Elon Musk just proved you can lose a billion dollars selling AI compute and still beat the street.
The Summary
- SpaceX's AI revenue tripled to $2.6 billion year-over-year, driven by compute deals with Anthropic and Google, positioning it as a neocloud competitor to CoreWeave
- Total revenue hit $7.8 billion, topping Wall Street's $6.81 billion estimate in its first earnings report since going public in June
- The AI division lost $1.26 billion this quarter, slightly better than last year, even as the company claims AI is "the source of most of its value"
- The company is now simultaneously building rockets, running satellites, and operating as infrastructure for the agent economy
The Signal
SpaceX isn't a space company anymore. It's a vertically integrated compute provider that happens to own the delivery mechanism. The AI business tripled revenue to $2.6 billion in a single year by signing deals with Anthropic in May and Google in June. That puts Musk's company in direct competition with neoclouds like CoreWeave, except SpaceX owns the satellites, the launch vehicles, and increasingly, the ground stations.
The economics are brutal right now. A $1.26 billion operating loss on $2.6 billion in AI revenue means the company is burning roughly 48 cents on every dollar it brings in from compute. But the loss narrowed from the same quarter last year, and Wall Street doesn't seem to care. The company beat revenue estimates by nearly $1 billion, and investors are treating it as a compelling buy.
"SpaceX's AI division is the source of most of its value, according to the company's IPO documents."
Here's what makes this different from every other money-losing AI infrastructure play:
- SpaceX controls latency in ways terrestrial clouds can't, especially for edge compute and real-time agent coordination
- The satellite network creates a moat that's expensive to replicate and harder to regulate
- Starlink subscribers are built-in distribution for consumer-facing AI services
The Anthropic and Google deals signal something bigger than revenue. Large AI labs are hedging against AWS, Azure, and GCP by diversifying into satellite-based compute. That's not about cost. It's about resilience, latency for agent-to-agent communication, and avoiding the hyperscaler oligopoly. When Anthropic signs with SpaceX, it's saying the future of AI workloads isn't all in Northern Virginia data centers.
The Implication
Watch where the next wave of AI infrastructure spend goes. If SpaceX can narrow losses while tripling revenue, other neocloud providers will follow the playbook: own the physical layer, sell to the labs directly, and position as the anti-hyperscaler alternative. For AI companies building agents that need to coordinate globally with sub-50ms latency, SpaceX isn't optional anymore.
For investors, the $1.26 billion loss is the cost of building a monopoly. Musk is spending to own the only compute infrastructure that doesn't rely on terrestrial fiber, doesn't bow to a handful of cloud providers, and can scale with orbital physics instead of real estate. If the AI division actually is "most of the company's value," then the rocket business is just the world's most expensive logistics operation.