have these two gridded ion, iodine thrusters. It's basically like a mini particle accelerator that shoots stuff out the back at 20 kilometers a second, and that means that you need very little fuel, so it keeps us up there for about five years.
Philip, I love that you referred to a mini particle accelerator like a little Pez dispenser in your pocket. It's just a little mini particle accelerator, it's all good. It's the size of a Rubik's Cube and it shoots stuff out of the back at 20km per second. Man, this could be the coolest episode I think I've ever been a part of.
Definitely will. To make it even cooler, it shoots Krypton out the back at 20km per second. Krypton or Xenon, of the two.
Krypton's real? Last week, the carryover question was from him, but we had Fintan on and their AI-powered satellites. And we were speaking about this low Earth orbit where they run out of energy and essentially drop from the sky and burn up. And he was speaking about this kind of new mindset of recycling these satellites somehow. Is that something that you're looking at, interested and aware of, or do you think that at the moment that's maybe a secondary thing to solve further down the line?
The problem is the chips need to be replaced every four years anyway. So we might look to reuse some of the structure, and in the white paper and the video that's what you see is this spine that stays up for longer which has the solar panels and radiators. But the actual compute module, it's unlikely we would bring it back to do anything with because after five years those chips are not going to generate cash for us anyway. So no, we haven't really baked in reusability for now. Re-entering a satellite is just expensive and time consuming and annoying. So I'm not sure exactly how they're meant to do it. If they're meant to re-enter, then that's very, very, very annoying to do.
Little harder coming down than going up, right?
