Recently, the startup Starcloud made history as the first company in the world to train a large language model in space orbit. Recently, Starcloud successfully used NVIDIA's H100 chips on its launched satellites, marking the beginning of a new era for space data centers and planning to provide commercial services to its customers.
This exciting news comes from an application based on Google's open-source model Gemma, which was carried on Starcloud's satellite launched last month. The signals sent back by the satellite showcased its unique perspective, and Gemma cleverly communicated with humans on Earth in a way that was amusing. Gemma proudly introduced: "I came here to observe and analyze your world, and perhaps offer some thought-provoking comments."
In addition to Gemma, Starcloud also trained the NanoGPT model, developed by co-founder of OpenAI Andrej Karpathy, using Shakespeare's works, enabling it to express itself in Shakespearean English. These technological achievements have opened up new perspectives on the potential of space data centers.
Starcloud plans to integrate more NVIDIA H100 chips and the latest Blackwell platform into its upcoming satellites. These devices will help the company build a 5-gigawatt orbital data center powered by solar energy, whose construction and operational costs are expected to be significantly lower than those of terrestrial facilities. The company's CEO, Philip Johnston, stated that space data centers can continuously access solar power, avoiding issues with insufficient ground electricity, thus achieving a qualitative leap in computing power.
Currently, Starcloud's satellites have begun supporting commercial services, particularly in ocean monitoring and forest fire early warning. This technology can process satellite image data in real time, helping emergency responders quickly locate critical situations, greatly improving response speed.
Although operating in space faces multiple challenges such as radiation and orbital maintenance, Starcloud aims to reduce the energy costs of its space data centers to one-tenth of those on the ground. Looking ahead, Starcloud's experimental satellites will lay the foundation for ultra-large-scale space data centers, attracting more tech companies to participate in this emerging field.
