Starcloud: The Rise of Orbital Data Centers

[MUSIC] Hey everyone and welcome back to another episode of the Delphi Intelligence Podcast. Today I'm delighted to introduce you all to Philip, the founder of Star Cloud. Philip, thank you for joining us. I've got to say, reading your white paper this summer on my honeymoon in fact. Probably the most energizing thing I've read in a long time. I'm extremely excited to chat to you and learn a little more. Yeah, maybe you could start with just a little bit of an intro and kind of explain what you guys are up to. >> For sure, but thank you so much for having me and apologies to your wife for running your honeymoon. Yeah, we're building data centers in space in order to be able to take advantage of the abundant energy in space and the ability to scale almost indefinitely. So we're launching a demonstrator satellite in November this year, November 2nd, which will be about 100 times more powerful GPU compute than it's been to space before. So we're launching the first H100 from Nvidia. And then next year we've got our second launch, which will be at least 10 times the first one in terms of power output. And then we basically scale up from there, happy to talk more about the long term vision or other things, but yeah, I'm stopping. >> Yeah, we'll jump into it. I guess before we do, I'd love to understand what the hell, seemingly only a madman would go after something like this. Like what shape do you want sort of instinct to go after space, to go after such a crazy idea of shifting heavy compute off-world? >> That's a good question. Well, my background, I started my career as an engineer, I spent the first five years on the engineering side. I studied applied math and theoretical physics undergrad masters. And then I moved over to the commercial side. So I was with McKinsey for a few years working with space agencies of various governments. And as we already started to notice that the launch cost was very rapidly falling. So it used to be about $60,000 a kilo with the shuttle program. And then with Falcon 9 and Falcon Heavy, it came down to less than $10,000 a kilo. And what's really interesting is where it's going, which is that with the Starship program, it's going to be potentially in the hundreds of dollars per kilo, but certainly less than 1,000. And the reason for that is it's the first ever fully reusable rocket. So I mean, I started looking at a bunch of different things. Actually, I went down to Starbase, Texas, where they're building that program in 2023. And that was before even the first launch, just to have a look around, just like as a tourist. And what struck me there is the capacity that building force is absolutely gargantuan. So they're building these two massive Starship Gigafactories, one in Florida, one in South Texas. They're basically Tesla production lines, Tesla Gigafactories, four Starships. So they're designed to produce in 18 months, three Starships a week and in 36 months, three Starships a day, which is just mind blank capacity when you consider that compounds on itself. So yeah, if you build a new Falcon 9 every day, for a year, the end of the year, you have one Falcon 9. If you build a new Starship every day for a year, because it's reusable, at the end of the year, you have 365 Starships. So yeah, not only does it have five times payload-based capacity of Falcon 9, but it also compounds. So we might see a thousand X increase over the next few years in launch capacity with a 10 to 100 X decreasing cost. So anyway, that got me thinking about what kind of business models have we been wanting to try for ages that we couldn't be in a sci-fi kid since the sci-fi nerd since I was a kid. One of those is space-based solar. And so my co-founders were and I started discussing what would the launch price need to be for a space-based solar to work that's basically a concept where you have huge solar panels in space and you've been that power down somehow. The problem with it is you have a 95% efficiency loss getting power from space to earth. So we went to a number around $50 a kilo, which is within the range of Starship and actually that's what we started looking at. And then we were like, well, most new energy projects to restrily are being built for data centers. So what would the launch cost need to be if instead we had a cheap way to get the data center of space so we don't lose that 20x in power sending it out? And we read around those numbers and came to a number around that sort of $500 a kilo number and that's where the white paper came from. So that became then the basis of a white paper and then from there we got off to the racists. Very cool. Yeah, I mean, it makes a ton of sense. When I heard on another interview you talking about that, I mean, if you look at the composition of humanity's energy consumption or the forecast, right, it is all compute. So it kind of makes a ton of sense if you can shift it out there. You make it sound so casual, the journey into going after something so crazy. I read or heard somewhere as well. Like, I'm curious, you mentioned I think that you're one of five brothers. I thought I'd flag it because I also am. So yeah, it jumped out to me. It's quite a rare thing. Where are you in the stack? I'm the second Elvis. So yeah, three are still in production. And one's fully operational, let's say. But yeah, I'm curious, like, what the hell are all the brothers doing? Like, did you guys all grow up with this sort of interest in sci-fi? Is this a shared thing? Because suddenly it's something I share with my brothers. So I was curious about that piece as well. I'm only really with my identical twin brother. So we're the youngest of the five. And I'm the youngest of the youngest. So yeah, we competed quite fiercely in high school and university and things. Basically have identical career paths, actually. But no, my other brothers didn't really, I mean, two of them didn't really go to university. One dropped out of school when he was 16 and became a chef. One, he manages a kitchen at a golf club now. One, like, the came of bar man, like, manages a bar. And then now there's something else, but not anyway. I see not not not shooting like a date senses into space. And no, no, no, no, no, always curious. So yeah, I guess they want to then ask, where the hell do you even begin after going? You mentioned, you know, with your co-founder, you guys are like, damn, this might actually be a sort of viable path. Like, where do you even begin in sourcing the engineering talent and convincing them that this is a viable path and also that you're the guys to helm the ship and help execute the commercial side and put everything together? Yeah, that is a great question. So we initially thought to get great space talent. We need to be in El Scundo and LA. So that's where we initially put the company. And then it turned out everybody who wanted to hire was in Redmond, Washington, which is where we are now. And the reason that is Starlink is here and Starlink produces, I think 90% of all satellites launched in the last few years where it designed and built in Redmond at the start. So they say, so if you like, you know, any any role you can think of. So LA is good for launch propulsion aerodynamics. Something that's like spacecraft thermal engineer or anything like that, you're going to find much more of up here where we are. So we basically then relocated to here. My third co-founder, Adi, was through Zit Space X, so that definitely helped, you know, convincing some space X folks to join. And also 20 years of Microsoft on the data center side. So we got some kick ass engineers from Microsoft through him. But yeah, I mean, I think if you, you know, so we raised a bit of cash as well. And we didn't hire until we'd done a 21 million dollars seed. So you know, once you've got a bit of cash, you've got credible technical co-founders as red, you know, as PhD in engineering and designed the NASA lunar pathfinder deployables. And you know, then I think people are interested in emissions. So I think it's actually once you've got a few is it's not as difficult you might imagine. Interesting. Yeah, I guess the intrigue, the mission is exciting enough that I've got, yeah, it kind of makes sense, right? And no wonder people want to work on it. So, yeah, I'm curious then in terms of like, all right, you assemble the Avengers, so to speak, bring some sort of crack engineers together. Like, maybe you can tell us like a crazy fucking story from the early idea validation, right? Like where does one even begin? Like what were the first sort of engineering pieces? Like what were the crazy early learnings? Yeah. Well, I mean, we went through by coming to the last summer. So we were not in the workshop and we're down in San Francisco. So like Ezra was like, you know, you have these like goals every two weeks of the group. I think one of those goals was Ezra had to prove that a deployable, like a mechanism for deploying the solar panel could be done. And so he ended up like making that mechanism in our living room in our house next to the YCL. But no, you know, I mean, we try to help much different stuff. Typically, so we are not doing for the first satellite, we are not doing the satellite bus. That's just an off the shelf manufacturer. There's about 50 different providers of that. We don't need to reinvent the wheel on things like reaction wheels and, you know, well, propulsion. We did do it. But, you know, talk ride like all of the different spacecraft control mechanisms. We don't really need to reinvent the wheel on. So the bit that we're doing for the first spacecraft is the compute module, the antennas and the thermal, like thermal management system basically. So that constrains what we need to test and try somewhat. You know, it's not like we're building a whole satellite from scratch. So yeah, in terms of the different thermal management solutions, you know, we tried every different, we've got basically this liquid immersion called system. And so for that, you know, through a region of form, or of potential thermal fluids that we could use to keep that cool. But now we're putting an H100 in a cyclotron, which is basically a high velocity proton beam shooting protons at it at 20,000 kilometers a second to find out where it fails. Because I think we might now be the only people in the world that know where an H100 fails if you shoot protons at it. - Very high speed. - I love that. I mean, I guess we didn't really touch on like, the stuff that you guys described in the white paper right is building this five gigawatt data center right, a four by four solar array, a four by four kilometer solar array, and like, it was a one by four kilometer radiates, like this colossal ding right. And so, I guess that leads me to one of my questions, like, I think in order to do something that big, it needs to be, well, something about your orbital paths, which we'll get into, right, there's, you need to go basically much higher up into the, this radiation belt right. And so that's like, the first major engineering challenge right, is the radiation shielding piece. Like, yeah, maybe you can tell us just a bit more about that. Like, I guess, firstly, like, you've said elsewhere that it's like half the engineering time is dedicated to radiation shielding, and the other half is on like heat dissipation, right. So, maybe you can just talk us through these two big, like, challenges and how you're approaching each and what the key like variables you're trying to solve for right now are. - Yeah, for sure. So, with the radiation shielding, there's three levers you can pull. So, one is shielding, another is software, like mitigating the effects through software, and the last is choosing where you fly, choosing the orbital path. And so, yeah, for shielding, that's just a lot of testing with different materials and other things down in this cyclotron, you know. And other ways of testing. For software, my co-founder, Ordnant, and they saw that one of our first engineers was at the IIT stuff for microsoft for ages. And so, he has a whole bunch of experience in writing redundant software. So, if you have a bit flip, you can, you know, voting algorithms and things to know where that's happened. And then the last one is choosing the orbital path. So, for the first few satellites, we'll be flying in a, what's called a VIO orbit. So, very low, I thought, that's around 350 kilometers out of shield. So, it's much, about 100 times the less radiation there, than there is where, you know, a styling can always fly at 500 kilometers out of shield. So, we're pretty okay for the first mission. For the second mission, we're at 600, and then for later missions, as you mentioned, we'll need to be much higher. And the reason we'll be higher in the later missions is, if you don't ever want to go behind Earth's shadow, you need to, the lowest you can fly is around 1,300 kilometers. So, yeah, then you really are in quite a nasty high radiation environment. I mean, the nice thing about being there also is nobody else wants to fly there, so you don't have to worry too much about orbital maneuvering and things. But what's good about flying there is, once the satellites larger, the compute volume is scaling with the, or the compute scales with the volume of the satellite, whereas the shielding scales with the surface area. And so, you have a natural tendency that when the satellite gets larger, the mass of shielding trends towards zero, as a percent. Interesting. Yeah, and I assume that's a pretty big burden on smaller payloads, right? Like the amount, as in like the shielding occupies a significant amount of the mass going up there. Yes. No, it's super cool. I was also reading around like, yeah, the orbital path, right? Like, then what did you call it? A dawn dust sun, synchronous orbit on gas up there? Like, super cool, right? I like, one of my questions was, I know space is huge, and anyone that doesn't know anything with space says you don't really know the extent of just how big it is, but like, assuming these multi-giggle watt things, let's say 10 years out, 15 years out, like, can you really stack tons of these four by four kilometers of a raise up there? Like, yeah, just tell us a bit more about that. Yeah, do you mean in terms of orbital collisions and things like that? Yeah, exactly. Yeah. Yeah. So space definitely is huge. We do need to be very careful and make sure that we're a responsible user of space. So yeah, one of those we can do that is flying up in that higher altitude. I think the probability of the Kessler syndrome in people's mind is often somewhat larger than it actually is in reality. Kessler syndrome for the listeners, you might not know, is where there's a concern that if you had one satellite collision, that would create a debris field of millions of small fragments of satellite, which would then or could then collide with other satellites and create more debris, and then that would collide with other satellites and create more debris, and then you would just end up with this sort of dust of fragments in lower orbit, and that could make a lower-thorough but unusable. As a few reasons why that is much less likely than people imagine, especially at the lower altitudes, anything below about 500 kilometers. One is, for every orbit you go around your altitude low is quite significantly, if you don't have propulsion, so in hundreds of meters at least. And because of low lows of drag from the upper atmosphere, and so the chance that you hit something on the second pass is very low. And you might think, "Okay, that's all right, very well to say, but what's the evidence of this?" And one interesting piece of evidence is, in the 1970s, the US government ran this experiment to put 400 million tiny needles about this big into lower orbit, and they put them at about 3,000 kilometers of altitude, which is in theory a terrible place to drop them off, because there's as much, hardly any drag from that with a specific, I'll hide it to deal with it, right? It would take a long time. Yeah, so as I understand, at least five nines of those are down, so like 99.999 are down, for a cent. So I think it's probably only, yeah, basically the entire, the chance of hitting anyone of those is basically zero. And the reason they deal with it is because there's, you have this three-body problem with, so you have pressure from the sun, in terms of, just, in terms of both radiation and actual photon pressure, pressure, then you have this earth moon system. And if any time anything comes slightly close to the atmosphere, it's basically, then it's just a matter of time, then it's like a year or two before it's gonna be down, because once it passes within a few hundred clomters, or within say, 600 clomters of the earth, that's gonna start creating drag that will just spiral it in. So, yeah. I think it's reasonably unlikely, we'll have some Tesla-Caslingen effect. But yeah, where we're flying out at 59 clomters, I think it's, we can put many of them before we start having problems with that. Well, let's hope none of those remaining zero points, zero, zero, one percent, and he does come flying during the time. Well, yeah. I am intrigued though. I've been reading about the sophistication of a lot of the orbital debris tracking, they're starting to put ones to actually go and orbit in the atmosphere. I guess the crazy shit is just that some of these things are orbiting at like almost 20,000 miles an hour, right? It's like the collisions, if they were to happen, are crazy. But sounds like you guys are sort of beyond that. I wanted to also ask then, obviously radiation shielding, big challenge, sounds like you guys have unlocked a lot of proprietary data by using whatever that device was called. Like one of my favorite things I've read about was spaces that this Carrington event that happened in like the late 18th century, this massive solar flare that set telegraphs on fire, right? Like, I mean, what degree of like, how many standard deviations of your expected like CMEs and solar activity do you need to account for? Because obviously if you're shifting a multi-billion dollar data central, whatever it ends up being into, into outer orbit, like this is a huge thing to consider. There's two raised answers to that. So one is, yes, it's pretty unlikely, but it's not impossible, obviously, because of the Carrington event. The other is, yeah, a lot of our satellites would be a little bit fried if the Carrington event happened. But so would most data centers on Earth. Right. Unless you're planning on building concrete around all the data centers on Earth, we have the exact same risk profile. So, yeah, I mean, at some point, we can do certain things. Like we can shift calling from the radiators in front of the solar panel as we can reorientate the spacecraft so that it's instead of flat on like this, it's angled on like this. And I think what the, it will really require early, some kind of early warning and the type of early warning that would be useful would be a probe out in the first or grand point between the Earth and the Sun that could give us 10 minutes of warning at least. For now, it's not ideal to be lost. Yeah. I mean, I've seen that they have some tracking systems in place, where they, supposedly, if you could get 24 hours, they could shut off data centers on Earth and the energy grid and whatever. But anyway, I wanted to ask, because I'm curious for these like free events that we do know what happens. Like, I'm fully brought into this, the assist of orbital compute now. And I'm just curious on that front. Let's see. And then yeah, I guess the heat dissipation was the other one I went into asking and understand a bit better. Like, yeah, what the hell does radiating five gigawatts of energy like look like? Yeah. Yeah. So I mean, it's a huge challenge. That's definitely where the most of our engineering time goes. As the rough calculation is one square meter of solar panel in space produces 800 watts per square meter. So 200 watts per square meter on solar panel side. And the radiator, if you keep it at 25 degrees, will dissipate around 800 watts per square meter, which means we need about a quarter again, the size of the solar panel in radiator. So if your radiator is four kilometers squared, you need a one kilometer square radiator to dissipate that amount of heat. And so that is the core of the IP we're developing is very large, easy to manufacture, low cost, low mass deployable radiators. Yeah, interesting. We're up some interesting to the latter. And then yeah, I'm also just curious, like with those, with that level of emissions, like if you do start scaling up to that, is they're not like forced exerted in space, like on the actual spacecraft, right? Do you have propulsion on the other side to net it out, or is it not big enough even at that scale? We'll have some station keeping to net it out, yeah. Interesting. Super cool. And then yeah, I guess one of the other things I want to ask around is just like the data uplink downlink. I mean, I assume you're not going to be doing like full training runs or whatever in space. It'll be primary inference based stuff for like, you know, AI workloads, but like, yeah, how do you, is that a primary constraint, or is it very easily solved, just bringing up an update to either by lasers, or I don't know, yeah, would love to learn about that. My co-founders were passionate about the idea of a data shuttle, like a little module that we can send up and dock in and dock out. No, I mean, we potentially will be doing training runs at some point, but you're right, not for the foreseeable future. And the way that we ship data around is just through Starlink. So Starlink has extremely high inter-satellite connectivity data rates. So I think each satellite, and I think this is a public, is each laser they have is 400 gigabit, a second data rate, which is more than most data centers have as their, or more than many data centers, have as their in an outpike. So, and you can have multiple lasers, so you can have much more than everyone. I mean, very cool. I saw also that they're adding with the new V3s going up next year, they're adding like four and a half thousand terabytes a second or something of like capacity for the like terrestrial network. So I assume it sounds like it's solvable by that. Interesting. Yeah, I also wanted to just touch on like, yeah, what are some of the more like attractive properties of space broadly, right? Obviously, there's the scale of it. Obviously, there's like solar potency, right? Like outside the atmosphere, you can harvest a lot more energy. I think one of the things that you've touched on elsewhere that I thought was really interesting is just like, if we are heading towards Kardashev type one or whatever, and we do 100 to 1000X like, our compute usage on Earth, like the heat dissipation alone on Earth would fuck the planet. And that was a really interesting thought that I hadn't come across before. Maybe you could just tell us more about, yeah, that period. Yeah, exactly. Yeah, exactly. People often talk about global warming from the context of carbon emissions and the extra heat that they'd wrap. And you're talking on the order of a few degrees of warming through, expected through that effect. That is a potential, you know, threat that we should definitely mitigate. The bigger threat coming down the line is if you have, as you mentioned, on the order of sort of 10 to 100 times the current data center capacity, even if all of the current energy used that it becomes carbon neutral somehow. So even if it's fusion or solar or something else, you're literally just pumping enormous amounts of waste heat into the atmosphere in terms of boiling our freshwater, which is how current data centers are keeping cool. So at some point we would definitely will need to get compute off of. And yeah, 10 to 100 times is certainly, well, 10 times is certainly within our lifetime. So it's certainly what I'm thinking about now, I think. Yeah, I think the other big advantage of space, which is often overlooked, is these sort of permitting cycles are much quicker than they are on Earth. So if you want to build a new nuclear project on Earth, you're looking at sort of 5 to 10 years, even longer sometimes, lead time on permitting, whereas we can launch a 40 megawatt data center in a week, you know, it's sort of first come first serve on orbital planes. That is surprising to me that the regulatory muscle memory has got to a point where you can actually run these processes like relatively headache free. Is that likely to remain the case as, you know, if if starships, truers, vision manifests over the next 10 years, there's going to be so much shit going up there, it's going to be like trivially, you know, cheap, whatever to send stuff to orbit. So will that remain the case? Do you anticipate that getting harder? Is it a question of building these relationships now and you guys can, there's some defensibility in that long term? Yeah, certainly we require a lot of collaboration, the FCC, and I presume that there will be more regulation coming down the line as starship comes online and more people want to put more stuff up. Yeah, but for now it seems that the one interesting potential of this business model is in the same way that people have people who bought Spectrum 20 years ago and are making a killing off it, you know, and if you saw just Echo Star just made an absolute fortune selling Spectrum to SpaceX. There's one interesting idea, which is that once you once you have a satellite in an orbit, that's basically then until it de-orbit or until something, you know, unless you let it go, that's basically then your orbit. And certain orbits are quite valuable, particularly the storm dust, sunsink when it's all that. So there's an argument you made that even just by getting their first where building some kind of value. That's a really cool idea. I guess that was what I was asking earlier, right? Like how, you know, how scarce is that, whatever, that 12 and 1300 kilometer range? It's really cool. It's an interesting point. Nice. It's like the cooler version of domain sniping sitting on these orbital parts. I like it a lot. Very cool. Okay, so I wanted to switch gears a bit and talk through like, I guess the business model. So like your ultimate model, am I right in thinking it's just like selling compute times to hyperscalers like in the long term? Yes. Okay. And so, like actually we'd rather be thought of as an energy provider in the long term. We don't necessarily want to have to buy tens of billions of dollars worth of chips. And so in the end state, we would rather just, we provide a box that has Earth-like conditions in terms of atmosphere and power and calling and networking. And then people can do whatever they want with that box. If they want to run compute in there, that's great. If they want to do manufacturing in there, that's great. Or whatever they want to do. So in the end state, yeah, we can be thought of as a provider of power quality connectivity. And then everybody can do what they want with that. Very cool. And then yeah, I guess one, one, one, talk through like the roadmap from here, right? You know, you've mentioned that there's this like November 2nd launch like, yeah, what exactly you guys validating with that? These H100s in space, the most compute by 100x, I think you said in space, ever like, yeah, let's talk like talk us through a bit what the next like 12, 24, 36 months, like in terms of roadmap. Yeah, sure. So yeah, most powerful GPU compute by 100x. Then the first thing to demonstrate will be, will be the first to train a modeling space, the first to do high power inference in space, the first to do fine tuning of a modeling space will be running a version of Gemini on that satellite in collaboration with Google Cloud. So there'll be a whole bunch of firsts, which should be very interesting. Then next year, we're launching with the Black, Black, Black architecture. That's going to be a sort of close to 10 kilowatt satellite, which is massive for a commercial satellite. We'll have by far the largest commercial radiator in space. That one will be the first to provide a proper service for customers. So the first one will test out what customer workloads, but the second one, yeah, we'll have proper contracts with people who rely on the service. So with mainly with the OD customers initially, but also with some earth observation and sensing companies. Yeah, that's one piece. So like these early revenues are going to largely be like, you know, third party contracts, both commercial and defense do same. I guess for both, there's a lot of like data sensitivity, whatever you're doing, that like edge compute it off-world for like, how do you do the sort of containerization and navigate all of those pieces? It's quite an interesting one as well. Yeah, I know. Yeah, no, it's very interesting. So we've got the guy from Azure who was doing golf cloud security. So that's basically where you can, we can both physically and in software segregate. So we'll have a whole bunch of GPUs on the second satellite. We can have physical barriers between we can say, okay, these two GPUs are for certain DOD customer and they don't have any software link or they don't have any networking link with the rest of the satellite. But also just in virtually, we can segregate for different customers who are less sensitive on the security side. But yeah, everything is end to end encrypted and it will be uscures sending data through style. Very cool. And then so yeah, assuming you use work, you start scaling the early revenues, you're building, you know, slowly slowly launching bigger and bigger satellites. I think I'd read somewhere that you said like Starship at a hundred ton payloads will equal like roughly 40 megawatts of compute and you guys are building this big solar ray and like a modular spine, right? So you can basically send them up and assemble them over time. Like at what point does the construction of that phase start, right? Like going for a big gig of what, you know, yeah, center basically. So not until we're launching for Starship payload base, what it makes sense to hook them up modularly because we don't really need to, we can just launch bigger and bigger payloads. My expectation is we'll probably have like on the order of a few hundred up before we even start hook in them together because if you're only doing input in it, if you're only doing inferences, no reason to hook them together. You can just use these optical links to connect them. And so in that instance, yeah, and until we start having lots and lots of these things that we would want to train in model on and to train a model in five years time, I'd imagine we'll be on the order of several gigawatts. And so that's yeah, you're looking at about a hundred of these modules. So yeah, until we have a few hundred up by expect, probably not. And that's like in the early 2030s that will have the first full Starship payload they are. Yeah, this is one thing I wanted to double click on. I mean, I know you probably have access to privileged information around this, but I guess like without asking for for any specifics but say like, I see one. varying forecasts for like a what timeline or where it's even going to get to right is it going to be $200 a kilo a hundred or 50 or 10 some people the Gigable say like what like obviously a lot of your thesis in terms of like the full fullest fruition of it is dependent on I don't know There's probably a break even point right at like a couple hundred bucks. I don't know Um Like how confident are you in those like forecast how aggressive are you like personally based on all of your experience assessing this stuff? Yeah, anything like 200 is perfect for us. I don't think in my fashion model we can not write anything like 300 and that starts around mid 20 30s. It starts to fall around. We're breaking even around. Yeah, between the around $500 a kilo mark on the competing with terrestrial energy market. And if it's not a reason it's delayed we'll just run the business for other satellites. So, um, yeah, we're building business that is profitable with the Falcon 9 launch cost serving other satellites for now. And then you know if that launch cost is delayed in coming down there we'll just run that smaller business and when it's ready then we can start skating up. Very cool. And I'm curious like if assuming it does drag should it like should it drag. What is some of the other cool like wacky ideas you guys are toyed around with like a beyond just like you know whatever like EO stuff like defense stuff. Um I heard him one interview you did that you were saying you know joking perhaps about shifting like Bitcoin mining off well doing key storage and stuff. I mean, let's get out and make sure by the way like Bitcoin is a great way to use energy when it's not being. It's very similar to the Crusoe model actually where they have these huge mining rigs that they would stick. You know 10% of that capacity they'll put in very high power GPUs. And when because you know the utilization on those is much lower than the utilization that you can get from Bitcoin mining rigs. Because you can basically get close to 100% sending the failure rate on the chips. It's not a demand whereas for high-powered users only so much demand a certain times of the day and whatever. So yeah, we'll have some mining rigs on the second satellite. That will be so sick. I love that idea. It makes sense. We put a report out a few years ago. Pre-crazy AI boom and like at that point our estimates were showing that 20% of the energy footprint of the internet was already BTC right there's another big energy energy demand. So that's a really cool vision. I like I like the sound of that. I look forward to that day. I guess like I'd also love to just hear from you what some of the other less obvious challenges about space about either like specifically technically what you were doing or just broadly operating in space the space starts up like what maybe wasn't obvious to you at the outset the now is. I think what surprised me at the very beginning this was just how bad connectivity is in space. So you have to for most satellites you have to wait half an hour or between 30 to 90 minutes before you get to a ground station or till you're flying above a ground station and then you're only about five minute window where you can download as much data as you can and you're getting you know less than a little bit of second data rate. Most of the time a few hundred megs and so yeah like I had sort of fairly naive it just presumed that space was similar to ground in that there was connectivity and it was a bit like the internet and you see a Wi-Fi or something that. And that does exist a degree with some of the other constellations like T-Dress and iridium but if you want cheap fast connectivity for a little satellite that's very constrained and that's one of the main reasons now why people are interested in using our compute so that they don't have to they can run all this high power in space rather than down and. See that's one thing that strikes me is strange the notion that it's more it's more likely sooner going to be economically feasible to put compute off well and run it on the edge like edge compute then just fix down link like surely all of what style is being doing like it's that surprising to me and but yeah let's see how that side evolves and. And then also like I mean you mentioned you're doing some defense stuff which obviously you want me to do much on but like I've been reading a lot about just the you know geopolitical side of space and this this very sort of like China Russia and then US led sphere of like space regulation even the like lunar programs that are ongoing like yeah I mean I've been reading about like Chinese debris removal satellites wiping out other communication satellites accidentally like. Is any of this weird shit that's going on a concern for you or your customers like if you're putting material amounts of chips and heavy infrastructure of world like yeah how much how often do you guys run into this conversation yeah I mean yeah certainly something to be cognizant of there's a few few aspects so one is that all you know any offensive action in space is very visible so anything that's above a centimeter cubed in space can be tracked and is tracked on the ground. So any you know if there's been any offensive action by Chinese satellites against our own we would definitely know about that and that would be considered certainly a destruction of satellite would be considered an act of war. So you know this treats is in place to govern what would happen in that instance. So yeah the both US and the Chinese are building offensive satellites or spacecraft that can intercept and can take actions that I think we are we're fairly dependent on your space force to keep us safe in that regard. It's nice cool. So if you're going to blip a data center space you could also blip a data center on earth and it's a lot easier to send a missile data center and it is to send one in space maybe there's different optics around that but you know absolutely it's just yeah it's interesting to me that that's a source of friction geopolitically now that they're starting to position for all of this stuff I couldn't actually believe it when the US space force was announced all there's years ago I was like it was very very so far right. Yeah, I'm also curious then like I don't know like you mentioned you grew up reading a lot of sci fi and obviously I mean I kind of touched on the card of chef type stuff right like if we can start harvesting planetary energy using all this compute like what are some of the ideas that really blow your mind around this like around like where this stepping stone that you're building validating this as a direction of travel like what's the crazy shit that excite you about where we might go say like within this. Yeah, within the century I'm very certain we'll have started sending spacecraft to the nearest star to alpha centauri it will take quite a long time to get that initially you know in the orders of tens of thousands of years but it's definitely something which is start doing soon I think. Yeah, I think it would take on the order of five to 10,000 years to build a proper full Dyson sphere or matrioshka brain. Dyson sphere hooked up to shutters. Yeah, so what we're doing I think will prove that it's you know a you know there's once we've built these big radishes it's going to be obvious to everybody that this is where compute lives. And then to start validating the Dyson sphere idea you could you would start putting them in a sun orbit rather than an earth orbit because at some point we certainly will run out space in Earth's orbit so for less time sensitive workloads you can have any. The bigger structures in the Lagrange points the first. Yeah, you probably want them in the yeah any of the four Lagrange points probably not the second but yeah right and that's super interesting. I mean as you said that the matrioshka brains was an idea that I thought was so well right like to the for the same reasons that you concluded that compute and not just a solo energy harvesting is the way to go like I don't know why we send a Dyson swarm rights is like how do you transmit the energy basically so I love the idea of putting. Massive arrays of compute you know orbiting the sun it will be so cool it'll be a cool way when that comes. I'm that's going to happen or we're going to wipe ourselves out and I think it's probably more likely we wipe ourselves out given the implications of the family paradox but we'll see. Totally I love yeah the family paradox boss from this great filter all this kind of stuff I feel like we're going to we're going to unlock incredible insights into both of those like theories I think in the next couple hundred years. I mean yeah on that point where do you I mean you guys are facilitating AI you know the ultimate goal of a G I summoning digital God like how do you rationalize with that how do you feel about it it's both equally you know exciting and terrifying like I mean yeah how do you think about all that stuff yeah it is very terrifying. I try not to call it something digital God but there is a sense in which is. I mean no I think I am extremely concerned about the idea that like a swarm of the killer a hydrange would make minutes me of the basically the entire planet within a very short period of time and that is what we're planning on building it looks like both the US and the Chinese. And if these two things go to war we are very screwed and as I say my working hypothesis is that that is at the moment the most likely outcome because if that wasn't the most likely outcome or it could be anything but like it seems to me the intelligent life is very short lived in our galaxy at least because it would be extremely simple to call us galaxy like you could do it in like a few million years. And that's sort of the blink of an eye in evolutionary and galactic time frames so it implies to me that there hasn't been very sophisticated intelligent life anywhere in the galaxy before that survived more than a few million years. And so then yeah the question is why did it not survive more than a few million years and I think I think I think I being self destructive is a very plausible explanation for. for that. Well, Chinat by Spurs, onwards. It's new match you can do about it. I mean, yeah, the alternative is we are really the first and we are really exceptionally rare and maybe that is the case and yeah, I don't have any evidence either way. There's a quote that I always love by a physicist, I forget his name, who said that there exist two possibilities in the universe, either we're alone or we're not and both equally terrifying. Yeah. And I think that's so true. Like the responsibility, the weight if we are the only. But anyway, let's see, let's hope sense prevails and we figure out our current qualms better. Anyway, I'm curious so less abstract sci-fi visionary terms having immersed yourself in the space industry and as a space enthusiast, what is some of the other shit that's happening in the next 10, 20 years that's really exciting you? Like I kind of touched on like, I don't know that both the US and China have lunar programs, they want a permanent human lunar base in the next 10 years. Like what's the stuff that's really exciting you about where this industry is going? By far the most exciting is the permanent base on Mars. So that's the whole ambition of SpaceX and given the SpaceX is heading to be by far the most valuable company in the world, that seems like they have the resources to do it. So I would not surprise me for them in the next 20 or 30 years we see large cities being built on Mars or a large city that's self-sustaining being built on Mars. I mean, it won't be self-sustaining in 20 or 30 years but where there will be maybe hundreds of thousands or thousands of people on Mars in that time frame, which I think is just mega exciting. Other things are asteroid mining is going to be a big business within a few decades once we have a launch cost asteroid mining is going to be huge. I think tourism will be interesting, particularly closer to home, lunar hotels and Leo hotels, I think will be a big business too. I would definitely love to go to a lunar hotel. I would pay any amount of money to go to a lunar hotel. As would I, it's always been my great dream to take my father to space. He loves space but having looked into it, I mean, it still seems to suggest it's going to be like 20-minute a seat for the foreseeable future. Maybe, I don't know, let's see, hopefully that changes. I don't forget our changes with Starship. There we go. I guess, yeah, it's a bit like, that's a joke, close to home. If all of your forecasts are here, what does Star Cloud look like in 2035, 10 years out? It could be the case that in 10 years time, most new data centers are being built in space, purely because of the constraints we're facing on energy trash really. In 2035, that's enough time to build out this supply chain for Starship and the volume production of Starship. I think, by that point, we'll have proved that we are much more cost effective than building data centers trash-free. It might not be the only player in town, and I imagine there'll be some others, even some of the large players could enter. It would not surprise me if we start seeing most new data centers being built in space. That still means probably 99% of all data centers at that point are still terrestrial just because of this massive build out that we're doing right now. Yes, sorry, it was a little loud there. I wasn't sure what was going on there. Engine is arrived on the motorbike. That would be exciting to see. If you had to guess, based on all of your expressions so far, who would those other people competing be? Obviously, Elon, SpaceX, XAI, your curious who else you have an eye on. AWS has Kuiper. I think that would make some sense as well. All of the other hyperscaters would need this capability. None of them currently have space arms, so Microsoft doesn't Oracle doesn't open AI, doesn't Meta doesn't Google doesn't. That's where I would expect they would want to partner with somebody with that capability like us. I don't see Microsoft using XAI or AWS's data centers in space anytime soon. Then on the startup front, there's a few startups now, none of which have expressed ambitions to build very large data centers yet, but a few of those are Axiom spaces, Sophia space, Madari and the UAE. A few others. Interesting. Let's hope you guys' composition is the incredibly neutral one for all the other people that don't yet have space arms. I kind of touched on some of the things that were peculiar to you that you learned about the space industry or operating in space. More broadly, what is one thing that you now know that you wish you did when you set out to found Star Cloud? I don't have a very good answer to that. It's just been like continuous learning, but it's nothing that I feel was too long to learn. Getting around dealing with government, maybe we should have is certainly an interesting and beast. It's a lot more relationship than I'd imagined. I'd imagined it was like, because I have a lot of application forms and things like this, which you could fill out. But unless you've socialized that with the people making the decisions informally, it's much harder to get contracts. I think that's probably one of the main things. Interesting. That's very cool. I always like to ask, sort of, in closing, what would you say is the most important or impactful book that you've ever read? The one that comes to mind recently is the Elon bar of feedback by E.W. to Isaacson or Isaacman. It's very inspiring. Elon in general is very inspiring character to me. I'm excited to see where all of his endeavors go. But I also love that book. It was a very cool read. It's one thing when I was digging into even just the Starship forecast and all the naysayers. It's like, he's done this shit so many times. They've done what is just completely beyond the bounds of pausability. Fingers crossed, I hope it happens with Starship and the launch costs. Philip, thank you so much for the time. I know you're a busy man. I'm really appreciate you coming on and wish you all the luck in the world with November 2nd launch. I hope we can stay in touch. I'm very excited about what you guys are working on. Thanks so much for the very efficient.