Why Thermodynamics Rules Future Orbital Data Centers (spectrum.ieee.org)

The emergence of orbital data centers, propelled by companies like SpaceX and Google, is transforming the landscape of computing in space. SpaceX is planning a constellation of data centers through its acquisition of xAI, while Google’s Project Suncatcher aims to deploy satellites with AI chips by 2027. However, the transition from terrestrial to space-based data centers reveals significant challenges, particularly concerning heat dissipation. Unlike Earth, space lacks a medium for conduction and convection, necessitating large radiators for cooling that considerably complicate designs and increase costs. For example, a single GPU chip, such as Nvidia's H100, requires an extensive radiator to manage heat in a vacuum, raising the overall cost and feasibility of space-based operations. Despite these challenges, space offers unique advantages for certain applications, such as preprocessing data from Earth-observing satellites and ensuring collision avoidance in heavily populated low Earth orbit. However, the need for redundancy and radiation-hardening adds layers of complexity and expense. Research indicates that operational costs for space-based GPUs can exceed those on Earth by an order of magnitude, though the allure of space's abundant solar energy remains. Nonetheless, the focus is shifting towards niche applications where the premium is justified, highlighting the intricate interplay of physics, engineering, and economics in the future of computing beyond our planet.