谷歌首颗AI卫星在轨运行，TPU首次生成太空响应

Google’s first AI satellite boots TPU in orbit, generates response using Gemma 4

Google confirmed its first satellite with an onboard TPU has successfully initialized and executed inference using a lightweight version of DeepMind’s Gemma 4 model—returning a text response to Earth. The system supports remote model updates and real-time command dispatch via low-earth-orbit links, with sub-second latency. This proves viable on-orbit AI inference—not just telemetry relay. For AI application startups, this expands the viable edge: use cases like autonomous maritime monitoring, disaster-response image triage, or global asset tracking no longer require ground-network backhaul. Founders should explore ‘space-ground co-intelligence’ patterns—e.g., satellite-side filtering + ground-side refinement, or offline-executable decision modules—rather than assuming all intelligence must reside in centralized clouds.