Data Processing in Space: Google’s Plan for Space Data Centers

Craig: let's talk about one of the biggest, most ambitious moonshot ideas to come out of google in years. it's called project suncatcher, and it's a plan to solve one of the biggest problems with ai. it's massive energy consumption by moving data centers off of earth and putting them into space. we're talking about constellations of satellites powered twenty four seven by solar energy, all to run the next generation of artificial intelligence. it sounds like science fiction, but the first prototypes are scheduled to launch by 2027. that's just two years away. google plans to launch a network of satellites that act as floating data centers in orbit. instead of building massive power hungry ai facilities on earth, they put their computer chips called the tpus in space where they can provide, be provided power twenty four seven by the sun. what do you think about this, john?

Jon: yeah. this is cool. so people might be asking why space? well, it's pretty simple. ai, as we know, is incredibly power hungry and we're running out of cleaner energy on earth to feed it. so in space, satellites can capture up to eight times more solar power than panels on earth because there's no atmosphere, there's no weather, and they can stay in constant sunlight. the other benefit is that we're talking nearly, so we're operating high performing computing hardware in space presents thermal and radiation challenges that differ fundamentally from terrestrial data centers. but space provides essentially an infinite heat sink through radiative cooling to 2.7 kelvin cosmic microwave. so that's crazy. there's a lot here. and like you said, instead of one massive space station, they're gonna launch these 81 small satellites. they're gonna fly in tight formation like a synchronized swimming team in orbit. and they're gonna each carry, chips, and they're gonna communicate using laser beams at incredibly high speeds. so together, they're gonna function as one distributed data center floating 400 miles above earth. this is cool. i think this is a great use of space, if that makes sense. thoughts? do you think this is a good idea, bad idea, or what are your thoughts?

Craig: well, it's interesting idea. the concept is very good. i guess the question is how many people will have to live in outer space to maintain it? true. is it gonna be bots only? are gonna be people? are we gonna be remote controlled? is this the beginning of skynet literally as, you know, the data centers in the cloud and we literally have no way of shutting it off anymore? there's all sorts of things that fly through my mind, but i love the concept of, hey. we're in space, so we can release as much heat as we want, and, you know, we literally can't affect anything.

Jon: yeah.

Craig: so you don't have any cooling problems. and if you can get, that kind of energy the, solar panels and solar, nets and stuff they use in space that will run these things, along with any sort of nuclear stuff they might have to use, great. i mean, getting it off the planet, giving us more space, not contributing to the climate dilemma here. the question become the climate, change heat and all that and giving us more energy. all sounds good to me. but i hesitate and say, okay. well, how much stuff is gonna be floating out there? yeah. and how soon till we can't see the sun anymore because there's so much, so many satellites doing so many things and so many data centers. like, the idea sounds great right now, but what happens when there's literally a thousand of these hive network, data centers in space?

Jon: yeah. so you bring up a good point about maintenance, and i didn't see anything in there about maintenance. but if we're talking about the economic viability so the economic feasibility, it actually hinges on dramatic reductions in launch costs. so currently, launch costs range from 1,500 to 2,900 per kilogram to low earth orbit. google's analysis suggests the break even point occurs when launch costs reach approximately $200 per kilogram, which makes space based infrastructure cost competitive with terrestrial data center energy expenses on a per kilowatt basis. the cost target could be achieved by 2035, and that's only if spacex starship system reaches operational status and maintains a launch cadence of a 180 missions annually. also, the analysis assumes that satellite operational lifespans of five to seven years to amortize launch costs effectively. so, obviously, this model is facing some uncertainty, because of evolving terrestrial energy cost, but, that's kind of where they're at. and that's a huge reduction from 2,900 to $200 per kilogram. i wonder if that, that centrifugal, launch can reach, those costs those low costs.

Craig: yeah. i don't know. that's very interesting. a good question to see if it could reach those low costs. yeah. they've, consist of, extensive radiation testing on their trillium, b six e cloud tpu hardware testing involved exposing the chips to proton beam exposure to simulate space radiation. the results are promising. so this looks, you know, very interesting what they're trying to do. the satellites will maintain formation of hundreds of meters to a few kilometers apart requiring precise station keeping maneuvers to maintain optimal position.

Jon: yeah. so the thing that system is gonna have to handle is dynamic network topologies because satellites are gonna enter and exit communication range with ground stations and each other. right? so they're gonna have to have some sort of advanced mesh networking protocols, some predictive link scheduling based on the orbital mechanics. they're also gonna need fault tolerance, distributed algorithms, and, like, they're gonna need a lot of complex protocols to maintain computational co coherence across the constellation for sure. like, there's a lot of challenges and hurdles that they're gonna have to overcome.

Craig: yeah. and one of the challenges and hurdles that, i talk about is the on orbit system, reliability demands redundant systems and remote maintenance capabilities because, you know, unlike terrestrial data centers, physical repairs become impossible once satellites are deployed. so they might be if they go down, they just replace it. yeah. remote maintenance for fixes and updates. but, you know, if you break a satellite, what happens? you know? what happens when, what happens when the

Jon: space debris collides with it?

Craig: yeah. when cloudflare does an update and it blue screens. satellites will be, built with backup components. if one fails due to radiation or defect, the backup part will automatically take over to keep the satellite running. but, you know, the plan is to make these satellites completely self sufficient and not to have it serviced by astronauts, which is interesting when it can actually service itself. yeah. we're in a different kind of world, and we're heading to that. i just don't think we're there yet.

Jon: think of it like this, guys. think of it like 81 satellites performing an incredibly precise aerial ballet at 17,000 miles per hour. so they have to stay in formation within meters of each other while communicating via laser beams. that's that's gonna be an, an incredible feat of engineering if they can accomplish it for sure.

Craig: and they assume all the propulsion systems will be powered by, solar?

Jon: yeah. so, apparently, be obviously, because of the the direct sunlight, that will give it enough power to power these, these data centers.

Craig: yeah. and i was more thinking of, normally, there's fuel or some sort in satellites to keep them rotating if they need to go back and forth. are they using something else for propulsion, to keep the things, synced up to do the minor adjustments to make sure everything can talk to each other perfectly? that's what i was wondering. and how long is there a fuel system that needs to be updated that, good question.

Jon: i'm sure they'll have batteries. yeah. well,

Craig: we'll have to keep a eye on this. and if they have batteries that run everything, i'd say the solar panel panels will charge all of that, and everything works great. yeah. so love, i love the idea. i'm a little concerned about, you know, as the satellites operate, dawn, dust synchronous orbit. that means they'll be following the sun. and as a result, as you get more and more clusters of them, we'll have all these satellite systems following the sun. again, we'll get a little bit of sky pollution for lack of a better word, and i'm curious as to where that will happen, whether the radiation hardening is gonna last, whether the redundant systems will actually work. you know? it's like when the light bulb burns out, you can't replace it. you just yeah. like, the light bulb, you gotta replace the whole satellite. so i'm curious to see how all of that, becomes sustainable. what do you do with that unit when it's no longer functional? something go pick it up? does it fall to earth? does you know, like, what happens? like, what's that? again, i'm looking far ahead on, you know, the whole cycle of this thing, but i do love the energy, the lower energy, putting that out there in space, the cooling so it doesn't affect our environment, and that it doesn't take up real estate on earth. it seems to be more and more expensive than ever.

Jon: yeah. absolutely.

Craig: so there you have it. google is literally trying to catch the sun to power ai. it's an incredible ambitious plan facing equally incredible challenges. it's it really highlights that the future of ai isn't just about software. it's a massive battle over hardware, energy, and resources. we'll have to wait until 2027 to see if those first prototypes even make it. what do you think? is this, the inevitable feature or the most expensive moonshot yet? let us know. leave us your comments.