Are data centers in space the new frontier?

Last year, investors pumped more than $60 billion into building more data centers, which are the server farms that power artificial intelligence. There are nearly three thousand data centers being built or planned in the U.S. Among the concerns is all the energy they suck up from the communities that surround them.

One potential solution is to put these data centers in outer space.

Google is among those pursuing this, with something called Project Suncatcher. Travis Beals, the project’s senior director, spoke with “Marketplace Morning Report” host David Brancaccio. The following is an edited transcript of their conversation.

David Brancaccio: Let's go through some of the benefits of this, crunching lots of numbers in orbit. So, I'll start with one: no cloudy days, if you're using solar power from space, right? You'll have pretty good access to the sun.

Travis Beals: That's right, yes. We will be launching the satellites into what's called a dawn-dusk sun-synchronous orbit. And the great thing about that is basically the sun always shines there.

Brancaccio: We know that terrestrial data centers, the ones that we drive past, run hot. It costs money on earth to chill them out. In space that's, I suppose, different.

Beals: It is different. The way it works is you radiate the heat away, out into space. So, if you've ever stood beside, say, a campfire or a radiator or something like that, you've felt that heat coming off of it, and that heat is reaching you in the form of infrared light. And you can radiate away heat in space the same way.

Brancaccio: So, Project Suncatcher has a big mark on the calendar for, I think, early next year, really, look at that we're in 2026! Two satellites. What are you going to test?

Beals: We're going to test running the TPUs — those are Google's AI chips — in space. We're also going to test out how we cool them. And then finally, the thing that we're going to do with having two satellites, is we're going to have them talk to each other. So, we're going to send data, in the form of light, between these satellites.

Brancaccio: And so, among your challenges, right, it's not cheap to put something into orbit. It is getting cheaper, but per kilo, it's still at least $1,000 but sometimes $20,000.

Beals: We've thought through this and modeled it. We talk about $200 a kilo as being an important milestone, and we think as launch costs come down, and we optimize a lot of aspects of the system, this really could make a lot of sense.

Brancaccio: So, have you ever seen visualizations? I mean, are there renderings of what a full-scale version of a data center in orbit would look like? I mean, is it like a floating Home Depot? Is it the size of a school bus?

Beals: Right. The way that we're approaching this is to make, I'd say, fairly large satellites, but still small enough that you could fit them in a launch vehicle, in a rocket. And then the way you scale beyond that, is you just launch more satellites, and you connect them together. I mentioned we would be aiming to launch into a dawn-dusk sun-synchronous orbit, which means you're flying right over the line between day and night, and that does help reduce the potential for the satellites to be visible from the ground.

Brancaccio: So, probably hard to tell. Are we talking decades, plural, for getting a full-scale one of these up?

Beals: I hope it's not decades, but you know, maybe more like, say, one decade, we're definitely working hard to go as fast as we can.