SpaceX Has Two AI Compute Stories; Only One Generates Revenue (davefriedman.substack.com)

SpaceX recently revealed its dual approach to AI compute infrastructure in its S-1 filing, showcasing a terrestrial data center strategy alongside an ambitious vision for orbital AI inference. On the terrestrial front, the company has committed billions to data centers, signing a significant contract worth $1.25 billion per month with Anthropic through May 2029. This existing infrastructure operates at roughly 1.0 gigawatt capacity, allowing SpaceX to capitalize on current high demand for AI compute power amidst a competitive landscape characterized by scarce resources. Conversely, SpaceX envisions a future where AI inference occurs in orbit, claiming substantial advantages in energy efficiency and cooling through solar-powered satellites. This orbital model aims to alleviate terrestrial limitations, particularly around energy supply and infrastructure delays. However, the S-1 filing lacks clarity on how the two strategies interconnect, particularly in terms of economic viability and workload migration. While the potential for lower operational costs in orbital AI is pronounced, significant technical and financial uncertainties remain—ranging from launch costs to satellite lifespan—which could impact the feasibility of this ambitious initiative. As SpaceX navigates these dual narratives, it highlights the urgent need for clarity on how these projects will ultimately intertwine and impact the broader AI/ML sector.