There is a constraint nobody in the AI boom is talking about loudly enough: Earth is running out of room. Land near reliable power grids, water rights for cooling, and grid interconnection slots that can take three to five years to obtain — these are the actual bottlenecks slowing AI infrastructure growth, not chips or capital. Bloomberg estimates nearly half of all U.S. AI data center projects will be delayed this year due to power constraints alone.
The solution being assembled right now bypasses all of it. Orbital compute — AI data centers launched into low Earth orbit, powered by constant solar energy, cooled by the vacuum of space, and beaming results back via laser link — is no longer science fiction. In November 2025, Starcloud launched a satellite carrying an Nvidia H100 GPU into orbit. In December, that satellite trained a language model on Shakespeare — the first time a neural network was trained in space. In March 2026, Nvidia CEO Jensen Huang unveiled the Vera Rubin Space-1 module at GTC, a chip platform purpose-built for orbital data centers, announcing that space computing had arrived.
Then in April 2026, SpaceX filed confidentially with the SEC for an IPO targeting a valuation of 1.75 trillion dollars — the largest in market history — with orbital compute as the central thesis. The company already filed with the FCC to launch up to 1 million orbital compute satellites. The SpaceX-xAI merger, valued at 1.25 trillion dollars, was completed in February explicitly to pursue this opportunity.
The economics are not there yet. Running one H100-equivalent GPU for one hour costs roughly one dollar in a terrestrial data center. In orbit today, that same compute costs around 142 dollars — about 60% of which is pure launch cost. The entire thesis rests on one number falling: dollars per kilogram to low Earth orbit. SpaceX Starship is designed to drop that from 3,000 per kilogram today to 50-100 at scale. The crossover point, where orbital compute becomes cheaper than terrestrial, is estimated around 2038.
But early adopters do not need to wait for 2038. Defense agencies, sovereign compute programs, and power-constrained regions in Europe and Asia all have compelling reasons to move now, even at premium prices. The companies building the infrastructure for this shift — launch providers, satellite chip designers, orbital systems — are in their early innings. The constraint driving this trade is real, the timeline is compressing, and the SpaceX IPO may be the starting gun.
