While everyone watches the rockets, SpaceX is building the pipes for the next economy—and the cash register is ringing louder than the engines.
The Summary
- SpaceX is signing deals with Google and Anthropic to rent AI compute capacity, projected to generate $26 billion annually
- The company is leveraging its satellite infrastructure to create distributed compute networks that sidestep terrestrial data center bottlenecks
- This positions SpaceX not just as a launch provider, but as critical infrastructure for the agent economy—owning the rails while AI companies fight for track space
The Signal
SpaceX is doing what every smart infrastructure company eventually does: it's renting the roads it built. The deals with Google and Anthropic represent a fundamental shift in how compute-hungry AI labs think about their stack. When you're training frontier models that cost nine figures and require exaflops of compute, you don't just need GPUs. You need them somewhere with power, cooling, and connectivity that doesn't brown out half a city.
SpaceX's edge isn't just Starlink satellites beaming internet to rural Montana. It's that they've built distributed compute infrastructure in places where land is cheap, power is abundant, and nobody's fighting over the same grid as crypto miners or legacy data centers. The Colossus data centers mentioned in the report aren't competing with AWS on latency for your Netflix stream. They're competing on raw throughput for training runs that take weeks and burn megawatts.
"The $26 billion revenue projection isn't from launching satellites—it's from being the landlord to the companies building Web4."
The Pentagon deals add another layer. When the Department of Defense starts buying compute from the same company that launches its spy satellites, you're watching dual-use infrastructure get built in real time. SpaceX isn't just a vendor. It's becoming a platform:
- Launch capacity for satellites
- Connectivity via Starlink
- Compute capacity via distributed data centers
- All vertically integrated, all controlled by one company
This is the Web4 stack taking shape. AI agents need three things to work at scale: data, compute, and connectivity. SpaceX is positioning itself to own two of those three. The implication for AI labs is clear—if you want to train the next GPT or Claude, you might need to go through Elon Musk's infrastructure. That's not a technology story. That's a power story.
The $26 billion figure also reframes SpaceX's valuation. Wall Street has been pricing it as a aerospace company with a satellite internet side hustle. But if a quarter of its revenue comes from renting compute to the companies building the agent economy, it's not a rocket company. It's an AI infrastructure play that happens to own the launch vehicles. The multiple on that is different. The strategic leverage is different.
The Implication
Watch who else signs compute deals with SpaceX in the next six months. If Anthropic and Google are in, OpenAI and Meta aren't far behind. This becomes a game theory problem: if your competitor is training on faster, cheaper infrastructure, you either match them or fall behind. SpaceX just became the arms dealer in the AI race, and the best part? They own the supply chain from orbit to power socket.
For anyone building agents or infrastructure in Web4, this is your signal that compute access is the next bottleneck. The companies that control where models get trained will shape what those models can do—and who gets to deploy them.