A startup named Starcloud has successfully trained an artificial intelligence (AI) model in outer space for the first time, CNBC reports. This achievement marks a significant step toward using data centers located in orbit around the Earth. Recently, the company launched a satellite carrying a high-performance GPU. This GPU is reportedly one hundred times more powerful than any computing hardware previously sent into space.
The satellite is currently operating "Gemma," a large language model. The system successfully sent a message back to Earth, proving that advanced computing can function in orbit. In another experiment, the satellite used the complete works of Shakespeare to train a different model, which subsequently generated text in a Shakespearean style.
Future applications
Starcloud aims to prove that space is a hospitable environment for data centers. Operating these facilities on Earth is becoming increasingly difficult because they consume billions of gallons of water and strain local power grids. By moving this infrastructure to space, companies can utilize constant solar energy and the natural cold of the vacuum for cooling. The researchers believe that orbital data centers could eventually have energy costs that are ten times lower than those on the ground.
This technology allows for the real-time analysis of data from above. For example, the system could detect the heat signature of a wildfire the moment it ignites or identify lifeboats from capsized vessels at sea. While the potential is vast, there are significant risks involved. Experts warn of challenges such as harsh radiation, the danger of colliding with space debris, and the inability to repair hardware once it is in orbit. Despite these hurdles, other major technology firms are now exploring space-based computing to harness the sun's energy and reduce the environmental impact of digital infrastructure on Earth.
Starcloud is backed by Nvidia. "Starcloud... plans to run Gemma - an open model from Google - in orbit on H100 GPUs, proving even large language models can run in outer space," notes a recent Nvidia blog post.