Starcloud co-founder and CEO Philip Johnston details the development of orbital data centers designed to overcome terrestrial constraints on energy and water usage. By leveraging solar energy in space and utilizing large-scale deployable radiators to dissipate heat into the vacuum of space, these satellites provide a sustainable infrastructure for AI inference. While training large models will likely remain Earth-based for the next decade due to complex internal networking requirements, space-based nodes offer a scalable solution for the rapidly growing demand for inference compute. The project utilizes NVIDIA H100 GPUs and relies on the increasing launch capacity and reusability of vehicles like SpaceX’s Starship to make orbital deployment economically viable. This transition represents a shift toward building physical, high-tech infrastructure in space, marking an early phase in the long-term development of large-scale computational systems beyond Earth.
Sign in to continue reading, translating and more.
Continue