Starcloud Shatters Earthly Limits: First AI Model Trained in Space Using Nvidia’s H100 GPU

Forget terrestrial data centers—the final frontier just became the newest training ground for artificial intelligence. Starcloud has successfully run the first AI training session in orbit, leveraging Nvidia's powerhouse H100 GPU to process algorithms where few servers have gone before.

Why Space is the Next Cloud Region

Training in microgravity bypasses traditional thermal and power constraints that throttle performance on Earth. It's not just a publicity stunt; it's a radical rethinking of computational infrastructure. The H100, already a legend in data centers, just got its astronaut wings.

The Hardware That Left the Atmosphere

This wasn't a stripped-down experiment. Starcloud deployed the full, unadulterated H100 GPU—the same chip driving the planet's most demanding AI workloads. No compromises, just pure silicon in a vacuum. It turns out radiation hardening and cosmic ray mitigation are now part of the AI engineer's toolkit.

A New Benchmark for Decentralized Compute

The mission demonstrates more than technical prowess; it sketches a blueprint for truly decentralized, resilient computing. If AI can train independently of terrestrial grids and geopolitical borders, the entire concept of server location gets a rewrite. It's edge computing, taken to the literal extreme.

Of course, the venture capitalists funding this are probably just looking for a story to tell before their next funding round—nothing says 'moonshot' quite like an actual moonshot. But behind the space-age hype lies a tangible shift: the ceiling for AI infrastructure is no longer the roof of a warehouse, but the Kármán line. The race for compute supremacy has officially gone orbital.

TLDR

• Starcloud launched a satellite with Nvidia H100 GPU in November 2025, running Google’s Gemma AI model in space for the first time
• The company also trained NanoGPT using Shakespeare’s works on the orbital chip, which is 100 times more powerful than previous space GPUs
• Starcloud plans to build a 5-gigawatt orbital data center powered by constant solar energy to reduce Earth’s digital infrastructure strain
• The next satellite launch in October 2026 will include multiple Nvidia H100 chips and Nvidia’s Blackwell platform
• Google, Lonestar Data Holdings, and Aetherflux have also announced space-based data center projects

Starcloud, a Washington-based startup backed by Nvidia, has trained an artificial intelligence model in space for the first time. The company launched its Starcloud-1 satellite in early November 2025 equipped with an Nvidia H100 graphics processing unit.

We have just used the @Nvidia H100 onboard Starcloud-1 to train the first LLM in space!

We trained the nano-GPT model from Andrej @Karpathy on the complete works of Shakespeare and successfully ran inference on it.

We have also run inference on a preloaded Gemma model, and we… pic.twitter.com/DDe7YpevZY

— Adi Oltean (@AdiOltean) December 10, 2025

The chip is 100 times more powerful than any GPU previously sent to space. The satellite is now running Google’s Gemma, an open large language model, in orbit.

This marks the first time an LLM has operated on a high-powered Nvidia GPU in outer space. The model sent a message to Earth saying “Greetings, Earthlings” and describing itself as ready to observe and analyze from its orbital position.

Starcloud also trained NanoGPT, an AI model created by OpenAI founding member Andrej Karpathy, on the H100 chip. The company used the complete works of Shakespeare for training, which resulted in the model speaking in Shakespearean English.

CEO Philip Johnston said the company wants to prove that space can host data centers. Earth-based facilities strain power grids and consume billions of gallons of water annually while producing greenhouse gas emissions.

Energy and Environmental Benefits

The electricity consumption of data centers is projected to more than double by 2030, according to the International Energy Agency. Johnston said Starcloud’s orbital data centers will have 10 times lower energy costs than terrestrial facilities.

“Anything you can do in a terrestrial data center, I’m expecting to be able to be done in space,” Johnston told CNBC. He said the reason is purely because of energy constraints on Earth.

The satellite can answer queries about its location and status. It can identify its position over Africa and predict when it will be over the Middle East in 20 minutes.

Starcloud plans to build a 5-gigawatt orbital data center with solar and cooling panels measuring roughly 4 kilometers in both width and height. The facility WOULD capture constant solar energy, unaffected by Earth’s day and night cycles or weather.

Commercial and Military Applications

The company is working on customer projects using satellite imagery from Capella Space. These systems could spot lifeboats from capsized vessels and detect forest fires at the moment they ignite.

Johnston said the satellites could enable real-time intelligence and immediately alert first responders. The company integrated the satellite’s telemetry so users can ask about its vital signs like altitude and speed.

Starcloud’s satellites are expected to have a five-year lifespan based on the Nvidia chips’ expected lifetime. The company plans its next launch for October 2026 with several Nvidia H100 chips and Nvidia’s Blackwell platform.

Industry Competition

Other companies have announced similar projects. Google unveiled Project Suncatcher on November 4, which aims to put solar-powered satellites into space with Google’s tensor processing units.

Lonestar Data Holdings is working to put a commercial lunar data center on the moon’s surface. Aetherflux announced plans to deploy an orbital data center satellite in the first quarter of 2027.

Morgan Stanley analysts noted that orbital data centers could face challenges including harsh radiation, difficulty of maintenance, debris hazards, and regulatory issues. Despite these risks, tech giants are pursuing the technology for access to solar energy and larger operations in space.

The next Starcloud satellite will include a module running a cloud platform from Crusoe, allowing customers to deploy AI workloads from space.