Particle.news
Download on the App Store

Orbital Data Centers Advance as Starcloud Trains LLM in Space and Aetherflux Targets 2027 Node

Earth’s energy limits are propelling early space compute tests despite unresolved radiation, cost and debris risks.

Overview

  • Starcloud says its Starcloud‑1 satellite ran and trained AI models in orbit on an Nvidia H100, including Google’s open Gemma LLM and NanoGPT, in what CNBC reports is a first for training an LLM from space.
  • Aetherflux announced a 2026 demo to beam power from low Earth orbit to the ground via infrared lasers and set Q1 2027 for its first “Galactic Brain” data‑center node.
  • The startup says its node will use continuous solar power and radiative cooling, with optical links and relay networks for availability; it also cites prior U.S. Department of Defense support for a proof‑of‑concept power‑transmission test.
  • SpaceX, Blue Origin and Google are exploring orbital compute: Reuters cites a WSJ report that Blue Origin has been developing enabling tech, Bloomberg reporting links SpaceX’s IPO planning to orbital compute investment, and Google’s Project Suncatcher targets prototype satellites by early 2027.
  • Proponents point to abundant solar energy and free cooling to ease terrestrial constraints on power and water, while analysts flag hurdles including radiation‑hardening, in‑orbit servicing, debris and traffic management, launch costs and data‑governance questions.