Elon Musk Pitched AI in Space. Wall Street Is Betting on AI on the Ground.

 SpaceX has spent the last year building a genuine AI narrative around orbital compute, data centers floating above the atmosphere, powered by endless solar energy. It's a compelling story. According to Wall Street analysts, it's also not where SpaceX is actually making its money right now, or where it will be for a long time.

The real business is refreshingly unglamorous: terrestrial data centers, sitting firmly on Earth.

SpaceX is already monetizing this through its Colossus supercomputer clusters, striking compute deals with enterprise customers including Anthropic, Alphabet's Google, and Reflection AI. Those contracts alone are expected to generate more than $28 billion in annual revenue. For context, that dwarfs SpaceX's entire 2025 AI revenue of roughly $3.2 billion, and it individually surpasses revenue from both SpaceX's rocket launch business and its Starlink connectivity business.

The spending backs up the story. SpaceX poured nearly $18 billion into AI infrastructure and development in 2025 alone, split between about $12.7 billion in capital expenditure and $5.1 billion in research and development. That's more than the company spent on its actual space and connectivity businesses combined. Colossus and Colossus II together already provide roughly one gigawatt of AI compute, putting SpaceX among the largest AI compute operators in the world. J.P. Morgan expects that terrestrial footprint to scale up to about 9 gigawatts by 2029, roughly four times the output of the Hoover Dam.

Here's the reality check on the orbital vision specifically. Anthony Milovantsev, a partner at consultancy Altman Solon, put it bluntly: the idea that orbital compute will meaningfully disrupt terrestrial data centers anytime soon is overblown, with any real displacement likely a decade or more away. J.P. Morgan's own read lines up with that timeline, expecting SpaceX to keep operating and expanding its terrestrial compute clusters through 2029, only pivoting toward orbital compute for incremental capacity after that point, not as a replacement.

The orbital bet isn't dead, it's just gated behind hard engineering milestones that haven't been solved yet: Starship achieving rapid, reliable reusability, meaningfully lower launch costs, and further advances in satellite engineering. If those pieces fall into place, solar-powered computing satellites could genuinely sidestep some of the costliest parts of running data centers on Earth, namely energy, cooling, and land. But "if" is doing a lot of work in that sentence.

Worth noting too: analysts caution that SpaceX's headline compute contracts include termination provisions, so treating that $28 billion figure as locked-in, guaranteed recurring revenue would be a mistake.

The framing analysts are converging on is a useful one. This isn't really a debate anymore about whether SpaceX can build and monetize serious AI infrastructure, it clearly can, and the deal sizes prove it. The real question is how quickly, if ever, that capability actually extends beyond Earth's atmosphere into something commercially viable.

Which is a more interesting story than "SpaceX is going to put data centers in space." The more accurate version is: SpaceX already built a massive AI infrastructure business, and it just happens to still be firmly planted on the ground.

Does the orbital data center vision still excite you as a long-term bet, or does this kind of analyst reality-check change how you read those announcements?