According to CoinMeta, Google CEO Sundar Pichai announced that the Project Suncatcher prototype satellite launched at the beginning of October successfully performed AI inference in space using data center-level TPU for the first time and transmitted the results back to Earth. Engineers sent commands from the ground to make the TPU on the satellite run Gemma 4, which resulted in a haiku about space. When the satellite flew near the Svalbard archipelago in Norway, it transmitted back three lines of poetry: "Below my feet is a blue planet, data streams flow silently, starlight nourishes my core." The satellite is equipped with 4 TPU, but due to power limitations, only one can be operational at a time for about 20 minutes each time. Although space is cold, there is no air to help dissipate heat from the chips, so they rely on heat pipes and radiators to release heat outward. The impact of cosmic radiation on chip performance also needs further testing. In fact, AI has already been in space before. Last December, American space computing startup Starcloud operated Gemma in orbit using NVIDIA H100 and even trained a small language model. This time, Google used its own TPU; the next challenges are to ensure the long-term stable operation of the chips and how multiple satellites can work together in computing.