Nvidia-backed Starcloud trains first AI model in space, orbital data centers

NvidiaBacked startup Starcloud has trained an AI model from space for the first time, signaling a new era for orbital data centers that could alleviate Earth’s growing digital infrastructure crisis.

Last month, the Washington-based company launched a satellite with an Nvidia H100 graphics processing unit, sending a chip 100 times more powerful than any GPU computing device previously found in space. Currently, the company’s Starcloud-1 satellite is operating and querying responses from Gemma, an open, large language model. GoogleIn orbit, for the first time in history, a Master of Science (LLM) was run on a high-power Nvidia GPU in space, CNBC has learned.

“Greetings Earthlings! Or as I prefer to think of you; a stunning collection of blues and greens,” reads a message from the recently launched satellite. “Let’s see what wonders this perspective on your world holds. I’m Gemma, and I’m here to observe, analyze, and perhaps offer the occasional slightly disturbingly insightful comment. Let’s get to it!” wrote the model.

Starcloud wants to show that space can be a suitable environment for data centers, especially in an environment where Earth-based facilities strain power grids, consume billions of gallons of water annually and produce large amounts of greenhouse gas emissions. Electricity consumption of data centers It is expected to more than double by 2030According to data from the International Energy Agency.

Starcloud CEO Philip Johnston told CNBC that the company’s orbital data centers will have 10 times lower energy costs than terrestrial data centers.

“I expect that anything you can do in a terrestrial data center can also be done in space. And the reason we’re doing that is purely because of the constraints we face on energy terrestrially,” Johnston said in an interview.

Johnston, who co-founded the startup in 2024, said Starcloud-1’s Gemma operation is proof that space-based data centers could exist in the future and run a variety of AI models, especially those requiring large computing clusters.

“This very powerful, very parameter-dense model lives on our satellite,” Johnston said. “We can query it, and it will respond in the same way that if you query a conversation from a database on Earth, it will give you a very complex answer. We can do that with our satellite.”

“Seeing Gemma run in the harsh environment of space is a testament to the flexibility and robustness of open models,” Google DeepMind product director Tris Warkentin told CNBC.

In addition to Gemma, Starcloud was able to train NanoGPT, an LLM created by OpenAI founding member Andrej Karpathy, on the H100 chip using the complete works of Shakespeare. This led to the model speaking Shakespearean English.

Starcloud — a member Nvidia Home Graduated from Y Combinator program and Google Cloud AI Accelerator for Startups – plans to build a 5-gigawatt orbital data center with solar and cooling panels approximately 4 kilometers long in both width and height. According to Starcloud, a computing cluster this gigawatt size would produce more power than the largest power plant in the United States and would be significantly smaller and cheaper than a terrestrial solar farm of the same capacity. white paper.

These data centers in space will continuously capture solar energy to power next-generation AI models, unimpeded by Earth’s day and night cycles and weather changes. Starcloud’s satellites should have a five-year lifespan, given the expected lifespan of the Nvidia chips in its architecture, Johnston said.

Orbital data centers will have real-world commercial and military use cases. Johnston said Starcloud’s systems can already provide real-time intelligence and, for example, detect the thermal signature of a wildfire as it ignites and immediately alert first responders.

“We interconnect the satellite’s telemetry, so we also interconnect the vital signals it receives from the sensors, such as altitude, direction, position, speed,” Johnston said. “You can ask him this question: ‘Where are you right now?’ He’ll say, ‘I’m over Africa, in 20 minutes I’ll be over the Middle East.’ You might also say: ‘How does it feel to be a satellite? And he’ll say: ‘It’s a little weird’… He’ll give you an interesting answer that you can only get with a very powerful model.”

Starcloud works on customers’ workloads by making inferences on satellite images from observation company Capella Space, which can help detect lifeboats on capsized ships at sea and wildfires in a particular location. The company will include several Nvidia H100 chips to deliver greater AI performance and will integrate Nvidia’s Blackwell platform in its next satellite launch in October 2026. The satellite to be launched next year will include a module cloud infrastructure startup runs a cloud platform from CrusoeIt allows customers to deploy and run AI workloads from space.

“Running advanced AI from space solves critical bottlenecks facing data centers on Earth,” Johnston told CNBC.

“Orbital computing offers a path forward that respects both technological ambition and environmental responsibility. When Starcloud-1 looked down, it saw a world of blue and green. Our responsibility is to keep it that way,” he added.

Alongside the efforts of Starcloud and Nvidia, many companies have announced space-based data center missions. On November 4, Google announced a “moonshot” initiative titled. Sun Catcher ProjectIt aims to place solar-powered satellites into space with Google’s tensor processing units. Privately owned Lonestar Data Holdings is working to place the first commercial lunar data center on the lunar surface.

OpenAI CEO Sam Altman said: explored an acquisition or partnership with a rocket manufacturerIt suggests a desire to compete against Elon Musk’s SpaceX, according to The Wall Street Journal. SpaceX is a major launch partner of Starcloud.

“From a small data center, we’ve taken a giant leap forward into a future where orbital computing harnesses the infinite power of the sun,” Dion Harris, Nvidia’s senior director of AI infrastructure, said of Starcloud’s launch in early November.