Data centers in space
10:32 Share

Data centers in space. Yes, really. From American Public Media, this is Marketplace Tech. I'm Novosafo. Every time AI generates a graphic or answers a query, a data center somewhere is involved. That means high-powered chips are involved, electricity is involved, as well as large cooling systems. The more AI utilization grows, the bigger the data centers.

In one corner of the startup world, there's an effort to solve the resource dilemma we're facing by putting data centers in space where there's lots of room, solar power, and a cold environment. Evan Robinson Johnson wrote about this in The Information.

If the audience is familiar with traditional data centers, sprawling campuses, large rectangular buildings full of tightly packed AI processors for the latest AI models, it's basically taking some of those same fundamentals and scaling it down drastically to something that is more the size of a refrigerator and has additional protections for space and has large solar panels to power them.

And are there startups out there that are about to make this happen? There are. So reporting focused on two, StarCloud out of Washington and Axiom Space, both of which are looking at initial designs and actually trying to send basically prototype satellites up into space earlier and by the end of this year.

And there are some big names involved in this effort as well, aren't there? There are. There's been expressed interest from Eric Schmidt, who was the former CEO of Google and has expressed interest through his new rocket company.

And Jeff Bezos, who has obviously ties to Amazon and their AI computing resources, as well as his rocket company, Blue Origin, has also expressed interest in moving several heavy industries into orbit. And AI data centers could be part of what he plans to build out with that infrastructure.

Now, why would these big guys be so keen on figuring this out, getting this all going?

I think we've already seen in a rapid number of years the growth of AI data centers in a way that didn't exist for traditional Cloud computing. The scale and magnitude has prompted conversations about whether the electrical grid can support these centers and just what it will take to actually build out that infrastructure on Earth. There are a couple of benefits to having these centers in orbit in addition to non-stop power.

non-stop power because they draw from the sun they'd have giant solar arrays right right but as we showed in our reporting kind of each each of these benefits that you might um get from going to space also comes with added challenge and technical hurdles which is sort of why we haven't seen it done at scale so far you said non-stop power what are some of the other benefits

Along the way, there's the idea that you can send information between satellites and down to Earth. And there are actually an increasing number of satellites that are taking pictures of Earth right now. It's kind of a new market where...

handful of companies and governments are making a lot of data in space and don't have efficient ways of sending it all down. So in the meantime, while we're also trying to do more computing for all of the AI functions that people are familiar with on the ground, having a center in orbit might actually benefit that industry and

I think there's a sense once you start talking to these people in space industry that there's going to be all sorts of additional research and development and business and commercial endeavors in orbit in the next few years. And all of that's going to need some of the same computing that we've seen here on Earth.

What about the financials? How do they pencil out? Is it more expensive, cheaper to do this all in space versus on the ground? It could be, but all of that hinges on having a cheaper ride to space. And a lot of that is still in the works as

much as we've seen out of SpaceX with a rapidly reusable Falcon 9 rocket, many of the people who are hoping to send more into space are waiting for Starship. It's a much larger rocket to come online and a handful of other competitors that might make it easier to send just more mass to orbit. Because as I started by saying, like the centers that we've seen on Earth are obviously quite large and there's some...

tricky and interesting scaling things you can do in space. But in order to actually make that cost effective, you need a much cheaper way to get things to orbit. And that still doesn't exist today. And cheaper by how much, potentially? Like, what are we talking about? What makes it cheaper to operate these data centers in space versus on the ground? Because, you know, logically, you might say, well, just going in space is so expensive. How could it possibly be cheaper?

Well, the biggest cost, at least that the one startup we talked to penciled out, was that you don't have to pay for the permitting and the power. Those costs can be pretty significant over, you know, say a decade of operating a data center. So even if the initial cost to build and construct and send the elements to space is costless,

comparable. If you think of like over several years, you're not having to pay for the land and you're not having to pay for the power in the same way. Their estimate for a 40 megawatt data center was about 100 million cheaper over the course of a decade. We'll be right back.

You're listening to Marketplace Tech. I'm Novosafo. We're back with Evan Robinson-Johnson, business reporter at The Information, who wrote about efforts underway to figure out how to launch data centers in space.

There are still a number of scientific hurdles and challenges here as well, right? Because one of the ones you wrote about that I was surprised, frankly, to read was that even the idea of cooling these enormous processing computers that would be sent up into space, even the idea of cooling them down is tricky in space where it's very cold. Why would it be hard to cool them in space? And how does that give us...

a taste of just how complex the scientific endeavor here is.

You're absolutely correct that like each piece of the technology presents a new hurdle. So in space, it's cold, but the actual technology that helps you eliminate heat that the AI chips develop is a radiator. And they're basically like these large black panels that you put on the backside of the solar panels. So they also have to be quite large. The idea roughly is that they would scale with the solar panels and there are some efficiencies you can take advantage of.

But the basic point being that there's no air in space, so you can't fan away the heat. We've already seen more sophisticated cooling techniques at the AI data centers that exist here. And you can take...

those techniques essentially and pair them with a radiator. It's just never been done at any significant scale. And so the idea that you could basically take the principles that you can do in a small demonstration satellite and scale up to a gigawatt data center seems like a pretty lofty goal to the experts that we've spoken to.

And these potentially, eventually, if these are done at scale, they would be very, very large structures. I mean, we've already had people talk about space debris, pollution, visual pollution, with scientists on the ground complaining about being able to observe the sky. How does all of that factor into what they're trying to do here? That is going to be a critical part of the conversation about any industry that we want to send to space. I mean,

Some of the acknowledgements from the companies that are looking to do this are that you have to have sophisticated monitoring. You have to have a means of propulsion to, uh,

negotiate the satellite around everything else in orbit. There is some flexibility in terms of all of these objects aren't orbiting earth in the same, same distance. So there are ways, you know, you can be strategic about, uh, helping to avoid collisions. And there are also similar ways that you can go about that by making them not so bright so that, uh,

people's initial frustrations with, especially in the scientific community with kind of brighter satellites in the night sky interrupting telescopic vision and things. There are kind of ways that you can try to make the satellites less visible. But all of that does become part of the conversation. And I think the question is just the trade-off about

The difficulties on building large scale infrastructure for these centers on earth, too. It's not like, you know, the more traditional data centers aren't without their own visual problems and resource problems. And certainly we've seen conversations about those centers in a similar way.

If these are to happen, and I understand it's a big if because of all the issues involved here and all the questions and scientific hurdles that have to be overcome, but if these were to happen, when would we see them?

Well, we'll see the first demonstrations, you know, if you take these companies that word by the end of this year. But to see something at the scale and magnitude that would actually be a meaningful shift in compute, it seems like that is still at the very least a decade away. And, um,

There will be kind of significant questions for any companies that's looking to scale about how they form the partnerships and prove that the concepts that they demonstrate from their initial satellites will be able to scale. That was Evan Robinson Johnson at The Information. By the way, the biggest data centers in construction today here on Earth will each consume enough energy to power 2 million households.

Daniel Shim produced this episode. I'm Novosafo, and that's Marketplace Tech. This is 8 p.m. This Old House has been America's most trusted source for all things DIY and home improvement for decades. And now we're on the radio and on demand. I think you're breaking into this wall regardless. I was hoping you wouldn't say that. I need to go and get some whiskey, I think. I would get the whiskey for sure. Subscribe to This Old House Radio Hour from LAist Studios, wherever you get your podcasts.