Starcloud Launches First AI-Powered Satellite to Train Models in Orbit

A U.S. startup called Starcloud has taken a bold step into space-based computing. In November 2025, the company launched its first satellite, Starcloud-1, carrying an advanced AI system. The move marks a push to bring high-performance data processing beyond Earth’s atmosphere. Starcloud-1, roughly the size of a household refrigerator and weighing 60 kilograms, hosts an Nvidia H100 GPU. This chip is about 100 times more powerful than any GPU previously sent into space. Its primary task is training Google’s open-source Gemma AI model while in orbit.

The AI made its first contact with users shortly after activation. Gemma’s greeting included a playful message: ‘Greetings, Earthlings! Or, as I prefer to think of you: a fascinating collection of blue and green.’ Beyond this, Starcloud has also experimented with another AI model, NanoGPT, trained on Shakespeare’s works to communicate in Elizabethan English.

Philip Johnston, Starcloud’s CEO, explained the company’s vision. He aims to replicate Earth-based data centre capabilities in space, driven by energy constraints on the ground. The long-term plan involves deploying orbital data centres with five gigawatts of power, generated by solar arrays stretching four by four kilometres.

Each satellite is designed for a five-year operational lifespan, matching the typical usage cycle of Nvidia’s AI GPUs. These space-based facilities could eventually handle the most demanding AI workloads without relying on terrestrial infrastructure. Starcloud’s launch of Starcloud-1 demonstrates the potential for AI training beyond Earth. The satellite’s powerful GPU and long-term energy plans suggest a future where orbital data centres play a key role in computing. If successful, the project could reshape how AI models are developed and deployed.