Elon Musk explained his merging of SpaceX and xAI ahead of a mega-IPO this year as a step in eventually building data centers in space.
For investors in companies like Axiom Space and Lonestar, Musk’s plans are validating for the space. But for astrophysicists like Jonathan McDowell, who worked at the Harvard-Smithsonian Center for Astrophysics’ Chandra X-ray Center for 37 years, Elon’s plans raise many troubling questions.
PitchBook spoke with McDowell, who
led the science data systems team for the NASA Chandra X-ray Observatory, working there from 1988 to 2025, about Musk’s ambitions. The conversation has been edited for length and clarity.
PitchBook: At the face of it, there seems to be compelling arguments in favor of orbital data centers. Cheaper power, no need for cooling...
Jonathan McDowell: Well, not no need for cooling. Perhaps simpler cooling. These things still generate heat, and we’ve got to get rid of it somehow.
In space, you can just do it with radiators. So that’s a big advantage, but still, radiators aren’t free. They add mass. And the bigger the satellite, the more radiators you need.
But that technology is ready to go, yes?
Yes, it’s mature. Maybe not on the scale we’re talking about, but yes.
So if we’re talking about the bulk of the centers needed for the AI buildout, that’s pretty sizable. How would that transform the landscape of Earth orbit?
In the last 10 years, we’ve gone from about 1,000 active satellites to about 14,000 active satellites. But with the SpaceX proposal and others, we’re talking about another factor-of-100 increase. Two orders of magnitude. That’s an enormous increase in this density of satellites in Earth orbit.
Space is big; there’s plenty of room. But in low Earth orbit, you’re traveling at 17,000 mph. So you have all of these spacecraft whizzing around in three dimensions, at these insane speeds. And so the risk of them hitting each other in the absence of careful management is very significant.
And that’s the Kessler problem?
The Kessler problem is: Two satellites hit each other, that generates debris; the debris hits other satellites, that creates more debris; and then you get a chain reaction.
We’re starting to already see an increase in minor collisions. There’s enough satellites in orbital shells—it’s enough to worry you that you really need to start removing some of this junk.
SpaceX will tell you they can manage these very large constellations by station-keeping, where each satellite is targeting a slightly different position. But now, China just filed for constellations totaling over half a million satellites. So now you’re looking at two million satellites in the same region of space [and] that have a history of not talking very well to each other.
Could that debris end up falling back to Earth?
With Starlink [satellites], they designed them so they pretty much melt entirely, so you don’t have to worry too much about stuff falling on your head, although that is a concern with rocket launches if a rocket fails. So there are risks to the public from that.
And what about environmental risks?
If they follow the Starlink model, which is to replace each satellite every five years—if they’re going to dispose of them in the upper atmosphere, that does raise concerns about what the atmospheric chemistry is going to be, using the upper atmosphere as an incinerator.
The general feeling is, at 10,000 satellites, we might be okay with the ozone layer, but at a million satellites, we may not be okay. And then there’s whether you count the change to the night sky as an environmental issue—I do. And then there may be things we haven’t even thought of yet.
VCs like to envision the future in a science fiction-y way. But when we look at today’s reality, what is really possible here?
The investment that’s going to have to go into satellite-building factories—let alone launch capacity—is vast.
And these satellites use propulsion, so they’re going to need a lot of propellant. SpaceX recently switched to using argon instead of xenon as the propellant because Starlink’s needs exceed the total annual production of xenon. I’m guessing the amount of argon you’re going to need for a million satellites is going to exceed the amount of argon we’re making these days.
There are entire subsidiary industries that are going to have to grow up to support the logistics of all of this.
Plus, they’d need to send all of this into orbit in the first place.
The claim is that with later versions of [SpaceX’s launch vehicle] Starship, when they finally get it working, is that we’ll be able to take 100 tons to orbit.
Now, that’s to a low orbit. Let’s pretend they can. If each satellite is two tons, that’s 50 satellites a launch, 20,000 starship launches total. With a five-year replacement cycle, that’s 4,000 starship launches a year. So that’s 10 starship launches a day, every day, forever. And that’s just for this one project.
Now, [SpaceX’s] Falcon 9 is doing a launch every few days, so it’s not completely inconceivable, but these are much bigger rockets than Falcon 9s.
That’s a huge ramp-up.
I don’t want to say they can’t do it. It’s amazing what they’ve done with Falcon 9, and they’ve got a really good safety record. So maybe they can get there, but it’s incredibly ambitious.
Meanwhile, they’ve got a commitment to NASA to support the Artemis program. One might worry that they’re not going to have a lot of spare attention to give to this orbital data center project in the near term.
Recent News
Why Wall Street wants to trade AI compute like oil
Conditions have been kind to PE returns. Why haven’t portfolios delivered?
I suppose that’s part of the reason they’re going public, for the access to capital.
That may be true, but then you worry about the workforce. Lots of extra money, great. But are there enough people with the appropriate skills? That could be a pacing [challenge], as you might have to wait to train a bunch of people.
Two or three years isn’t looking likely, then?
I can fairly definitely say no.
Maybe for a prototype satellite, but for deploying this system, I can’t see that happening sooner than a decade, personally. Space [exploration] is incredibly capital-intensive. So the question you have to ask is: On that timescale, are there other technical advances that could happen to make Earth-based data centers more efficient? It’s very much an open question whether this is the right move.
Learn about our editorial standards.
