SpaceX Valuation: The AI Premium Debate

Morgan Stanley is essentially arguing that if SpaceX hits a $100 billion valuation, the market is pricing in its rocket business and Starlink, but assigning exactly zero value to its AI potential. This is a wild take because it frames SpaceX not just as a transportation company, but as a massive data and robotics play.

When you look at the hardware side, SpaceX is essentially building the most complex autonomous systems on the planet. The landing sequences for Falcon 9 aren't just pre-programmed scripts; they are real-time physics calculations handled by onboard systems that mirror the "embodied AI" trend we're seeing with humanoid robots. If we treat SpaceX's software stack as an LLM agent for physical space, the valuation gap becomes obvious.

To understand why the "AI value" is being ignored, you have to look at how the market separates "Software AI" (like OpenAI or Anthropic) from "Industrial AI." Most investors only count GPUs and tokens. But SpaceX is operating in a loop of: Data Collection → Model Training → Physical Deployment → Real-world Feedback.

If I were to model the AI value of a company like this from scratch, I'd look at the autonomous navigation telemetry. For example, the precision landing logic requires a specific type of reinforcement learning that is far more valuable than a chatbot.

Here is a conceptual look at how that "Physical AI" loop differs from a standard LLM workflow:

AI_Workflow_Comparison:
  Standard_LLM:
    input: "Text/Image Prompt"
    process: "Token Prediction"
    output: "Text/Image"
    feedback_loop: "RLHF (Human Feedback)"
  
  SpaceX_Physical_AI:
    input: "Sensor Telemetry (Inertial/Radar)"
    process: "Real-time Trajectory Optimization"
    output: "Actuator Command (Grid Fins/Thrusters)"
    feedback_loop: "Physical Outcome (Landing Success/Failure)"

The "zero AI value" claim is likely a result of the current bubble where "AI" only means "Generative AI." But for anyone doing a deep dive into AI workflows, the real frontier is the intersection of LLMs and robotics. Starlink provides the low-latency backbone for this, and the rockets provide the edge-case data.

If you are trying to build a practical tutorial for valuing AI companies, you can't just look at ARR (Annual Recurring Revenue). You have to look at the "Data Moat." SpaceX has a moat that no other AI company can touch because they own the physical infrastructure.

For those tracking the deployment of autonomous agents, the logic used to land a booster on a drone ship is the ultimate "agent" success story. It's a closed-loop system with zero margin for error. If the market starts pricing SpaceX as an AI company rather than a launch provider, that $100 billion figure will look like a bargain.

The real question is whether the "AI premium" will eventually migrate from the cloud to the stratosphere. If the market realizes that controlling the orbital data layer is the ultimate AI play, we're looking at a valuation multiplier that makes current SaaS benchmarks look tiny.