Particle.news
Download on the App Store

Bezos and Musk Join Orbital AI Data Center Race as First In‑Space GPU Runs an LLM

Unresolved engineering and economic obstacles mean orbital compute remains far from large‑scale reality.

Overview

  • Jeff BezosBlue Origin has worked on in‑orbit data center technology for over a year, while Elon Musk’s SpaceX is planning upgraded Starlink satellites to carry AI payloads, the Wall Street Journal reported.
  • Startup Starcloud says it powered an Nvidia GPU on a satellite and ran Google’s Gemma large language model, describing the experiment as the first use of a cutting‑edge AI chip in space and noting a small model was trained on Shakespeare.
  • Google announced a Planet Labs partnership to fly two prototype satellites with TPU chips, targeting an early‑2027 launch under its Project Suncatcher effort.
  • China’s Institute of Computing Technology is promoting a low‑Earth‑orbit “space supercomputer” concept integrating 10,000 high‑performance cards, with a lead researcher claiming an early edge for China.
  • Industry reporting and company papers underscore unresolved risks including cooling in vacuum, radiation damage, debris and fuel needs, bandwidth limits, and uncertain economics, even as proponents cite relief from Earth’s energy and siting constraints.