What to know

  • Launched last month into orbit, Nvidia-backed startup Starcloud has trained an AI model from space.
  • Starcloud-1 satellite runs Nvidia H100 GPU and Google's Gemma model in orbit, achieving a historic first for space-based AI training.​
  • Orbital data centers promise 5x more efficient solar power, passive cooling, and scalability beyond Earth's limits.​
  • Competitors like SpaceX, Blue Origin, and Aetherflux accelerate plans amid energy demands for AI.​
  • Challenges include radiation, debris risks, and regulations, but Starcloud eyes 5GW clusters next.

Starcloud launched Starcloud-1 last month via SpaceX, packing the Nvidia H100 GPU—100 times more powerful than prior space chips. This Washington-based startup uses an LLM created by OpenAI founding member Andrej Karpathy. The model is further trained on works of Shakespeare, and its first apropos response was: "Greeting, Earthlings!"

The Starcloud-1 satellite is also querying responses from Google's open Gemma model in orbit. This marks the first time an LLM running an Nvidia GPU is operating in outer space. From its vantage point, the model called earth "a fascinating collection of blue and green" and went on to say: "Let's see what wonders this view of your world holds. I’m Gemma, and I’m here to observe, analyze, and perhaps, occasionally offer a slightly unsettlingly insightful commentary. Let’s begin!"

This marks the first step towards improving the digital infrastructure crisis on Earth, a promising start for data centers in space where, being in constant sunlight in sun-synchronous orbits, they won't have to face the same level of resource challenges as on Earth. On top of that, deep space cooling (near 4 Kelvin) eliminates water use and chillers, targeting PUE like top Earth centers.

Starcloud CEO mentioned that the space data centers will consume 10 times less energy than those on Earth.

Shifting AI to space eases Earth's energy crunch - with tripling demand projected - while enabling GW-scale training unimpeded by permits or grids. Real-time apps like disaster monitoring will further gain a low-latency edge as emissions drop post-launch. This milestone positions orbital compute as AI's sustainable frontier.