The NEXT AI Gold Rush? Space-Based Data Centers Explained

Added: 2026-05-19

[00:00:00]Two of the biggest investment themes of space and AI seem to be colliding with space-based data centers being touted as the next billiondoll buildout. What I want to go into today is is this actually feasible from a physics standpoint and also the economics of it.

[00:00:16]And I'm going to do that by focusing on a company called StarCloud who have actually already launched an GPU, an NVIDIA GPU into space and have done real AI work on it. And with the topic of SpaceX coming up more and more now, they're touted to IPO very soon at a huge valuation, it's becoming a big topic that's spoken about in the investment opportunity because this really could be a huge buildout with a lot of value to be unlocked and hopefully investors can also capitalize on that. So, we're actually going to look into StarCard because they have already got an Nvidia GPU in space. And this is actually a private company that has been working on this actual concept for a couple of years now. And they have released a white paper actually a couple of years ago now that really goes into more of the details of the physics, the science behind it and also the economics of why this really could be a fantastic opportunity. So, this is StarCloud. This is a private company that was founded only back in 2024 with a goal of building huge data centers in space. And you can see here they've actually launched the first Nvidia H100 into space and they've been doing real uh AI compute and training from that as well.

[00:01:24]They have much bigger ambitions and um they're certainly their ambitions are certainly not small as well and they're already getting great funding behind this. They've recently just uh within the past couple of months secured $170 million of funding. And that's at a huge $1.1 billion valuation. And from the last time I checked as well, this company has under 20 fullterm uh full-time employees as well. So, it's getting a huge valuation. There's certainly a lot of people excited about this. What I'm actually going to go into today is not just kind of skim over the website and stuff like that, but they actually released a white paper when they started building the company. Now this is back in 2024 that they released this white paper but a lot of the findings are still really really relevant and it goes through just the basic case of why you would put AI compute in space. So what I think is going to be really valuable to do is to go through this and look at again like I said the physics of it and also the economics of it to see is this just kind of some sci-fi project that's being used to raise a load of money that will turn into anything or is this actually the future of where infrastructure is going to go. So we're going to go through first of all well they start off on this talking about power which is the main issue.

[00:02:37]Power is the main uh cost that is going into running these huge data centers. It starts off talking about um kind of all of the the big names in the AI space. Uh Elon Musk talking about and then just electricity shortage in general in about two years. Mark Zuckerberg, we would be building bigger clusters than we current than we currently can if we could get the energy to do it. Sam Wman again, we still don't appreciate the energy needs of this technology. So, it really is an energy first. That is why we're going into space to build these data centers.

[00:03:09]We'll get on to the first main point and it's uh operational cost which comes from power. So they've got here, significant operational cost savings can be achieved by using inexpensive solar energy without the limitations of terrestrial solar farms discussed below.

[00:03:24]Orbital data centers can leverage lower cooling costs and passive radiative cooling in space to directly achieve low coolant temperatures. Perhaps most importantly, they can be scaled almost indefinitely without the physical or permitting constraints faced on Earth.

[00:03:37]So those are kind of the three reasons they've kind of put out there. Number one, cheap uh very very cheap power.

[00:03:44]two, cheap cooling because you can use that in uh by radiating the heat away and then also you can scale pretty much indefinitely because you don't have all of these permitting constraints of building massive data centers in someone's backyard at home.

[00:03:57]So now which I think I don't want to go too much into the details of this but this is essentially the main core premise of why these data centers are getting put up into space and that is solar power and it's not just that solar power is cheap but it is also much richer up in the atmosphere. You don't have clouds and it is also and you don't have the atmosphere to block it and it's also running 24 hours a day if you're in the right orbit. So terrestrial solar farms in the US can achieve a medium capacity factor of just 24%. And that capacity factor is essentially of the available energy that is coming down from the sun, how much of that can actually be absorbed. And also it says while solar products in temperate regions such as northern Europe or the UK for example typically achieve capacity factors of under 10%. So of all the energy that can possibly be absorbed or taken in by the sun, only 10% in some case is actually getting absorbed.

[00:04:48]perhaps because of things like uh the atmosphere, atmospheric kind of distortion from that, clouds, weather, and of course the day and night cycle.

[00:04:56]And they say this here, a capacity factor over 50% is impossible because of this day and night cycle. But they said for proposed space-based solar array is greater than 95%, you don't have day and night. You don't have weather and you don't have clouds up there as well. And just to look at the rough kind of numbers, they're saying we will be able to offer an equivalent energy cost of uh 0.0 02 uh per kilowatt hour. For comparison, at the moment, they're looking from anywhere from 0.04 dollar per kilowatt hour up to 0.17 per kilowatt hour. So, a big big jump up potentially if you kind of go on the math of how just cheap you could get the power up up there. Number two, which is the cooling. These data centers put out a huge amount of heat and cooling them actually also takes a lot of energy as well. But what you do have up in space is uh a lot of radiative cooling opportunities. So you don't have conductive heat because there's no atmosphere to conduct it away. But you can take advantage of huge radiators. So deep space is cold which which is accurate in that the effective ambient temperature is around minus270° C corresponding temperatures of the cos uh cosmic microwave background. To use deep space as a heat sink uh to dissipate waste heat, a deployable radiator is needed. So these radiators are going to have to be very big just like the solar panels because it's going to have to kind of radiate this heat away. But kind of going into the maths of it, a 1x one meter black plate kept at 20° C will radiate around 838 watts deep space. Um essentially saying as a result radiators need to be less than half the size of the solar arrays. So essentially the biggest factor of this is still going to be the solar panels.

[00:06:37]If you have solar panels that big, the radiators are not going to be any larger than that. they're going to be quite a lot smaller than that.

[00:06:44]Also, they say as well, orbital data centers, they don't have any fluctuations or ambient changes in temperature. So, data centers just from where they are on Earth, they're going to have like day and night cycles where the heat changes depending where they are. And then also seasons throughout the year, and you're going to have to engineer these data centers to be able to deal with that, just like absolutely pretty much everything we have to engineer on Earth has to. It's going to be consistent up in space. They're saying that you don't also have to engineer for those fluctuations as well.

[00:07:15]This is a breakdown as well. So they kind of give a breakdown of what the cost is of actually running and I believe this is for a 40 megawatt cluster over a 10ear uh 10ear span amortized. So they're talking about the cost of the solar arrays. Um one thing which we will get into they're assuming a really really low cost of launch for here which we will get into as well. But you can see when they're looking at all of the cost of building a data center.

[00:07:39]So for example, building the enclosure, they're assuming that that will be the same. All the other data center hardware, assuming that will be the same. There's also um some other stuff that you won't need backup pl usage and stuff. Um but they are assuming a very very cheap cost of launch. What that basically gets you down to is 167 million versus 8.2 million. So if all of these things were to hit and of course you're compounding variables in a positive manner which you you're kind of really looking for a best case scenario here but you can see the cost savings will be huge scalability. Another thing they talk about here is what we mentioned which is without the physical and planning constraints that plague terrestrial projects of this size. You'll know if you invest in any data center stocks, actually being able to secure power and then get the the kind of uh the approval to be able to build a data center is a whole job in itself. And kind of with this as well, you're dealing with space.

[00:08:34]You just need to send more up there and then attach them to your data center which is already there like this as well. So speed of deployment in western countries, new large scale energy and infrastructure products often take decades or more to complete due to all the permitting. this would get around that. They're saying um given the larger physical size, orbital data centers must be especially responsible users of low Earth orbits. They're also saying that they are going to have to maneuver these as well, just like satellites or just like a ginormous uh space station that's up there already with stuff like spacecraft maneuverability and also what to do if this was to break up or when you need to re-enter back into the atmosphere.

[00:09:15]They do note that well if we're talking about the solar arrays I think they said on here it's around 4 km squared for these solar arrays they're going to be absolutely massive um and with that that is going to be the biggest part and that is kind of going to be up to be hit by any orbital debris but they are also saying as well for example the space station the space station itself is around the size of a football pitch and that's been up there for years and years and years now there is a lot of learnings that have already been done with space and orbital debris hitting solar panels and they're saying like they can basically deal with this over time networking architecture. So this is basically how they're looking at building them because you do have to think of how you're putting these together to make them very modular because you need to build much of it a lot of it down on Earth and when it gets sent into space it kind of needs to be ready to really simplistically or as simple as possible be attached and to start running. So they're talking about basically building these as set kind of containers. So with server racks inside them. Um and this will basically have one mechanical port, a docking port at one end which you're going to do everything for which is like connections, um networking, power, calling connectivity which will then go basically onto a branch to build this massive cluster of data centers. And these will kind of be managed separately. So if these start to fail, they will de-orbit it. it'll put it back into the atmosphere and equally they can replace them as they want or just extend and make the data center larger uh over time connectivity. So this is something that we do need to kind of be very wary of which is there's a huge amount of data that's going to be going on up there and they need to be able to connect this data and also beam this back down to earth and they're saying like they're basically relying on laserbased connectivity with other constellations such as Starlink or Kyper or Kepler. So they're very much kind of relying on these other laserbased connectivity uh constellations are going to be needed from what I can see here. So they're basically relying on them relying on them to be able to transfer data across and again laser links and if you're you're invested in RocketLab you'll know that laser links is a company that they are they have just acquired a laser link business and it's something that they are very very excited about in the future and the more you read about kind of the constellations of the future.

[00:11:33]This is going to be essential with transferring data because you can transfer a lot of data a lot faster and there's no permitting needed with like the radio frequency side of things as well.

[00:11:45]Power. So this is what I mentioned as well the size that these solar arrays are going to have to be. So a 5 gawatt data center would require a solar array with dimensions of approximately 4 km by 4 km. Absolutely huge. Um and they said that these cells like that is not a constraint in terms of being able to build that many cells. Um today, this is back in 2024 over 300 gawatt of power was deployed. This is back in 2023. Uh so it seems like they're saying that's not a problem in terms of the volume that is done there. Of course, putting that together and uh launching it, putting it together or unraveling it or unfolding it is going to be a whole another technical challenge. But that is something that I think they are really working in and that might be one of the core pieces of engineering that needs to be worked on that StarCloud are working at as well deployment. So like I said so deployment these arrays can feasibly be accomplished using design concepts have been demonstrated before in orbit such as Zfold roll out or picture frame designs and they've said this is a core area of development with star cloud. So this is one of the big areas of engineering. they need to build um solar panels that unfold to be this big and they need to launch it in a compact rocket to get it up there in the first place as well. So it does seem because when you're looking at what they're doing here, they're launching Nvidia GPUs which they don't make. They're probably not going to be making the solar panels themselves. They're going to be using Starlink. They're not going to be launching themselves, but it is around this area of actually building the actual infrastructure kind of around the data center and also managing that.

[00:13:16]That seems to be what StarCloud are like doubling down on and they see as their main kind of differentiator um in their business thermal management. So the f thermal load from compute modules needs to be transferred from the center of the data center which is basically where all the heat is going to be building up um and that needs to be sent out all the way to the radiator so they can get rid of that heat. The way to think about this is is, you know, it's really cold up in space, but they're taking sun energy from the sun from those solar panels. They're basically turning that into heat.

[00:13:48]They're using it for compute, but they're making a load of heat as well.

[00:13:51]They need to get rid of that energy through the radiators so they can take some more on and keep that uh the heat level at a maintainable level so the GPUs can run. So there's going to be several cooling calling loops using two-phase systems where pract where practical to reduce mass flow requirements, therefore reducing pump losses, but they're going to be using cooling loops inside here and then taken out to the radiators.

[00:14:15]Again, essentially all the power generated by the solar arrays and the remaining solar energy absorbed by exposed services needs to be dissipated as uh heat waste. And that is the other kind of main te uh technological challenge that they have there. bring enough power in and then also getting all of that heat out on the other side as well.

[00:14:34]There's a lot of stuff. I'm going to give you the link to this white paper as well that will be in the description as well. So, he's kind of got all the maths that go behind all of these assumptions as well, but I'm just giving you a run through of kind of what the bare what they need to be doing and what the kind of the top line of the science is. So, launch. So this is where if you follow kind of the launch industry and SpaceX and Rocket Lab, you'll see that this is their very very big expectations on the cost of launch going down. That is a big factor and that is assumed in these space-based data centers working. Um they're basically assuming two things and Peter Beck of Rocket Lab has said this as well. You're basically assuming that the cost of energy down on Earth is not going to get any cheaper or keep on going up. And you're also assuming that the cost to deliver stuff to orbit is going to keep on going down, which it is at the moment, but they're assuming a really big drop in that as well.

[00:15:26]Basically looking at the numbers that they're giving for Starship from SpaceX.

[00:15:29]They don't mention it here, but that is the vehicle that they are touting. So they say that the world's on the verge of a step change in la la la la la la la la la la la la la la la la la la la la la la la la la la la la la la la la la la la la la la la la launch costs thanks to the development of several partially or fully reusable heavy lift launchers which are expected to offer a launch price of around $5 million per launch long-term with a payload capacity of 100 tons to low Earth orbit um sun-synchronous orbit. This translates to approximately $30 per kilogram uh and it suggested that cost could drop as low as $10 per kilogram. So, this is taking this kind of the really top future best case assumptions that's been given on Starship, which is which is the vehicle that they're referring to here. And from interviews that I've heard with the founder of Star Code as well, they're assuming that they're going to be launching on uh Starship, which is a huge launch vehicle. It can bring 100 tons into low Earth orbit. And because it's fully reusable, it's essentially just going to be down to the costs of fuel and then some operational costs around that as well. But this is a massive step down. If you're assuming $30 per kilogram or even $10 per kilogram, just to give you some context, the Falcon 9 I believe is two 2,000 something like that in a best case scenario. 2 to $25,000 per kilogram at the moment. That is a rapidly launched superefficient vehicle. So that is a big jump down from what we're talking about.

[00:16:50]we can achieve at the moment that they're assuming to to make all these all this work. I don't know how the numbers would change and how feasible it would be if the cost to go to orbit wasn't this cheap. And then on to what this can actually do in terms of launching compute. Assuming a power density of 120 kW per rack, equivalent to Nvidia GB200, uh one launch can deploy around 40 megawatts of compute with rack level mass savings. Power densities are projected to rise dramatically in the coming years. So this estimate is conservative. So they're also saying power density densities are only going to increase as new GPUs come uh come out. So that is also going to be on the side of these AI data centers as well.

[00:17:36]Onto the radiation topic. So the choice of orbit must balance factors including radiation, aerodynamic drag, network availability and latency and space debris. So on radiation as well you are going to have to have some protection of radiation that comes into there. Um but when they're talking about the orbit the most important factor is having maximum kind of exposure to the sun. So the most important factor is continuous solar power generation and thus low earth dawn dust sun synchronous orbit has been selected. Basically if you're up in this specific orbit you're not going to have a day and night cycle when you've constantly got solar power. So if we go back down to this um diagram here, this is kind of what they're saying with the orbit and the solar panels will be constantly in the sun. So generating uh going back to that capacity factor, generating power 247 maintenance. So this is another factor despite advanced shielding designs, ionizing, radiation, thermal stress and other aging factors are likely to shorten uh the lifespan of the electronic devices. It says that the data center architecture has been designed such that compute containers and other modules can be swapped out in a modular fashion. So as they stop working, you can basically drop them out. You can deorbit them. You might be able to take them back safely if that is needed. And then another one can be put in. Of course, you're not really going to be able to do much servicing on these. So what I'd assume is this as well is there is going to be a lot more loss where if it was down on Earth, you might be able to do some small fixes or tweaks. That's not really going to be you're not really going to be able to do that in space or especially not with the technology that we have at the moment and they will just be burning them up back into the atmosphere most likely.

[00:19:17]You can see here we've got a view of basically what this looks like. They're going to build be building these big shells where they've got their racks in there and this will basically attach like into a branch of the big data center where you're going to have these huge solar panels and these can be attached and deployed and de-orbited uh as needed.

[00:19:36]Then into their conclusion, gigawatt scale orbital data centers are among the most uh ambitious space projects of all time sitting at the intersection of four trends. So why are we doing this? And it's these four trends. A drastic fall in launch cost that is making this available to launch a huge amount of infrastructure into space. The upcoming electricity demand crunch which we're all hearing more and more about. The growth in demand for large energyintensive GPU clusters and the proliferation of lowcost connectivity from mega constellations. All of these things are happening together and that is why the ideas of space-based data centers are becoming a lot more a lot more spoken about at the moment. And of course we've got StarCloud talking about this, but then we've got the big upcoming IPO of SpaceX who basically planning to do this but at huge scales and this is going to enable them to hopefully grow into that massive valuation they're looking to IPO at. So you can see here I wanted to go through this because it gives you a topline level and it gets gives us a way to discuss why are people speaking about this and it's not just sci-fi. If you look at kind of the topline physics and the maths of it, it does really seem to add up. Of course, we have whole other question of the actual execution here.

[00:20:47]There are a lot of things that really need to happen. That launch cost particularly is going to have to come down dramatically and they're going to have to do that consistently up to launching Starship like three times a day to be able to get those kind of costs down. So, we've got a lot of assumptions in there, but with companies like this getting this much funding and with SpaceX planning on doing what they're doing, there's no doubt that other companies, we know Rocket Lab have already started building solar arrays that can work for infrastructure such as this, like for massive data centers. And with how big these solar arrays are going to have to be, that's a massive part of making these data centers work.

[00:21:22]So, it does actually make me quite excited for the future of what this can mean for Rocket Lab as well. So, I hope this was useful. As I said, that was back in 2024 when they first kind of launched the company. They made this white paper. So, there's going to be some stuff that has changed from there.

[00:21:36]I'm assuming their assumptions have probably only got more bullish as the AI demand, the demand for AI is only going upwards. And I think I'm going to do a lot more on this topic as well. So, if you found this useful, let me know in the comments because with StarCloud interviews with the founder and also of all the stuff and the data that's going to come out with SpaceX when they're going to IPO as well, there's a lot of stuff we can talk about here and it could be a massive opportunity to invest in whether it's a new company or just as double down in the space infrastructure plays we already can invest in such as Rocket Lab. So, I hope you found this useful. If you have any questions or thoughts on this, do let me know in the comments and apart from that, I'll be back in the next video very very soon.