Data Centers in Space Geek Out

went up and kept going. And so yeah, it got to Jupiter in record times, slung shot around and then basically gets shut down into a dormant mode for seven years. And twenty fourteen, they'll light it back up as it's starting to get close to Pluto. Just a hint. It got to Jupiter in a year, but it'll take another seven years to get to Pluto. It's just a long, a long way. I said it before. Those photos of Pluto were just amazing. Yeah, unbelievable and unexpected. Right, Yeah,

nobody knew. Nobody knew what it was, what it looked like. So it was really a great mission that it surprised everybody. In March, the second flight of a SpaceX Falcon one. This is at Omolek, which is a Quandullan Islands funded by the United States Department of Defense. Well, let me

tell you how this thing went. So apparently they got the first stage fuel mix wrong, so the engine wasn't burning as efficiently as possible, and that meant they had they were deeper in the atmosphere at second stage separation. That additional drag plus the fuel slash means that the first stage actually bumps into the second stage, possibly causing it to destabilize a bit and causing more slosh in the second stage. So it does fire up, but shuts

down early, doesn't reach ARP. But the good news is since the previous mission had failed as well, this time it had no real payload, just a weight essentially, and Elon called it a ninety five percent success. Okay, it's also the last time they fly on what they call a Merlin one in a engine, which is the oblatively cooltile. They'll move the next flight, they'll use the version they

call the one C, which is regeneratively cool. The modern engine that is not on the Falcon nine will be the one E. So that's their second flight, both of both of which have failed. But just you know that's reasonable. In June the return of more Shuttle missions. So' only gonna do three Shuttle missions in two thousand and seven because of the new rules coming out of the Columbia disaster. So the June Atlantis mission brings in the stree and

S four trust of the International Space Station. They're all space station flights now, and this is the second set of This is the third set of solar panels and the S three S four. So they remember when they first put together the station, they had one set that was kind of in the middle, up standing up high on the Z trust. Yeah, and then they then they started putting in the wings and so now the port wing and the starbough wing are in. This was the

starboard wing. Then in August, just a couple months later, they fly the S five trush, which is a connector to put on a second set. Not the most important thing for the space station. But what was interesting about the Endeavor mission in August was Barbara Morgan. Barbara Morgan, remember this is two thousand and seven, was the backup

to teach your Krista mccauliffe. Oh yeah, who lost their life in the Challenger disaster in nineteen eighty six and solo literally twenty one years later, her backup gets to fly as the first educator astronaut. Wow, I remember she stayed in all that time and got to actually do one of the first last Space Lab missions and have

that experience. Awesome. Also, in August, Voyager two cross the terminator shock, just like Voyager one done back in two thousand and fourth, is the point at which the solar radiation from the Sun is being pushed back by the interstellar waves and so our furthest objects out Forture one two. It's September. The Dawn spacecraft one of the first multi destinations spacecraft editor and it's going to the Internet. It's

going to the asteroid belt. It's going to visit vest the asteroid Vesta in twenty eleven, one of the largest, probably a dwarf planet. And then in twenty twelve it'll go to Serre's one of the very very largest, the largest, and it will see it will eventually run out of fuel in twenty eighteen and they will leave it orbiting Serres. So there's some hope that sometime in the future we'll get back there and find the old down spacecraft still pinn spin around it there. In October lasts of the

Shuttle flights for the year. This is Discovery flying the Harmony module, also known as No. To two, and so that's a big connector module, smaller than the habitat or anything like or the Discovery the big laboratory, but it is a big connector module. So that's No. Two. This is also when they take those central solar panels, the P six panels that had been folded up since the wings had come in and actually moved out to the

port end trust. Also in October the first commercial flight of the airbus A three eighty, a Singapore Airlines flight from Singapore to Sydney, and last of the space flights. In November, China's Chang E one orbiter goes to the Moon. It launched in October. It makes to the Moon in November. It will spend a couple of years orbiting the Moon

doing a detailed map of the lunar surface. They will deliberately impact it into the Moon in two thousand and nine, and then Changy two will launch in two twenty ten, beginning China's march towards the Moon, which still continues to this day. All right, on the computing side, clearly, the most important thing in computing to happen this year was

the first episode of run Ass Radio in April. Yes, that's not self serving at all, Oh my goodness, no, no, I think probably a little more cloud than that would be the release of say, the iPhone in June two

thousand and seven, a kind of a big deal. It's useful to note that at that time Steve job says, if you're going to program for the iPhone, you're going to do it in Safari, because he had already seen the writing on the wall for HTML five well before it was real, and didn't believe any that he was not going to give anybody direct access to the phone whatsoever.

He's of course wrong, and by July, within a month the phone is jail broken and people are starting to put software on, and he's about to lose control of his phone. Also in phone News, by November, Google will release Android, so the smartphone wave is underway, big way. On the Microsoft side. In January, Microsoft Vista is released to the public. There was a release in November of two thousand and six that was the enterprise only edition,

but the other editions release in January. At this point, people have already been criticizing Vista so heavily that sales are appallingly bad.

and all the codings done in JavaScript. Dot Net has not yet come to that land, although in November we get Studio two thousand and eight which has the dot Net framework three point five with link and stuff.

Gothrie decided, because he was going to All dot Net, who was very much about more testable web development, better quality of web development, all that sort of thing, he wanted to show this off. So though they apparently they did write the demo on the plane wow going down there to show this thing off, he did the keynote,

and I do remember that story blew them away. I would argue the most famous thing from that entire conference is a photograph of a terribly young Scott Hanselman holding up a large post it note from the open Spaces section that says, why so mean? Yeah? I remember that because, you know, the all dot net folks, at least some of them, were very vociferous about this idea that open source should be an important part of Microsoft's plant. Yeah,

I know, all point out NBC was open source. It was released via Coplex, and they.

I know, and that's what I got. You'll never need that much Okay, that's crazy.

happily retired software and database engineer. I recently read a sci fi story called Backyard Starship, where the satellite's orbitan never start to fall due to a cascade of collisions A big object it was hit. That's hundreds of smaller pieces into other satellites. How likely would a cascade effect like this be in our current age? And by the way, this story also talks about a specialized fleet of space ships cleaning up orbital satellites. I thought it was terribly

appropriate comment for the show we're about to record. Indeed, so Wade, I will answer this in line as we get deeper into the conversation about building data centers in space. Because the thing you're thinking about is called Kessler syndrome, and I ended up in order to make this whole talk work, I had to study it extensively from a statistical and engineering point of view. So I will answer it. So thank you so much for your comment, and a copy of music co Buy is on its way to you.

And if you'd like copy of music Gobe, I write a comment on the website at dot NetRocks dot com or on the facebooks we publish every show there, and if you comment there and I read it on the show, we'll send you copy of music go by.

for quite a long time. Although it really this current wave really starts at the end of twenty twenty five, which makes sense because we are coming to the edge of the AI hype cycle and people are starting to see that we can't scale as far as we need to. The lashback get data centers is real, and so meeting an alternative is an important part of that equation, and

this data centers in space is an effective alternative. Its question is doesn't make any sense, and I was struggling to figure this out, so I just ended up running all of the numbers myself and referenced a lot of material. And while I was putting this together, SpaceX was alfterually publishing stuff that seemed to line up pretty close. I would point out there is already a GPU in space.

So there's a startup, a y combinator startup called star Cloud that was on the Bandwagon four Falcon nine flight. This is in November of twenty twenty five, and so they had put up a microsatellite. It was about there was eighteen microsats on this thing. That's why they call it Bandwagon, right, It's this big, massive launch of microsatellites. And on board the star Cloud satellite was an H one hundred GPU. They were able to load a copy of NANOGPT on it and have a couple of conversations

with it. This, of course, is a microsat or a NanoSat. It won't last very long. They typically don't last longer than a year before there.

only last a few months. They one of their reasons to do this was to profile the radiation and environment in lower th orbit. So it'll be interesting to see what they get back from that. I mean, this thing's still flying as of recording, but it won't last too much longer, so they want to see how it survives. Yeah, and star Cloud's gone on to say that they want to put up eighty eight thousand satellites as already data centers.

trijunction once the expensive ones you'd use on satellites. So in order to collect five gigawatts of power, you need a solar array roughly four kilometers square. That's really big.

not what SpaceX is proposing. The orbit matters a lot and so to build something large like that, if we're ever going to do anything like that, you'd probably do geostationary orbit and geostation orbits about thirty six thousand kilometers up. And what makes that orbit special is that the orbital velocity at that altitude is the same as the rate of the rotation of the Earth, and so the illusion then is that the satellite is hovering over a point

in the land. Right. Sure. Now, we put satellites up there routinely, typically, whether satellites broadcast and so forth, because at that distance the Earth is quite small, and the solar power runs all the time. There's no nighttime when you're that high up, and you can use three satellites at one hundred and twenty degrees evenly around that orbit to cover the whole Earth. So it's a great place

where weather satellites are broadcast. The downsize is your thirty six thousand kilometers away, and so the ping times like six hundred milliseconds, which and so.

thousand up to thirty six thousand kilometers. Now your latency's anywhere gonna be built between fifty and five hundred milliseconds, that's normal. The area where you put GPS satellites, glonde, neass cloud AO, those kinds of things. So satellite are

rays again that are transmitting purely one way. Like GPS satellites are just constantly transmitting a signal of who they are and what time it is, and you as long as your receiver can see four of them, you can figure out exactly where you are, how fast you're going out Zude. All that KOD stuff that's NEO is great for that. If you care about latency, you want to be in low Earth orbit, right, So it's one hundred

and sixty to maybe one thousand kilometers up. So your latency is going to somewhere between twenty and forty milliseconds. And this is the realm of starlink okay, and other communication satellites, those kinds of things where latency matters, and it's a more crowded space without a doubt. It also is close enough when you're down to one hundred and sixty, two hundred and three hundred, even four hundred klometers enough, it's not like the atmosphere just ends. It just gets

more and more tenuous. And so what really finds Leo is are you still having your satellite slowed down by the atmosphere once you get higher than that. It doesn't happen. For the most part. Your orbits are pretty stable. Yeah, And it's not like the atmosphere is static either. It expands and shrinks depending on the amount of solar radiation it's hitting. So we happen at this moment to be a solar maximum where our star is particularly active and

kicking off large corolla mass injections. And that that's why we get such awesome auroras right now. Yes, you know, usually and when the stunt, when the sun is calmer, you'd only see auroras up in the Arctic or in the Antarctic. But right now, because it's so active, it's throwing so much energy at the Earth that we get these awesome auroras all the way down, you know, into the US. And when that happens, the atmosphere actually expands, it gets bigger.

get the Space Shuttle up there to rescue it. Only because we went through a solar maximum in the late seventies, there was a lot more drag on Skylab than expected, and so it deorbited well before the time they expected. Plus space upletok, way too long. We'd lost Skylab in nineteen seventy nine and didn't get Shuttle up till eady

one anyway. And I bring them all that up because when you are trying to track satellites, you actually have to scan them from the ground to figure out where they are because the atmosphere disrupts them on a routine basis, the same for a debris. Another aspect of satellite is important, So where you orbit matters, and then your inclination of orbit matters also. So if you're going to have a geostationary satellite, in order for it to be geostationary, it

has an inclination of zero akaa over the equator. That's the only way geostationary works, so that it stays in place essentially. But as soon as you're in any other orbit and you want to cover more of the Earth, essentially, then you change the inclination. For example, the International Space Station is an inclination of fifty one fifty one point six degrees, and that inclination allows it to pass over Russia on a regular basis because as much of Russian

modules there and they want to communicate with them. You know, communications have to relay from the ground, and so where it flies matters.

can do the polar orbits from there. It takes more energy, but that allows you to cover more of the Earth. So when you talk about starlink, the largest area of satellites out there, it's there in about a fifty five fifty six degree inclination orbit. There's a few variations on that, and that covers most of the land that's inhabited. It doesn't cover the poles, but it's close enough. Speaking of Starlink and SpaceX in general, this is an important part

of the equation. So before Starlink, which rerely started in about twenty nineteen, there was only about six thousand satellites had been flown. Ever at about three thousand of them operational at that time. But with the introduction of Falcon nine by SpaceX, which lowered the price to flight traumatically. Today, not counting Starlink, there's about forty five hundred operational satellites.

Anybody who has a choice of launch vehicles to fly into space is going to choose Falcon nine because it's so much cheaper. It's literally like five times cheaper than the next closest competitor. The notable exception, of course is China, who's not allowed to fly on Falcon nine, but everybody else already does. But Starlink alone, they do one hundred plus launches a year of twenty five to thirty satellites,

and so today there's about ninety five hundred starlink satellite. Wow, so double everybody else out of twelve thousand launch So they've already lost three thousand of them now and they go ahead.

The only when they need to will they relay between satellites, which they can do. So this is how you get coverage, say in the middle of Pacific Ocean, right where there is no land immediately accessible, and so they will relay across a couple of satellites before they come back. Doesn't have to go very far, and lasers work pretty efficiently, but it is slower to do that. But your best performance on starlink is literally a single pin back down

of the terrestrial network. Understand that the fully populated starlink network, which is not quite there yet, they should finish it this year at twelve thousand satellites will be the complete network. You're only got about as much about seven six hundred to seven hundred gigabits of total bandwidth terabits of total bandwidth capacity, which is one percent of the traffic over the Internet. Basically, the total capacity of starlink is the equivalent of one or two top of the line under

sea cables. It's just not that much bandwidth, relatively speaking, and there's a reason for that. You can just cram a lot more data down a piece of fiber than you can through an atmosphere or even across space. It's just not as focused and so you know the cable. The undersea cables are long, but they're not as long as space is long, so they're very efficient at moving data arounds. So almost all data moves under the NDERC

cable network. Very little of it moves across systems like starlink, but it's a last mile solution, right, you can get that relay point and get it back down again. All right, But we were talking about building data centers in space right, specifically for AI compute. And this is where this whole story comes up, as the situation with the AIH cycle has gotten really serious and folks don't want too many more data centers. So the sort of lash back has

been Okay, we'll just build them in space. The question is what's practical. So we got to start. We're going to build a satellite, so first us, we have to start with our payload, because our payload's going to define a bunch of things around the satellite. The unit of work that I've chosen for this is an Nvidia NVL seventy two. Okay, so this is a seventy two U RAQ hence the name that has thirty six Grace GPUs and seventy two Blackwell GPUs for total compute power about

seven hundred and twenty petaflops. So that's powerful enough to run GPT four full on n FP four mode, so trillion parameters could run in this rack. Yeah, okay, it's powerful. Just to be clear, it's one point three metric tons. Admittedly that's because it's in a housing and it's got metal cases and stuff like that that that weight is noted for making your floor strong enough if you're going

to build out a bunch of these things. And they run about three million dollars a piece and consume one hundred and thirty five kilowatts of electricity, which is the same as the entire International Space Station. Actually wow. Now, just to get scale on this, Microsoft recently announced in one of their giant data centers they installed forty six hundred of the okay, and normally these racks sit side

by side. They're liquid cooled, so they have pipes running through them to do their cooling, and they're strung connected together within finiband connections. And finiband connections are four hundred, eight hundred or one point six gigabits per second. That's terribits per second if you get to one point six, so four hundred gigabits eight hundred gigabits are one point

six terabits per second at sub millisecond latencies. So when you want to do a model build, if you're going to make a new version of a model, you need a lot of compute, huge amounts of compute, and it has to be tightly tied together because there's lots of communication going between all of these machines. And that speaks to if we're just going to fly satellites with one rack in them, we really can't build models with that.

We can do work right, you can utilize the model up there, but you'd build models elsewhere, maybe in geostation orbit one of these massive systems, or maybe on the ground. But if you want low latency in low Werth orbit, you're going to be smaller than this. And so a rack like this is kind of as big as you can go realistically. Now, the pay that racks only about a ton ton and a half. Maybe we can lighten it up with aluminum framing and so forth and get it under a ton, But now we have to build

a satellite around it. So what are the components that go into a satellite. Well, you need a bus, so basically a carrier that this payload's going to sit inside. It's going to need some on board compute for operating the satellite itself, managing, doing health monitoring, command and control, and so forth. A big piece of every satellite is what we call attitude determination and control. So I need to know which way I'm pointing and be able to

change my pointing direction as I need to. So you have a need, you have a set of tools in there. Main one will be inertial measurement units. So these are a kind of gyroscope that basically measure the orientation constantly the satellite. They're very precise, but they do drift over time, so they need they do what we called relative tracking. You need correction every so often, so then you need correctors, and they're basically a couple of species of those. One

is what we call a star tracker. So these are small telescopes electronically controlled that are watching for certain constellations and knowing roughly where you are and what time it is and what stars you're seeing, you can correct your im used to be more precise. Obviously, you can't use those when the sun is out, so we also have sun trackers, and the sun tracker does the same sort

of job. It can infer a location based on the location of the sun and the time, and so between those sets of instruments is and they also use GPS in certain cases for this as well. You can figure out where you are and which way you're pointing, so that is your attitude determination. Next comes your attitude control. How do you turn the satellite, because you're going to need to do that. You need to keep your intennas pointed towards the ground for communication, you need to keep

your solar panels pointing towards the sun for power. You need to keep your cooling away from the sun to disperse heat. And so you have reaction wheels. Reaction wheels literally are weighted wheels spinning in different orientations inside of the chassis. And when you increase the spin on one, just make it spin a little faster. It'll force the satellite to rotate in an accent and you just keep doing that to keep your alignment the way you want to.

Although eventually those flywheels hit top speed, which they call saturation. And it's saturation you're gonna need to d spin, and so you need to have a backup pointing system for while you're d spinning or desaturating, and that's when you might use thrusters. Or at least if you're in the lower orbit, you can use magneto torkers, which you'll literally press against the gravitational field of the Earth to orient the satellite while you're desaturating.

really have to do with the payload. We have to power it and we have to cool it, so obviously it's going to be solar panels to power it. The challenge here is it's one hundred and thirty five kilowatts plus whatever the system itself needs. But we can probably go with one hundred and thirty five kilowots. We get thirteen hundred and sixty watts per square meter while we're in orbit. We're going to be able to collect about

thirty five percent of that. So let's go with about three hundred watts per square meter, So to get one hundred and thirty five kilots, that's four hundred and fifty square meters, So that's panels thirty meters long and seven and a half meters right. Actually, it nets out almost to exactly what the solar arrays are for the International Space Station, although there are older rays up there and there are new arrays, and new arrays are smaller, so

we have gotten more efficient with trying quadjunction panels. These still would be big, big, big wings, and they're heavy relatively speaking. So you know, our operating pala was only about a ton. This would be three tons of solar panels to generate that much power. Now, if you've got one hundred and thirty five kilowatts of power generation, how much cooling do you think you need?

do conduction or invection with. So your only way to dissipate heat is radiatively. And so you've seen these on the International Space Station, these cooling arrays. They are or a U panels, these big white panels. So they actually pump liquid ammonia through. Wow, because ammonia functions very well in space, tends not the freeze, tends that it doesn't boil too difficultly, and so and the phase change is useful,

will take the heat away. So you use ammonia in the cooling loops on the processors and as that vaporizes, it takes a lot of heat away. You then pump it out into these radiators and they dissipate the heat radiatively, and then that liquid comes back and loops in again. Now again we can count on the International Space Station for this is about the same power. There are two sets of radiators with three spars each eight of eight panels.

They actually are heavier than everything else. They are about three metric tons each and you need two of them, so it's six tons worth of cooling for three tons worth of power in a one to one and a half ton paid Now.

in the dark. Think about the power system that that involves. So for forty five minutes you're collecting electricity off those panels, and you're also charged up the batteries for the forty five minutes that you're in the shade where you're running entirely on battery, and then you have to do it again over and over and over and over again. So that also speaks to we're gonna need some honk and big batteries for this as well.

the one that Elon's talking about. It's actually going into a sun synchronous orbit where you're never in the shades, so you don't have to hold the batteries run because batteries we're out and they need chargers, and those chargers can generate more heat. So as soon as you go down the battery path. You need more heat management. Ramming that power into those batteries that quickly is very hard.

Final piece on this thing would be the communications systems, and again we're used to InFine ban, which is incredibly lit latency, incredibly high bandwidth. We don't have the equivalent in space. The fastest system is actually made by SpaceX. It's their optical mesh, which does do one hundred gigabits, which is pretty compelling, and you could probably gang these

up if you need more bandwidth than that. Which you're not going to be able to get away from is the latency because you're still transmitting some distance and so it's still several milliseconds between satellites, which is fine when it's starlink, but if you were trying to mesh a bunch of these satellites together, as if they were in a data center, it's not fast enough. You're going to have problems. And they're not particularly heavy, not even particularly

power consumptive, but they are. They're just not as efficient and relating to the ground. You're going to go with various kinds of microwave transmitters, and so we've got a rough concept of a satellite now something in the neighborhood of twelve metric tons, which is big but not unprecedantly big, and big solar arrays and cooling panels on this and if you actually go on the SpaceX website, they talk about the AISAT Mini and that's pretty much exactly what

I've described. So, but they are big. The largest commercial satellite in the world today is a satellite called Jupiter three and it's about ten metric tons. There are military satellites bigger than that Keyhole was fifteen, like, there's probably some twenties out there. So we're in the ballpark at twelve metric tons. That's very feasible, but that also now gets into our launch and location problems. Before we get into that, why don't we take a break.

blew up from Blue Origin, we could lift three. Starship Block three. The current one that we're testing right now could fly eight. Right, So it's going to take a while to lift a lot of these. You remember Elon was talking about flying millions of these, which is kind of crazy. Uh, for no other reason? Then, how many satellites can you reasonably fly? Yeah? You know how many

of these aisats are you're actually going to need? When you think that Microsoft alone in one data center put in forty six hundreds NVL seventy two's you know you could that's forty six hundred satellites. Like that's about you know, five hundred launches of starlink alone. That's a crazy number. How many things to reasonably need? And they also makes the question like how many can we actually have in

lower th orbit? Like what's feasible right now? We have somewhere in the neighborhood of twelve thousand satellites in lower thorbit, most of which are starlink, and they're not the only network right Amazon is working on their LEO project, formerly known as Project Kuiper, which I'd like to name better, but now they're calling it LEO is in lowerth orbit. They've already flown three hundred to their satellites. Their goal

is thirty two hundred satellites. They think they can provide a good array of thirty two hundred satellites, so I guess we'll see. I know this folks who would like to use something other than starlink, So that's a competition coming up. And then there's the Chinese, because the Chinese are actually proposing three different mega constellations. The one that's the farthest along is they call the thousand Sales constellation.

It's out of Shanghai. They've flown seven hundred and fifty satellites already and they plan on global coverage by twenty twenty seven, coming maxing out by twenty thirty at fifteen thousand satellites. So if you're doing the math, we got about fifteen thousand satellites now all told, plus another fifteen thousand China three thousand plus from LEO, So we're thirty

three thirty four thousand satellites. And then there's two other networks that Chinese want to fly, one specifically military orient and the other one's a different project altogether.

bit secretive. Not everybody's talking about this in large scale, but NASA has to publish their numbers for the Space station and so I went back and looked at the space station being the largest object in order by far, it's the size of a football field. Before starlink started, back in the day twenty nineteen, when there was about three thousand operational satellites total, which the space station was one of them. There was also twenty thousand tracked objects. Okay,

so that's is that space junk space junk exactly. And the goal here is to keep everything at least three kilometers away from each other. And so you're using radar on the surface, especially in Leo, because the orbits keep changing based on atmospheric effect, and they would have a couple they get a couple dozen mornings a day, most

of which could resolve without a maneuvering at all. Is like, yeah, no, they're not going to get that close because they would have to do more precise calculations each time to sort of adjust that. Most satellites pre twenty nineteen never needed to maneuver to avoid anything. The space station, as the largest object up there, would maneuver once or twice a year. Now.

Partly that was because the projective boundary around the space station is much larger because there are people on board, and so we're way more careful, but also it's very very big. As starlink has grown as a network, SpaceX is instituted a mechanism where the satellites are talking to each other and avoiding each other because they're all in similar orbits, and so they now all maneuver almost every day.

They're ninety five hundred so satellites, and the space station's reevaluate its location is reevaluated three times a day to cope with potential junk and so forth, wow, and is at least a maneuver every other month to avoid debris. And that's at fifteen thousand satellites and about fifty thousand total objects in order now is twenty twenty six. So the fifty thousand objects that we're tracking, fifteen thousand of them are working satellites, fourteen hundred dollars defunct satellites no

longer operational. There are nineteen hundred rocket bodies. So these are the largest inert objects in orbit. Now, what's a rocket body? These are typically top stages that were used

to boost into high orbits and then weren't rendered. Most boost stages or second stages, deliberately deorbit, so they save a little fuel and after they push the payload off, then they wait for the payload to get away, they turn around and they do a burn to deorbit to land in the ocean, especially with larger spacecraft, where it's very likely that the upper stage will make it to land will actually hit the ground, so they aim to land in water.

to not generate more debris. Please, no more any satellite tests, stupid deorbit, your bodies, doorbit your satellites, like, be responsible about it. But there is a whole conversation about, well, how many satellites are reasonable in lowerth orbit, like how many could we really handle? And I found a study published by the Miter organization, which is a non governmental

entity that does this kind of research. They published it in Breaking Events in twenty twenty four, and what they talked about was these mega constellations, these huge eras of satellites like Starlink, all talking to each other, all cooperating to avoid each other. So that would mean you know, Amazon and SpaceX getting along and doing shared communications to avoid each other. And the Chinese government also playing ball on this as well. Wow, the belief is they could

get to one hundred thousand satellites in lower th orbit. Jeez, now, considering we're already talking about forty fifty thousand satellites just for communications, these low latency communications solutions. It doesn't leave a lot of room for aideas. There's certainly not the millions that SpaceX has talked about.

their satellites is only five years twelve thousand satellites. That means every year they expect twenty percent of the fleet, or twenty four hundred satellites to deorbit. How much is that? You know, that's fifteen hundred and eighteen hundred metric tons of debris burning up in the atmosphere. It's just a

lot of aluminum. It's a lot. Yeah, And it also means you're going to have to replace it, and if you only flit twenty five per fulk and nine means it's one hundred Falcon nine flights just to maintain the network right now, presumably with Starship, they're going to be able to fly bigger satellites and more of them, so they will you know, that equation will change a bit. But yeah, there's a lot of debris coming in the nineteen hundred. Rocket bodies are by far the most concerning

thing today. Again, they generally deorbit these things that they can, but they don't have enough fuel to do that. At least they depressurize them. Because if you leave a pressurized vessel in orbit, sooner or later something's going to break and they're going to explode and make debris. And so today, if you're doing a geostationary transfer to something, you're probably not gonna be able to get the booster back onto the Earth anytime soon, and so after you release your payload,

you'll at least depressurreized. So it's in a nerd object. It's not dangerous with satellites. Many satellites now it's normal practice is simply maintain enough fuel on board that before you're fully out of fuel, you will burn the last of your maneuvering fuel to put yourself in a deliberate deorbit into the ocean. That's just you know, response behavior.

There's also some passive deorbit strategies which have always been tested but never used, like a solar sale is a real thing, a very thin film of capon or something similar like that, like aluminum foil that you stretch out that the solar radiation, the particles coming from the Sun actually pressing against and will slow you down. So you show it in the prograde direction so that it's causing

a retrograde acceleration. So you put it up when the Sun's in front of you and it'll slow you down, and then what's behind you, you close it up so it doesn't accelerate you and you can speed up your deorbit. Depends on the size of the satellite, of course, only smaller ones. So that's going to work for our twelve ton beast that is this AI data center. That's probably

not a practical thing for. Further to Wade's comment on the show about managing debris, there are now as cost of flying the orbit has gotten cheaper, we're starting to see more initiatives to deliberately remove debris, so like actually netting debris and dragging it down inner accepting the larger stuff like old rocket bodies and hard putting them or capturing them, and then redliberately reanturning because you have fuel

on board. And then the coolest thing is for the smallest the debris lasers because laser and space are always fun, but the reality being, if I can point a laser at a small object and heat it up, it'll actually cause more drag on it. So if I time it the right way, I can actually force these things to re enter just by hitting them with microwaves or optical lasers. So we can put together more initiatives to reduce the amount of debris, but we're limited to how many satellites

we can put in lower th orbit. To be clear, it's definitely a limitation, and it's not comparable to anything that modern data centers can possibly handle. It were it's not going to build a million of these satellites. Ten thousand probably is the reasonable limit, and even then it's questionable. Yeah, and so there's a few places trying to raise money. But the only reason I think this is AI data

center thing has got any traction. Google did a paper on it called Suncatcher, But it's SpaceX that gives it credibility because Elon jumped all over it. And that really leads us to this whole conversation about the SpaceX ipo, which just happened shortly before we recorded this. Why the heck did SpaceX ipo money? It's weird, isn't it. Well, does he need it? He's already the wealthiest man in the world. Well, now he's a trillionaire. Oh boy, great, yeah,

I know that's a feature. And the more interesting thing is if you go back and see him talk about SpaceX. He always said, I'm going to keep this as a private company because the goal is to put a city on Mars, and that's not what a public company would do. Sherylders aren't going to go for that. There's no money in that. This was supposed to stay at private offering. So why did he change his mind?

strike price, they're fine. You know, you just collect a low interest on that which we can cover. No big deal, and you keep going. And this is what billionaires do, right, They just borrow against their stock.

on new cars per se. And I think it's simply because he would never get the offering that the valuation that he need if he's just making more electric cars, which is why he's talking about stuff like robot taxis and androids and things like that. Good for keeping the stock price up, because if it reaches the strike price on his loans, he's got big problems.

already dominates the launch platform. Eighty percent of everything that goes into space goes via SpaceX Falcon nine anyway, so you know, he had to publish the numbers of what his company actually can do, right what they call a total addressable market. He was talking about three hundred billion or so for flying other people's payloads in the space. Eight hundred billion for starlink is awesome, but it's not

a multi trillion dollar company that way. And then there was the AI side, Like the AI side is where he could say, oh, this is a twenty six trillion dollar offering, right that the opportunity to address this market is trillions of dollars. So it's very self serving to say, hey, we're going to change it now. And see, if you have to read the whole IPO. In it, it says we may have an issue with the number of satellites we can lunch. So they're admitting it, just people aren't

reading it closely enough. But the reality it is he's fundamentally crippled SpaceX's original mission of puking a city on Mars. Not that I ever thought that was a good idea. I never did either to I'm sorry be able to IPO this thing and fix his own loans, and he's leveraging that on the back of these AI data centers, So he's got a clear incentive to do this. Is he going to build some of them? Sure? Does it really make sense?

from people. Treat it like heavy industry, where it should be away from people. And you're also responsible for your own power, right, And that really, if you're really going to lift this much stuff into space, if you really got that capability, shouldn't we go back and look at

space based power again. We did a whole show on this, We did Starship Comes Online, and it's really as cheap as they believe it's going to be and by all cases it should be twenty to two hundred dollars kilogram to low Earth orbit compared to fault at ninety eighteen hundred, which is a bargain without a doubt. But this would be literally in order or to magnitude less. So I went and dusted off the latest plans. The British Astronomical Society does a review of space based power and a

regular basis. The latest set is from twenty twenty four. The design they call SPS Alpha Mark four and the specsond are pretty compelling. A gigawant is about thirty two hundred metric tons. Now you'd put this in geostationary orbit, so it would always have sunshine all of the time. There's lots of concerns with how big this thing would be be several kilometers long, so it would be an

incredible thing to build. And you are transmitting the power to the surface with microwaves high frequency microsoft nine gigahertz or so. The power the energy is not dense enough to like cook birds in flight or anything like that. Yeah, but it would land on a circle about four kilometers across. You'd put us at a rect tennison that to collect

that electricity. You're probably gonna want to put a fence around that thing because you don't really want people walking around in there, not that they're going to burst into flames or anything. But you could beam a gang a lot of power onto the surface of the planet and that gets rid of your power problems because they run

that runs twenty four hours a day. Like, instead of doing this AI data center thing, why not mature this technology and actually be beneficial for all kinds of things, not just data centers, but making electricity that zero emission and works twenty four hours a day. Right. We could put quite a few of these up into geostationary orbit.

of the time. So the AI data centers you're hearing about right now, they were already in development before chat GPT showed up. There's been no power plants built for AI because it takes longer than that. There's just no way to do it, really, And the real issue here is that as long as you're building data centers close into cities and trying to use city power, people don't like them. No, they don't, and that's always been true. They know it's causing more trouble than it's worth and

demanding tax breaks and things. It's just kind of stupid, like this business is average profitable or not, but these big companies have learned to manipulate local governments right right. I've heard it described as it's like they're trying to get Taylor Swift tickets, so they open ten browsers to get one pair of tickets. That they're using third party entities to lock up as much land and doing as

many orders as possible. And many of them are never going to build in the first place, but right now while they're able to get away with it because AI they're doing as many as they can.

maybe two square miles for that power plant. It's cheap maybe a billion, billion and a half dollars half a million a month a year to run. You'll get it done in three to five years. That's why you build so much natural power plants these days. They're really efficient, they're really cheap, they build really fast. They just emit a bunch of carbon now megawatt for megawatt. They emit

far less carbon than say coal. But don't consider that an achievement, right, Like you really want to have no emissions, Ideally you'd go with solar, except that solar ad best is still half load. Right we talk about about capacity. In reality, solar doesn't even get to fifty percent. It's typically around thirty percent. So the only way you solar work is you need a gigga WoT with the solar panels and a couple of gigawats with a battery, and

that's the tough part. Well, first off, a gigawatt of solar panels and maybe you'll throw some wind in there. You're talking about a hundred square miles like, it's big. It takes a lot to collect that much solar power, right, and that's expensive, Like you're going to spend five to ten billion on that, But then you need forty eight gigawatt hours of battery right so that you can run.

That's even more expensive. I extended period of time twenty billion, easy, and that'll be the largest battery pack ever made, which actually has reliability problems, so things tend to catch on fire. Yeah, so you know this is not a trivial problem to do. It's going to take longer to build, it's going to cost more to operate at that scale. It would almost

be more rational to mix solar and natural gas. So if you did a solar system, that is enough power to run at least during the day with some battery. So if you get good solar and only when you really need to are you kicking over to gas. That gets more down to reasonable size and a reasonable price point, and you can build it relatively quickly. Obviously at a gigawatt nuclear power makes perfect sense. Sure, the challenge is

it's more expensive. It's not as expensive as that straight solar and storage you're talking if you you know, we take the Voguetel reactors in new Westinghouse ones they just finished, which they spent almost twenty years and forty billion building,

and actually build them efficiently. You should be able to knock them out in maybe ten years at maybe ten billion a shot, right, and build a few of them, and those are a deal candidate in many respects, in the sense that that's a gigaut of power that likes running full ball or twenty four hours a day. They work the most efficiently like that maybe a mature small nuclear The problem is the regulatory regime and time to build. It's twice the amount of time of these other solutions.

There has been some breakthrough in geothermal, but geothermal doesn't come that big. So if you were going to do this with geothermal power, you're actually talking about building a dozen geothermal sites over maybe one hundred square miles and that's going to take a while as well, and only to be fair, because space based power is in if it is immature realistically, if you're going to do space based power. You're talking probably as much as forty billion,

maybe more for the first ghost. That's expensive and at least a decade to build it out, Like, we have to learn a bunch of things to pull that off. Operating costs would be low, and the landing site is small, so it's pretty compelling. None of these answers are easy.

The hybrid is probably the best choice. Like, I think it would make sense for municipalities to agree that if you want to build it a data center, you build enough power to power it at least most of the time, right, and maybe I'll grid tie for the rest of the time. And like any heavy industry, you negotiate for consumption of grid power. You don't get it by default any more

than an aluminum smelter gets it by default. But they sure these power problems are real, and then none of them as quickly as anybody wants, right, including AI data centers. ADI data centers are not going to build that fast and have a meaningful impact any more than Starlink has had a huge impact on the total amount of bandwidth on the Internet, right, It took building out that whole network to have widespread coverage, but it's still not a

lot of bandwidth. You know, fifteen thousand of these satellites, how many customers is that twenty thirty, forty thousand. It's not enough, right, We just can't build enough to actually make a huge impact on the business as a whole. It's great for the marketing. It certainly made the SpaceX IPO come true. He's postponed the problem for another couple

of years. He gets the mature starship. Maybe he'll switch over to space based power at some point, which really could make a difference in the world, but it will take longer to execute on. But data center wise, if you just build data centers like reasonable people do in a reasonable way, they don't have to be a problem. They can be a good member of society. Essentially, something that provides its own power, participates in the system overall,

provides some jobs, and does some compute. It's just that right now, inside of this AI bubble, everything is a bit dumb. It's a bit dumb.

is in the combined cycle natural gas. Just because it emits relatively a little carbon for the amount of power emits, capture systems costs more money, so it's going to increase the price of it. But they were already pretty cheap. They're not they're not high load systems, so they do need more tinkering and so forth to keep it running

at capacity. And you do have to do something with a carbon, right, whether you can turn it into calcium carbonate, or you can even turn it into graph Like you can turn it to costs and carbinate with a little bit of chemistry, but you need a lot of calcium. You can turn it into graphite with a whole bunch of heat. Blow the oxygen off of it, and it's just dry carbon which has got industrial applications so forth. It just takes more power. So you know, I think

governments would not have to make these decisions. They would just have to make the rules. Hey, you can do whatever power you want as long as it's zero mission. And if that means use natural gas but capture it all fine. If that means go a small module or nuclear that doesn't have those emissions in the first place, great build out solar panels with huge raise of batteries, awesome. If you got the space, go for it and the money. My shoulder takes up a lot of space. All of

it costs money. But last time I looked, all of these companies are making record profits. They can afford it, right, Yes, yes, yes, anything else? I don't know. I think I ran the gamut there.

They keep us in business. Now go write some code.