SpaceX acquired xAI on February 2 in a $1.25 trillion deal, the largest M&A in history. Musk's pitch: vertical integration of rockets, Starlink, and AI to build orbital data centers. The logic is seductive. Space offers free solar power, no real estate costs, and happens to need the giant rockets SpaceX already builds.
The economics don't hold up.
First, scale. Training frontier AI requires hundreds of thousands of GPUs. xAI's Colossus cluster runs 200,000. OpenAI is planning millions. Competing would mean launching hundreds of thousands of satellites, dwarfing the 15,000 currently in orbit. At that density, the risk of Kessler syndrome (cascading debris collisions that could cripple space access) stops being theoretical.
Second, upgrade cycles. When new AI hardware ships, ground data centers start deploying immediately. In orbit, you launch an entirely new satellite fleet. Every generation.
Third, the moving target problem. Google's 2024 study suggested orbital compute might be competitive by 2035 if launch costs hit $200/kg. But that assumes ground alternatives stand still. Solar panel costs have dropped steadily for decades. Every efficiency gain in terrestrial energy makes the space case harder.
So why is serious money piling in? SpaceX is targeting a $50B IPO this year. xAI burns $1B monthly and needs capital infusions to survive (Tesla and SpaceX each recently put in $2B). Long development timelines mean hype can run well ahead of physics. Investors don't need the tech to work. They need the next buyer to believe it will.
Google, Lonestar, Axiom, and Nvidia-backed Starcloud are all announcing orbital compute plays. None have solved the fundamental trade-offs. The pattern is familiar: when timelines are long and the pitch is ambitious, market excitement can sustain projects that wouldn't survive contact with a spreadsheet.
The real question is what happens when the timelines compress. SpaceX derives 80% of revenue from Starlink launches. If orbital AI pivots from moonshot to revenue dependency, the physics won't care about the valuation.