We’ve all seen the headlines lately talking about the proposed data centers across the U.S. While nobody wants these things in their backyard, putting them in space isn’t going to solve our problems.
With just a quick look into some basic physics and logistics of putting a data center in space, the idea becomes comical. First, you have to put massive amounts of hardware, cooling systems, and power into space. Then you have to completely reengineer the data center for space. Then you have to think about maintenance. Oh, and cooling these things down in a vacuum is no easy task.
To put it simply: no, we won’t be putting any large AI data centers in orbit.
Hey, everybody. Peter Zeihan here, coming to you from Colorado. Today we are taking a question from the Patreon crowd. So it’s a question that I have resisted answering because I don’t like it, but so many people have asked, I figure, what the hell? So it’s on space based data centers. Is that the future? Will this solve some problems?
How much is it? It’s just a little bitty snake. Is this something we should take seriously? Let’s explain what a data center is. You have dozens, hundreds, thousands of racks that contain modules that are packed with things like graphics processing units, Dram and Rand memory, and then threading through it all, some sort of coolant.
You see, most data centers are based on crunching data, the GPU part, and that generates a lot of heat. So you incorporate the cooling system into the modules and if you can, into the gypsum cells, although that’s really cutting edge and no one’s really doing that at scale just yet. Soon, hopefully. Anyway, those racks have weight, the GPUs have weight, the Nand has weight, the Dram has weight. You need a lot of electricity. You need a lot of coolant.
You need a liquid system. You need a pumping system. And if you’re going to take a one gigawatt system and move it into space, the theory is that because space is cold, cooling is cheap, and because solar power in space generates at least five times what it does on the surface, power is cheap. No, no. Let’s start with getting this stuff up there.
So just the computational stuff, the racks and the things in them, not counting the cooling system. That’s for something of this size, roughly 20 million pounds of weight. Then you need another 10 million pounds of weight for the liquid coolant. Then you need another 5 to 8 million pounds of weight for all of the assorted things that keep it alive the pumps, the power cables and such.
Okay, that by itself is about 250. Starship launches. Starship is the vessel that Elon Musk and SpaceX hope is going to revolutionary space travel. It’s currently in the prototype stage. It has not been going particularly great, but let’s just hand wave that away and say that it works 250 of those to get the equipment up. You’re not done.
Now you need a construction facility in orbit, which means you need a habitat in orbit. You also need a habitat at the data center, because a typical data center has 100 people working at any given time. Now, the ES International Space Station has never had more than 13 people in at a time. And really it’s only rated for seven.
So you’re talking about the equivalent of 20 to 40 ES equivalents attached to this data center in order to carry out the work. And that assumes that the work is possible, which are probably isn’t GPU’s really. All the high end chips are not rated for vibration, for cold, for heat, for humidity, and certainly not for radiation. And you’re in space so you don’t have the planet’s magnetic field and atmosphere protecting you.
So now you need to clad the whole thing in radiation proofing, which is at least another 20 million pounds. There’s another 200 Starship trips. Also, a data center is a living thing. Chips burn out, you have to remove them and replace them. So now you need a space based launch system just to maintain your one data center. Oh, and the biggest thing heat.
Yes, space is cold. But what people forget is that space is a vacuum, and a vacuum is the most effective insulator we have ever discovered. So yes, you can use radiator fins, but their area of effect has to be an order of magnitude bigger than what they would be on Earth, and a very different configuration because they need to extend significant distance.
You’re talking about literally tens if not hundreds of kilometers square kilometers of radiator surface in order to radiate the heat. Yeah, it’s cold out there, but it’s also insulating. So you’re now talking about a system that needs a pumping system to get the liquid out to the radiator fins, and you are in zero g. So you don’t only need more pumps to cover more territory.
Pumps don’t work as well in space because you don’t have gravity doing some of the work. So now we’re basically conservatively up to 500 space a star. 500 star. What’s it called? Starship. Yeah, 500 Starship launches and a note on the radiators. This sort of design, we don’t build on Earth because, you know, we’ve got gravity. So not only are you talking about having to invent a fundamentally new system here for the cooling system, but just the cost of the radiators alone is probably about the same as the GDP of New Zealand, about $250 billion, and that is by itself 20 times the cost of a one gigawatt facility all in on Earth.
There’s so many levels of dumb to this that I forgot. One that’s really kind of critical. The radiators. Not only are there big and expensive, but remember spaces and insulator. So if you’re radiator fins are flat on to sunlight, they’ll actually absorb heat rather than get rid of heat. So you have to make sure they’re always point in to the sun, which means that your space based data center has to do something that no Earth based data center has to do.
It has to be able to move. So you got to put engines in maneuvering thrusters on this thing, complete with the fuel. So yeah, so much dumb. So anyone who tells you that data centers in space are coming because we don’t have room is lying to you. Anyone who is telling you it’s the only way that AI is going to work is lying to you. And anyone who’s trying to raise funds off of a story like this is just thinks you’re stupid and is going to get your money.
Don’t fall for it. Does this mean there can’t be computing in space? Of course not. You can have small, small emphasis on small like a rack systems attached to some sort of orbital platform. But all of these rules apply. Massive radiator fins, atypical cooling system, huge cost to install. And unless you’re going to permanently station people up there when something fails, it’s broken for good.
So no, this is not revolutionary. But if the rules of physics change, I reserve the right to change my mind.
