The sharpest technical objection to SpaceX's orbital AI plan is not power or heat. It is that modern AI models are designed to run across hundreds of interconnected chips... so we model how a typical 72 GPU NVDA rack would perform as a node in comparison. Forcing a frontier model into one satellite forgoes up to 44% of achievable output. The penalty is real but not permanent, better software, slimmer models, and models designed to fit the satellite could reduce it to near parity over time.

A physics-first look at how fast SpaceXAI can build frontier AI models, and what its C software rewrite actually changes. We time every stage of the training cycle, calibrated against the one run the company has shown us, then test the speedup claims, how long the edge lasts against open tools, and how the picture shifts as models grow. The model behind the analysis is attached for members.