SpaceX is not really pitching Wall Street a rocket company anymore. It is pitching the first vertically integrated AI infrastructure company with access to orbit.
That is the real story behind this IPO.
The headline numbers are already historic: a $75B raise, a $135 share price, a roughly $1.75T valuation, and about $150B in institutional demand. The largest IPO ever, about 2x oversubscribed.
But the more important shift is buried underneath the bookbuild.
SpaceX is no longer just selling launches, Starlink subscriptions, or Mars optionality. It is now selling compute.
Google is reportedly paying SpaceX $920M per month from October 2026 through June 2029 for access to around 110,000 Nvidia GPUs and related infrastructure.
Anthropic is reportedly paying another $1.25B per month for compute capacity.
Together, those two contracts alone imply roughly $26B in annualized AI compute revenue - more than SpaceX’s reported 2025 revenue.
That is why the market suddenly has to reprice the company.
The bull case is no longer “rockets are cool.”
It is: SpaceX controls the launch layer, the satellite layer, the communications layer, the compute layer, and potentially the orbital data-center layer.
Reuters now reports that SpaceX is aiming to begin orbital AI compute demonstrations by late 2027, ahead of the “as early as 2028” deployment timeline disclosed in its IPO materials.
The ambition is staggering: space-based AI computing infrastructure, with Starship as the cost-reduction engine that makes mass deployment possible.
In other words, SpaceX is trying to turn orbit into the next data-center geography.
Not Virginia.
Not Arizona.
Not Abu Dhabi.
Orbit.
That is why this IPO matters beyond finance. It is a market event, but also an infrastructure event. If SpaceX can make orbital compute economically viable, it changes the geography of AI power.
Power constraints, cooling constraints, land constraints, sovereignty questions, and latency trade-offs all enter a new regime.
The bear case is obvious too.
Starship is still the bottleneck. Orbital compute at scale is not proven. A $1.75T valuation assumes execution across rockets, satellites, AI infrastructure, and public-market discipline. That is a lot to believe at once.
And 2x oversubscription is solid for the biggest IPO in history, but not euphoric by normal hot-IPO standards.
Still, the market is no longer valuing SpaceX as a launch company.
It is valuing it as the company trying to fuse rockets, satellites, GPUs, cloud contracts, and orbital infrastructure into one stack.
The question is not whether SpaceX is expensive.
Of course it is.
The question is whether Wall Street is watching the first AI infrastructure company whose data centers may eventually leave the planet.