Starcloud Achieves First AI Training in Space, Marking New Phase in Orbital Data Center Race
December 10, 2025
In a milestone for the burgeoning field of orbital computing, Starcloud, a startup backed by Nvidia, has successfully trained and run an artificial intelligence model from space for the first time. This achievement signals a potential paradigm shift for the global data center industry, which is grappling with an escalating infrastructure crisis driven by soaring energy demands, water consumption, and greenhouse gas emissions from terrestrial facilities.
Last month, the Washington-based company launched its Starcloud-1 satellite, equipped with a powerful Nvidia H100 graphics processing unit (GPU). The company now confirms the satellite is operational and has successfully queried responses from Google's open-source large language model, Gemma, in orbit. This marks the first instance of an LLM running on a high-performance Nvidia GPU in outer space. In a demonstration of its capability, the model transmitted a message: "Greetings, Earthlings! Or, as I prefer to think of you — a fascinating collection of blue and green... I'm Gemma, and I'm here to observe, analyze, and perhaps, occasionally offer a slightly unsettlingly insightful commentary."
Starcloud's mission is to prove that space can be a viable environment for data centers, offering a solution to Earth-bound constraints. According to the International Energy Agency, data center electricity consumption is projected to more than double by 2030. Starcloud CEO Philip Johnston stated that the company's orbital data centers aim for energy costs ten times lower than their terrestrial counterparts. "Anything you can do in a terrestrial data center, I'm expecting to be able to be done in space. And the reason we would do it is purely because of the constraints we're facing on energy terrestrially," Johnston told CNBC. He cited the operation of Gemma as proof that space-based centers can host sophisticated AI models.
Beyond Gemma, the company trained NanoGPT—a model created by OpenAI founding member Andrej Karpathy—on the satellite's H100 chip using Shakespeare's complete works, leading the AI to respond in Shakespearean English. Tris Warkentin, product director at Google DeepMind, noted that "seeing Gemma run in the harsh environment of space is a testament to the flexibility and robustness of open models."
The startup, a member of the Nvidia Inception program, has ambitious plans to build a massive 5-gigawatt orbital data center complex with solar and cooling panels measuring roughly 4 kilometers in both width and height. According to the company's white paper, a compute cluster of this scale would generate more power than the largest power plant in the United States while being substantially smaller and cheaper than a terrestrial solar farm of equivalent capacity. These orbital facilities would capture constant solar energy, unimpeded by Earth's day-night cycles or weather, to power next-generation AI. The satellites are designed with a five-year lifespan based on the expected durability of the Nvidia chips.
Johnston outlined immediate commercial and military applications, such as enabling real-time intelligence and spotting the thermal signature of a wildfire at ignition to alert first responders. The company is already processing customer workloads, running inference on satellite imagery from Capella Space to aid in tasks like locating lifeboats from capsized vessels. Starcloud's next satellite, scheduled for launch in October 2026, will integrate several Nvidia H100 chips and the Blackwell platform for greater AI performance, and will feature a module from cloud infrastructure startup Crusoe, allowing customers to deploy AI workloads directly from space.
However, significant challenges remain. Analysts from Morgan Stanley have highlighted potential hurdles including harsh radiation, the difficulty of in-orbit maintenance, space debris hazards, and complex regulatory issues surrounding data governance and space traffic. Despite these risks, the prospect of nearly limitless solar energy and gigawatt-scale operations is attracting major players. Alongside Starcloud and Nvidia, several companies have announced space-based data center initiatives, including Google's "Project Suncatcher," Lonestar Data Holdings' lunar data center project, and Aetherflux's plan to deploy an orbital data center satellite in early 2027.
Referring to Starcloud's launch, Dion Harris, senior director of AI infrastructure at Nvidia, said, "From one small data center, we've taken a giant leap toward a future where orbital computing harnesses the infinite power of the sun."
Source: CNBC
