Toward a future space-based, highly scalable AI infrastructure system design

The sun is the largest energy source in our solar system, and thus it warrants consideration how future AI infrastructure could most efficiently tap into that power. We explore a scalable compute system for machine learning (ML) in space: fleets of satellites equipped with solar arrays, free-space optics inter-satellite links, and Google tensor processing unit (TPU) accelerator chips. To facilitate high-bandwidth, low-latency inter-satellite communication, the satellites would be flown in close proximity. We illustrate the basic approach to formation flight via an 81-satellite cluster of 1 km radius and describe an approach for high-precision ML-enhanced models to control large-scale constellations. Trillium TPUs are radiation tested. They survive a total ionizing dose equivalent to a 5 year mission life without permanent failures and are characterized for bit-flip errors. Launch is critical to overall system cost; a learning curve analysis suggests launch to low-Earth orbit (LEO) may reach ≤$200/kg by mid-2030s.