why jeff bezos wants aws to move from the cloud into orbit
Are we going to have new AWS regions orbiting overhead by 2050? Physics says no.
Jeff Bezos is best known from his viral appearances in Epic Rap Battles Of History, a shout out in a smash hit song by Bo Burnham, and… oh, right, having made global logistics and digital infrastructure his personal playthings. Having retired from active day to day work at Amazon, he’s been doing a lot of public brainstorming, including on the future of computing and AWS, which powers most of the apps on your phone and large websites you visit.
And one of his grand visions? Data centers. But not your typical data centers. No, no, nothing that boring in pedestrian. The data centers he’s dreaming of are in space and will be floating over our heads in no more than 20 years, consuming a terawatt worth of energy from the sun each.
This sounds like it makes sense. Space is really cold. Computers get hot by forcing something like a septillion electrons to run through tens of billions of transistors per chip. Naturally, if you put the hot thing in the cold thing, add some solar panels, and you’re golden, right? After all, the sun puts out about 1,300 watts per square meter across Earth’s orbit, and will continue to do so for over a billion years.
Except that’s not how physics works. One of the counterintuitive things about space is that spacecraft tend to have more trouble cooling down than heating up because of what heat is. Heat is the energy of rapidly moving particles, and that motion has to be transferred to other particles for the heat to radiate out, as it desperately wants to do thanks to a little something called entropy.
On Earth, we have air to provide particles to absorb and transfer that energy, letting hot things cool off. Space is a vacuum, so there are virtually no particles to which to transfer energy, which means that the hot thing just stays hot. And if that hot thing is continuing to do whatever made it hot in the first place, it will just keep getting hot to the point where it melts and becomes a ball of heat floating through the cosmos.
how to build a data center in orbit
Then there’s the logistics. For a heavy duty data center with about 100,000 servers, a few Falcon Heavy launches at $100 million each could be enough to get all the server racks into orbit, plus another launch or two to make sure the new station itself is fully assembled, probably inside a self-inflating object very similar to designs by Bigelow Aerospace. So far, so possible. Just very expensive.
Day to day maintenance would need to be done remotely by robots, the power would need to come from enormous solar panel arrays which produce a million times more electricity than the International Space Station. Just a little back of the envelope math shows that we’re looking at an array at least two square kilometers large to meet the stated goal. Which is basically the Principality of Monaco proper, but it orbit.
At this point, your center isn’t even up and running yet, but you already spent at least two orders of magnitude more than a similar data center would cost to build on Earth, plus had to design a state of the art cooling system that absolutely cannot fail since you can’t just open a window and get some fans going while you repair it, and simply shutting down the servers won’t do much because, again, physics.
Maybe filling the space with chilly nitrogen to absorb some waste heat until your rapid response team gets there and makes the necessary repairs, meaning that you need a crew of astronauts and rockets on standby, ready to launch and handle whatever any of the robots doing the day to day maintenance and repairs can’t manage themselves or with remote control from a center on Earth, staffed around the clock.
But suppose you do have a few billion burning a hole in your pockets and you want to send astronauts into orbit to perform repairs on a semi-regular basis at similar costs, and none of that was a showstopper. Your data center is now up and running. You just need to connect it to the internet. For which you’ll need to create and launch the kind of satellite swarm that can support terabit speeds, almost 2,000 times faster than the current setup ran by NASA. Which is an endeavor has its own major issues.
how to overpromise and never deliver
Now, I don’t want to say that we’ll never have data centers in space, or have no use for them. If the goals of the Techno-Optimist’s Manifesto and Bezos’ dream of a trillion of us swarming the solar system for the glory of Amazon do come to pass, we will need quite a few of those. But by that time, we’ll most likely have things like photonic CPUs which don’t generate nearly as much waste heat, and quantum computers that need to be cool to operate properly in the first place.
The notion that in 20 years, we’ll have orbital data centers when we can build them on Earth and power them with clean energy for three orders of magnitude less while also taking full advantage of existing infrastructure and with a lot less risk like creating vast clouds of dangerous space junk should an accident happen, or a rocket blowing up in the middle of a crewed mission, seems absurd. So why would Bezos even say it?
Because it sounds cool. Because it worked for Elon Musk. Make grand predictions of sci-fi technologies people would love to see, if only for the cool factor, give yourself a very long timeline in which to repeat the claim to make it sound like you really know the future and are actively creating it, enjoy the fawning press coverage, deliver a sad shadow of what you predicted — if anything at all — and move on to the next thing as you feign amnesia about the thing you previously predicted.
And this is the biggest problem with the tech industry today. Those with the loudest voices, most access to funding, and making most of the funding decisions, no longer care what actual users and customers want and are busy shaving imaginary yaks. It’s no longer about building things people need, but becoming digital gods and absurd exercises in self-indulgence and self-aggrandizement.
Bezos doesn’t care that we don’t need data centers in space anytime soon, that when we do, they’re going to be on board spacecraft and use a infrastructure that needs to be designed from scratch and built over decades, and that we’ll need these servers to use post-silicon CPUs and quantum chips. He cares that it sounds hella awesome and will get him a headline, that he has a lot of money, and that no one has told him no so far. What you actually want or need from technology? That’s irrelevant when there’s a solar system wide empire to build and immortality in the cloud to chase.