Themata.AI
Follow coverage
Themata.AI
← All topics

Google tensor processing units

Source reports covering Google tensor processing units. Stories will appear when analysis and community reactions are ready.

1 reports · page 1 of 1

Browse reports ↓Search this topic

Connected topics are temporarily unavailable. You can still browse stories below.

Source reports

  • Google’s first data center satellite is online and writing poetry

    news

    3 min read

    Google’s Suncatcher prototype space data center is operating in orbit after launching earlier in October 2026 on a SpaceX Falcon 9 mission carrying more than 100 spacecraft. The satellite has four Google Tensor Processing Units, although it can run only one at a time for roughly 20 minutes because its power system generates up to 1 kilowatt. Following a three-week commissioning process, engineers used the preloaded Google DeepMind Gemma 4 model to generate a space-themed haiku and to report orbital conditions. Google said the spacecraft was operating nominally. Suncatcher is testing whether TPUs can perform inference workloads under orbital thermal conditions and radiation exposure. Radiation can produce memory bit flips that corrupt data without conventional monitoring alerts. Ground-based particle-accelerator tests indicated radiation should have limited effects on inference but could pose greater problems for model training. If the first satellite validates the hardware, Google plans to launch two higher-power satellites in 2027 with inter-satellite optical links. Google envisions close-flying networks of satellites sharing distributed AI workloads, a model dependent on high-speed laser connections and on fully, rapidly reusable Starship launches. Suncatcher researchers estimated earlier in October that SpaceX would need 1,800 Starship flights to match the learning-driven cost reductions achieved with its earlier vehicles.

    SatellitesCloud computingAI infrastructure
Explore all topics →