Starcloud trains AI in orbit

Startup Starcloud reportedly trained the first large language model in orbit in December 2025 using an Nvidia H100, the same class of accelerator that powers terrestrial AI data centers, roughly 500 kilometres above Earth. The achievement is early and symbolic, but it points to a serious question for AI infrastructure: whether space can eventually offer advantages for energy, cooling, latency to satellites or data generated off-planet. Running an H100 in orbit also tests radiation tolerance, power stability and remote operations. If space-based compute advances, today’s launch companies, chipmakers and cloud providers could converge in a new infrastructure market.

Sources from the briefing

• In December 2025, startup Starcloud trained the first large language model ever trained in orbit, using an NVIDIA H100 — the same class of GPU built for Earth's AI data centres, now running roughly 500 kilometres above the planet. - ScienceBlog.comScienceBlog.com
• In December 2025, startup Starcloud trained the first large language model ever trained in orbit, using an NVIDIA H100 — the same class of GPU built for Earth's AI data centres, now running roughly 500 kilometres above the planet. - ScienceBlog.comScienceBlog.com