Orbital data centers are possible but brutally difficult
SpaceX is planning a massive orbital data center constellation of 1 million "AI1" satellites generating 120 GW to power frontier-class GPUs, though experts debate whether the concept is highly viable or mostly valuation hype. The technical and financial hurdles are immense, with estimated constellation costs ranging from $1.45 trillion to $9.8 trillion depending on Starship's ultimate launch capacity and cost efficiency. Key physical challenges include mitigating space radiation, which limits cutting-edge chip lifespans to about five years, and developing highly efficient, lightweight radiators to dissipate massive heat without convective cooling. Additionally, while latency from satellite-to-satellite distance would cripple tightly synchronized AI training workloads, less dependent tasks like AI inference could still run effectively. Ultimately, the feasibility of orbital data centers depends on unprecedented launch frequencies and massive capital, but no fundamental laws of physics stand in the way.