If Al is a foundational general-purpose technology, we should anticipate that demand for Al compute -and energy - will continue to grow, The Sun is by far the largest energy source in our solar system, andthus it warrants consideration how future Al infrastructure could most effciently tap into that power.This work explores a scalable compute system for machine learning in space, using fleets of satellitesequipped with solar arrays, inter-satellite links using free-space optics, and Google tensor processingunit (TPU) accelerator chips. To facilitate high-bandwidth, low-latency inter-satellite communicationthe satellites would be flown in close proximity. We illustrate the basic approach to formation flightvia a 81-satellite cluster of 1 km radius, and describe an approach for using high-precision ML-basedmodels to control large-scale constellations. Trillium TpUs are radiation tested, They survive a totalionizing dose equivalent to a 5 year mission life without permanent failures, and are characterized forbit-flip errors, Launch costs are a critical part of overall system cost, a learning curve analysis suggestslaunch to low-Earth orbit (LEO) may reach s$200/kg by the mid-2030s.