The rapid expansion of artificial intelligence has fueled interest in placing data centers in orbit, where abundant solar energy and freedom from terrestrial power constraints appear to offer an attractive solution. Companies including SpaceX have suggested that orbital data centers could eventually support the growing computational demands of AI. However, the IEEE Spectrum article argues that the enthusiasm surrounding these proposals overlooks fundamental engineering realities that make space-based computing far more difficult than it first appears.
One of the biggest obstacles is thermal management. Although space is extremely cold, it is also a vacuum, eliminating the two most effective cooling mechanisms on Earth: conduction and convection. The only remaining option is radiative cooling, which is significantly less efficient. To dissipate the enormous heat generated by high-performance AI processors, orbital data centers would require exceptionally large radiator panels, adding considerable mass, complexity, and cost to every satellite.
Radiation presents another major engineering challenge. High-energy particles in orbit can damage semiconductors, introduce memory errors, and shorten the lifespan of advanced GPUs and AI accelerators. Hardware failures that are routine to repair in terrestrial data centers become extremely expensive in space, where servicing missions remain limited and costly. In addition, AI hardware evolves rapidly, raising concerns that expensive orbital systems could become obsolete long before the end of their operational lives.
The article also questions whether orbital data centers make economic sense. Deploying enough computing capacity would require an enormous increase in satellite manufacturing and launch rates, far beyond current capabilities. A constellation of AI satellites would also increase orbital congestion and the risk of space debris, while adding pressure to an already crowded low Earth orbit. Although reusable rockets continue to reduce launch costs, the economics still favor terrestrial facilities for most AI workloads.
Rather than dismissing the concept entirely, the article suggests that orbital data centers could become viable over the next 5–10 years if launch costs continue to decline and advances are made in thermal management, radiation protection, and satellite servicing. For now, however, the idea remains more aspirational than practical. The future of orbital computing will depend less on visionary announcements and more on solving the demanding physics and engineering problems that continue to limit large-scale computing beyond Earth.