Google Paper Details 81-Satellite Cluster for Orbital AI Datacenters
sebkrier · x · 2026-11-05
Google researchers including Blaise Agüera y Arcas published a paper on scaling AI datacenters in space.
- An earlier Project Suncatcher milestone sent TPUs to orbit, where Gemma 4 returned the first AI response from a TPU in space (a haiku), but that was only step one.
- Scaling requires dozens of satellites flying in tight formation; the paper lays out an 81-satellite cluster design.
- Placed in a dawn-dusk sun-synchronous low Earth orbit, the cluster gets near-constant, efficient sunlight—up to 8x the annual solar yield available on Earth.
- The authors call it a major engineering challenge in its own right and detail the formation and power design.
More from Infra
- AMD ships Day 0 support for Kimi K3.1 Pro: 2.8T params, 1M context on 8x MI355X — Kimi_Moonshot · 2026-11-05
- SpaceX makes its own transformers and turbines, skipping the 1-2 year compute lead time — beffjezos · 2026-11-05
- Malaysia now has 6x more AI compute per capita than the EU — basedjensen · 2026-11-05
- Chamath: Chinese models run 112x cheaper than Anthropic — $0.50 vs $56 per million tokens — 2C_ornot2C · 2026-11-05
- 10 open-source repos to slash your AI agent's token bill, from Ollama to LiteLLM — victor_explore · 2026-11-05
- Celesto launches hosted cloud-sandbox MCP: run agent tasks with one prompt — aniketmaurya · 2026-11-05