Article

Starcloud, in collaboration with Nvidia, successfully trained Google's Gemini LLM, Gemma, aboard the Starcloud One satellite in low Earth orbit using an Nvidia H100 GPU. This milestone demonstrates the viability of space-based data centers for AI, as Starcloud aims to establish a 5-gigawatt orbital data center to significantly lower energy costs. The company plans to launch infrastructure equivalent to the entire US power grid's capacity annually within the next 10 to 20 years, while addressing challenges related to radiation and orbital debris.