Chinese and U.S. research labs are racing to become the first to operate a supercomputer that runs AI systems from space.
Companies including Starlink, Blue Origin, Google, and the startup Starcloud are reinforcing satellites and chip designs to withstand launch stress, radiation, and extreme temperatures. The long-term goal is to shift AI computing off Earth, where data centers consume enormous amounts of energy and water.
This new contest echoes the Cold War–era space race, but instead of reaching the Moon, today’s competition is focused on low-Earth orbit. Nations and tech giants are now vying to place AI infrastructure above the planet.
China appears to be ahead. Researchers at the Chinese Academy of Sciences are working toward launching an orbital AI data center built around roughly 10,000 high-performance computing cards. At the same time, U.S. tech leaders such as Elon Musk, Jeff Bezos, and Google’s Sundar Pichai are pouring massive investment into their own space-based AI efforts.
Blue Origin has spent more than a year developing an orbital AI hub. Starlink plans to upgrade its satellites to carry AI workloads. Google is testing Project Suncatcher, which installs compact computing racks on satellites. But Nvidia-backed Starcloud may be furthest along.
Last month, Starcloud launched a satellite carrying an Nvidia H100 GPU—by far the most powerful chip ever sent into space—and successfully trained the NanoGPT model while in orbit. It marked the first time an AI model was trained in space and the first use of an Nvidia GPU to run a large language model beyond Earth.
Supporters argue that orbital AI data centers could dramatically cut energy use and water consumption by relying on solar power. Starcloud estimates its space-based systems could use ten times less electricity than Earth-based facilities.
China continues to advance as well. A joint project between Guoxing Aerospace and Zhejiang Lab recently launched 12 satellites, forming the first space-based computing constellation. Other Chinese firms have operated experimental space computers in orbit for years, steadily refining the technology.
Despite major technical challenges—from launch vibrations to radiation exposure—researchers believe fully operational space supercomputers could arrive in the 2030s. The only open question is who will get there first.