Starcloud Trains First AI Model in Orbit with NVIDIA H100

Engineers at Starcloud announced they trained a language model in orbit. This happened on December 10, 2025. What's interesting is an NVIDIA H100 processor, similar to those in Earth's data centers, was used aboard the Starcloud-1 satellite. This spacecraft travels at roughly 500 kilometers above the planet. The H100 learned from a Shakespeare dataset, saved the model, and then generated new text while orbiting Earth. The model trained in space was Andrej Karpathy’s NanoGPT. Starcloud fed it Shakespeare's complete works, allowed it to adjust, and then generated Shakespeare-like text. This 60-kilogram technology demonstrator launched on November 2, 2025. Nvidia described the H100 as carrying 100 times more GPU compute than any earlier space-based operation. The bottom line: this achievement demonstrates the capability of training AI models directly in space, opening new possibilities for orbital operations.