AI Model Launches into Space! NVIDIA and Starcloud Collaborate to Make H100's First On-Orbit Run of LLM Gemma

American startup Starcloud announces successful deployment of NVIDIA( H100 GPU into outer space, and its operation of Google open-source LLM Gemma in orbit, setting the world’s first record for “LLM execution in space.” The team hopes to demonstrate that outer space has the potential to become the next-generation data center, which may alleviate Earth’s pressures on energy, land, and infrastructure. As companies like Google, Lonestar, and Aetherflux continue to invest, orbital computing )Orbital Computing( has become an emerging battleground in the tech industry.

Starcloud Launches Space Mission in November, H100 Capable of Space Computing

Based in Washington State, Starcloud launched the Starcloud-1 satellite in November this year, equipped with NVIDIA’s H100 GPU, making history as the first time such a high-performance processor has been sent into outer space.

The H100 GPU’s computing power is 100 times that of chips used in previous space missions, representing a major breakthrough in space computing capability. Once in orbit, the satellite immediately activated Google LLM Gemma, which can receive questions and reply directly in space. The first message from Gemma in orbit stated:

“Earthlings, hello! I am Gemma, here to observe and analyze.”

Starcloud states that this is the first-ever successful execution of LLM in outer space using NVIDIA’s high-end GPU.

Starcloud Promotes Next-Generation Data Centers, Space Energy Advantages Emerging

Starcloud CEO Philip Johnston pointed out that global data centers are facing high energy demands, massive water usage, and carbon emission pressures, and space offers a whole new option. He believes that solar energy in space is unaffected by day-night cycles and weather, and energy costs could be ten times lower than on the ground, emphasizing:

“What can be done on the ground, can also be done in space.”

Currently, Starcloud has successfully operated two AI models in space orbit, namely Gemma and NanoGPT. They were trained on orbit using H100, and incorporated Shakespeare’s complete works as training data, enabling the models to generate responses in Shakespearean tone.

Google DeepMind Product Director Tris Warkentin praised this achievement, considering Gemma’s smooth operation in such harsh space conditions as the best example of the robustness of open-source models.

Starcloud Space Data Center Blueprint Revealed, Planning 5GW Supercluster

In its latest white paper, Starcloud proposed a broader long-term blueprint to build a 5GW space data center in orbit, equipped with large solar panels and heat dissipation structures, covering approximately 4 km by 4 km. The company notes that building such space facilities would cost less than equivalent-sized ground-based solar farms and offer better energy efficiency.

Johnston added that the satellite’s lifespan is estimated at about five years, similar to the cycle of NVIDIA GPUs, ensuring continuous system operation.

Real-time Monitoring and Disaster Relief, New Capabilities from Space AI Models

Starcloud has begun assisting clients in deploying inference tasks in orbit, including integration with Capella Space’s satellite imagery for maritime rescue, forest fire detection, and hotspot identification in emergency scenarios. Space-based LLMs can also respond about the satellite’s own status, for example:

“Where am I now, where will I be in 20 minutes, and what does it feel like to be a satellite.”

Johnston said these new interactive capabilities stem from LLM-based real-time reasoning in space. The next satellite launch by Starcloud is scheduled for October 2026, which will carry more H100 units, adopt the new Blackwell architecture, and integrate Crusoe’s cloud platform modules, allowing customers to perform AI tasks directly in space.

Space Computing Race Heats Up, Google and Startups Accelerate Deployment

In addition to Starcloud, Google has launched Project Suncatcher, planning to combine TPU with solar-powered satellites for computation. Lonestar Data Holdings is working on the world’s first commercial lunar data center. Baiju Bhatt, former CEO of Robinhood and founder of Aetherflux, announced plans to deploy a space data center in the first quarter of 2027.

NVIDIA officials believe that Starcloud’s success marks an important milestone, stating:

“Starting from a mini data center, we are heading toward a future of orbital computing powered by unlimited solar energy.”

Space Data Centers Still Face Risks, Radiation and Maintenance Challenges

Morgan Stanley analysts pointed out that space data centers still face multiple challenges, including risks of hardware damage from space radiation, difficulty of in-orbit maintenance compared to ground environments, and increased collision risks due to space debris.

Additionally, regulations on data governance and space traffic management are not yet mature, creating uncertainties for commercial deployments. Nevertheless, the near-infinite solar energy and vast deployable space make companies eager to invest in in-orbit computing technologies.

Starcloud Emphasizes Environmental Responsibility, Technology Ambitions and Sustainability Go Hand-in-Hand

Johnston expressed that in-orbit computing is a path to balance technological ambitions with environmental responsibility. He said:

“Starcloud-1, when overlooking Earth, sees blue and green. Our responsibility is to keep it that way.”

Image: A scene of NVIDIA and Starcloud’s Starcloud-1 satellite traveling in space

This article: AI models in space! NVIDIA and Starcloud team up to enable H100 to run LLM Gemma in orbit, first appearing in Chain News ABMedia.