Starcloud, supported by Nvidia, has become the first company to train an AI model in space, marking a key step in developing orbital data centres. Its Starcloud-1 satellite carries an Nvidia H100 GPU and runs Gemma, an open large language model from Google—a first for such advanced AI in orbit.
The satellite greets Earth with, “Greetings, Earthlings! Or, as I prefer to think of you—a fascinating collection of blue and green.” Starcloud aims to ease the strain on Earth’s data centres, which face rising energy use and emissions, by using space’s abundant solar power.
CEO Philip Johnston said orbital centres could cut energy costs to a tenth of terrestrial levels.
Starcloud has also trained NanoGPT on Shakespeare’s works in orbit, showing AI’s adaptability in space. The company plans a massive 5-gigawatt solar-powered data centre in orbit, offering uninterrupted energy free from Earth’s day-night cycle.
Beyond saving energy, these centres can enable real-time insights like wildfire detection. The next satellite launching in 2026 will expand AI capacity with more Nvidia chips and cloud tech from Crusoe.
Though challenges remain—such as radiation and debris—tech firms including Google and Lonestar Data Holdings are racing to build space-based data centres. Nvidia’s Dion Harris described Starcloud’s launch as “a giant leap toward a future where orbital computing harnesses the infinite power of the sun.”