On Friday, when SpaceX filed plans with the Federal Communications Commission (FCC) for a million-satellite data center network, you might have thought Elon Musk was having a bit of fun with us. But a week later, it is clear that he is dead serious.
The most obvious step, of course, is the formal merger between SpaceX and xAI that went forward on Monday, officially drawing together Musk’s space and AI ventures in a way that makes a lot more sense if there’s some kind of joint infrastructure project planned.
But even beyond the merger, we’re starting to see the idea of orbital AI data clusters — essentially, networks of computers operating in space — cohere into an actual plan. On Wednesday, the FCC accepted the filing and set a schedule seeking public comment. It’s a pro forma step normally, but FCC chairman Brendan Carr took the unusual step of sharing the filing on X. Throughout his tenure as chairman, Carr has shown himself eager to help Trump’s friends and punish his enemies — so as long as Musk stays on Trump’s good side, the proposal is likely to sail through without issue.
At the same time, Elon Musk has started to flesh out the argument for orbital data centers in public. On a new episode of Stripe co-founder Patrick Collison’s podcast “Cheeky Pint,” which also featured guest Dwarkesh Patel, Musk laid out the basic case for moving most of our AI computing power into space. Essentially, solar panels produce more power in space, so you can cut down on one of the main operating expenses for data centers.
“It’s harder to scale on the ground than it is to scale in space,” Musk said in the podcast. “Any given solar panel is going to give you about five times more power in space than on the ground, so it’s actually much cheaper to do in space.”
Close listeners will note that there is a bit of a gap in the logic here! It’s true that solar panels produce more power in space, but since power isn’t the only cost in operating a data center and solar panels aren’t the only way to power a data center, it doesn’t follow that it’s cheaper to do the whole thing in orbit, as Patel noted in the podcast. Patel also raised concerns about servicing GPUs that fail during AI model training, but you’ll have to listen to the full episode for that.
Overall, Musk was undeterred, marking 2028 as a tipping point year for orbital data centers. “You can mark my words, in 36 months but probably closer to 30 months, the most economically compelling place to put AI will be space,” Musk said.
He didn’t stop there. “Five years from now, my prediction is we will launch and be operating every year more AI in space than the cumulative total on Earth,” Musk continued.
For context, as of 2030, global data center capacity will be an estimated 200 GW, which is roughly a trillion dollars’ worth of infrastructure when you’re just putting it on the ground.
Of course, SpaceX makes its money by launching things into orbit, so all this is pretty convenient for Musk — particularly now that SpaceX has an AI company attached to it. And with the new SpaceX-xAI conglomerate headed for an IPO in just a few months, you can expect to hear a lot more about orbital data centers in the months ahead. With tech companies still pouring hundreds of billions of dollars into data center spending each year, there’s a real chance that not all the money will remain earthbound.
