Elon Musk just turned a rocket company into an AI infrastructure company, and the strangest part is that it actually makes a certain kind of sense.
Here’s what happened. On SpaceX’s second-quarter 2026 earnings call, the company laid out plans that have very little to do with launching satellites and quite a lot to do with computers. An exclusive partnership with Nvidia. A target of $100 billion in annualized revenue run rate by the end of 2026. A $1 trillion target by 2030. And $60 billion of revenue coming from AI alone by 2027.
Then Musk posted this on X: “I am highly confident that SpaceX will be launching Nvidia VR NLV72 AI computers in space next year.”
Not satellites carrying cameras. Not internet relays. AI computers. In orbit.
Why would anyone put a computer in space
If you’re not technical, this probably sounds like a solution in search of a problem. Computers work fine on the ground. We have buildings for them.
But the buildings are the problem. AI models — the things powering the chatbots and agents you’ve been reading about — need enormous amounts of electricity and enormous amounts of cooling. A modern data center is essentially a warehouse full of expensive chips that get very hot, sitting next to a power substation and a lot of industrial air conditioning. Cities are starting to push back on them. Power grids are straining. Land near cheap electricity is getting expensive.
Space has some interesting properties for this particular problem. Sunlight is constant and unfiltered by atmosphere, so solar panels do better. There’s no zoning board in orbit. And you’re not competing with a suburb for grid capacity.
There are also enormous problems, which I’ll get to. But the logic isn’t nonsense, which is more than you can say for a lot of AI announcements.
What this means if you’re just trying to use AI tools
Most readers here care about AI agents — software that does tasks on your behalf. So let me connect the dots, because they do connect.
Every agent you use runs on someone else’s computer. When you ask an AI assistant your inbox or draft a reply, that request travels to a data center, gets processed by chips that mostly come from Nvidia, and comes back. The reason agents are still expensive, still rate-limited, and still occasionally slow is that there isn’t enough of that computing capacity to go around.
So when a company announces it’s building a genuinely new place to put chips, that’s a story about supply. And supply is what determines whether AI agents stay a premium product or become as unremarkable as email.
That doesn’t mean orbital data centers will solve it. It means the people with the most money are now looking in unusual places, which tells you how tight things are.
The honest skepticism section
I try not to be the person who repeats press releases with enthusiasm, so let’s be fair about what’s uncertain here.
- Cooling in space is hard. There’s no air to move heat away. You have to radiate it, which is slower and needs large surface area. This is a real engineering constraint, not a detail.
- Maintenance is nearly impossible. A chip fails in a Virginia data center, someone swaps it out. A chip fails in orbit and it stays broken.
- The revenue targets are targets. $100 billion by the end of 2026 and $1 trillion by 2030 are projections a company made about itself. Treat them accordingly.
- The market wasn’t uniformly convinced. Nvidia gained after the announcement. SpaceX shares fell. Investors seem more certain about who’s selling chips than about who’s buying them.
There’s also a financing wrinkle worth understanding. Musk has been selling assets — including some Nvidia stock — and borrowing money to fund his commitments. That’s not automatically alarming; big bets get funded with use all the time. But it’s a reminder that these plans depend on capital markets cooperating for several more years.
How I’d read this
Two things are happening at once and they deserve separate scoring.
The first is a business realignment. SpaceX is positioning itself Given that Nvidia reportedly holds a stake in SpaceX, the partnership binds together the chip supplier and the delivery mechanism. That part is already real and already affecting stock prices.
The second is the orbital computing plan, which is a genuine experiment. Musk says next year. Musk’s timelines have a known relationship with calendars. I’d file this under “watch closely, don’t reorganize your assumptions yet.”
For anyone using AI tools day to day, nothing changes this month. But the question of where AI runs — and who controls that ground, or lack of ground — is quietly becoming one of the more consequential questions in tech. Now it involves rockets.
🕒 Published: