Starcloud Trains First AI Model in Space Using Nvidia H100 GPU

Washington-based startup Starcloud has achieved a milestone in space-based computing by training an artificial intelligence model aboard its Starcloud-1 satellite. Launched in November 2025, the satellite is equipped with Nvidia's H100 GPU, a chip 100 times more powerful than any previous spaceborne graphics processor.

The mission successfully ran Google's open-source Gemma large language model in orbit and trained NanoGPT on Shakespeare's complete works. This demonstration marks the first time machine learning operations have been conducted entirely in space, bypassing Earth's latency and infrastructure limitations.

Starcloud's ambitious roadmap includes developing a 5-gigawatt orbital data center powered by solar energy, with plans to deploy additional Nvidia H100 chips and Blackwell platforms in an October 2026 launch. The project joins similar initiatives from Google, Lonestar Data Holdings, and Aetherflux in the emerging field of extraterrestrial computing infrastructure.