SpaceX's $75 Billion Is Heading to Orbit, to Build AI Data Centers in Space

The biggest IPO in history was sold as a rocket company. The money is going somewhere stranger. SpaceX raised 75 billion dollars last week, and a big chunk is earmarked for building data centers in orbit, putting AI compute on satellites circling the planet. That is the real bet behind the 2.8 trillion valuation, and it explains why the market is treating SpaceX as much like an AI infrastructure company as a launch business.

Here is the plan, in SpaceX's own words. The proceeds go three ways: expanding Starlink, growing the Falcon and Starship launch fleet, and building an orbital compute layer. That last one is the headline. SpaceX wants to fly its compute, on satellites it calls AI1, with two prototypes targeted for early 2027 and a ramp toward roughly 1 gigawatt of orbital AI capacity a year by late 2027. From there it talks about 10 gigawatts, then 100, with no dates attached.

It already owns the pieces. SpaceX bought xAI in February, folding in the Colossus data centers and the Grok models, so it has frontier AI in-house. It is building chip packaging at its Bastrop, Texas campus this year and has floated a 55 billion dollar Terafab chip plant in rural Texas. Starship is the delivery truck, each launch can carry 30 to 50 AI1 satellites at once. And the demand is showing up early, Google reportedly signed a compute deal worth around 920 million dollars a month.

The pitch for space compute is power and cooling. Data centers on Earth are running into energy limits, grid bottlenecks, and water for cooling, while orbit offers constant solar power and the cold of space as a heat sink. Put the GPUs up there and you sidestep the two things choking AI on the ground. That is the theory, anyway. It is a genuinely big idea, and SpaceX is the one company that already launches enough to attempt it.

Now the cold water. None of this is proven at scale. Orbital data centers have to survive radiation, dump enormous heat, and get serviced where no technician can reach, and the whole timeline leans on Starship flying reliably and the FCC signing off. The gigawatt targets are roadmap, not reality. So the 2.8 trillion price tag is partly paying for rockets and Starlink today, and partly paying for a space-compute future that has barely left the prototype stage.

That is the SpaceX trade in one line. The valuation is a bet that the next great data center is in orbit, and that SpaceX is the one that builds it. If it works, today's price looks early. If it slips, it looks like faith priced as fact. Either way, the launches to watch now carry servers, not just satellites.