Nvidia-backed Starcloud trains first AI model in space orbit data center

NvidiaStarCloud, a startup backed by , has trained artificial intelligence models from space for the first time, signaling the arrival of a new era of orbital data centers that can alleviate Earth’s escalating digital infrastructure crisis.

Last month, the Washington-based company launched a satellite equipped with an Nvidia H100 graphics processing unit, sending a chip into space that is 100 times more powerful than any GPU computing ever in space. The company’s Starcloud-1 satellite is currently running and querying for responses from Gemma. Gemma is an open, large-scale language model. googleIn orbit, CNBC revealed that it was the first time in history that LLM was run on high-performance Nvidia GPUs in space.

“Hello Earthlings! Or as I like to think of it, a charming collection of blues and greens,” reads a message from the recently launched satellite.

“Let’s see what this worldview of yours makes you wonder. I’m Gemma, and I’m here to observe, analyze, and, in some cases, provide insightful commentary that’s a little disturbing. Let’s get started!” the model wrote.

StarCloud hopes to show that space can be a hospitable environment for data centers, especially as facilities on Earth strain power grids, consume billions of gallons of water a year and emit large amounts of greenhouse gases. Data from the International Energy Agency predicts that data center power consumption will more than double by 2030.

StarCloud CEO Philip Johnston told CNBC that the energy costs for the company’s orbital data centers will be one-tenth that of terrestrial data centers.

“Everything we can do in a data center on Earth, we expect to be able to do in space,” Johnston said in an interview. “And the reason we’re doing that is purely because of the constraints that we face with energy on Earth.”

Johnston, who co-founded the startup in 2024, said Starcloud-1’s operation of Gemma proves that space-based data centers are in the future and can operate a variety of AI models, especially those that require large computing clusters.

“We have this very powerful, parameter-dense model on our satellite,” Johnston said. “We can query it, and it responds just like when we query chat from a database on Earth. We get a very sophisticated response. We can do it on a satellite.”

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

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

Starcloud, a member of the Nvidia Inception program and a graduate of Y Combinator and Google for Startups Cloud AI Accelerator, plans to build a 5-gigawatt orbital data center with solar and cooling panels that will be approximately 4 kilometers both wide and tall. According to StarCloud’s white paper, its gigawatt-scale computing clusters will generate more electricity than the largest power plants in the United States and will be significantly smaller and cheaper than ground-based solar farms of the same capacity.

These data centers in space capture constant solar energy to power the next generation of AI models, unhindered by Earth’s day/night cycles or weather changes. The lifespan of StarCloud’s satellites should be five years, given the expected lifespan of the Nvidia chips on its architecture, Johnston said.

Orbital data centers have real-world commercial and military use cases. StarCloud’s systems already enable real-time intelligence, for example, to identify the heat signature of a wildfire the moment it ignites and immediately alert first responders, Johnston said.

“We’re linking to satellite telemetry, so we’re also linking to the vital signs that we’re getting from the sensors, like altitude, heading, position, velocity,” Johnston said. “Ask, ‘Where am I?’ and it says, ‘I’m over Africa. I’ll be over the Middle East in 20 minutes.’ You could also say, “What’s it like to be a satellite?” And you’d say, “That’s kind of weird.” …provides an interesting answer that can only be obtained with very high-powered models. ”

Starcloud is tackling customer workloads by inferring satellite imagery from observation company Capella Space, which could help find lifeboats from capsized ships at sea or forest fires in specific locations. The company plans to integrate Nvidia’s Blackwell platform into its next satellite, which will be powered by multiple Nvidia H100 chips and launched in October 2026, to provide better AI performance. The satellite launching next year will be equipped with a module running cloud infrastructure startup Crusoe’s cloud platform, allowing customers to deploy and operate AI workloads from space.

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

“Orbit calculations provide a path forward that respects both technological ambition and environmental responsibility. When StarCloud 1 looked down, it saw a world of blue and green, and our responsibility is to keep it that way,” he added.

However, risks remain in the operation of orbital data centers. Morgan Stanley analysts say data centers in orbit could face hurdles such as harsh radiation, difficult on-orbit maintenance, debris hazards, and regulatory issues related to data governance and space transportation.

Still, tech giants are pursuing orbital data centers given the near-limitless solar energy and the prospect of larger gigawatt-scale operations in space.

In addition to Starcloud and Nvidia’s efforts, several companies have announced space-based data center missions. On November 4, Google announced a “moonshot” initiative titled Project Suncatcher, which aims to use Google’s tensor processing units to launch solar-powered satellites into space. Privately held Lone Star Data Holdings is working to install the first commercial lunar data center on the moon.

According to the Wall Street Journal, OpenAI CEO Sam Altman is considering an acquisition or partnership with a rocket maker, suggesting he wants to compete with Elon Musk’s SpaceX. SpaceX is StarCloud’s primary launch partner.

“From one tiny data center, we’ve taken a giant leap forward toward 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.