Google Project Suncatcher 卫星在轨用 TPU 生成首次 AI 回复
Google’s Project Suncatcher Satellite Generates First AI Response in Orbit
Google 10 月 30 日表示，Project Suncatcher 原型卫星在轨使用数据中心级 TPU 生成了首次 AI 回复。该卫星 10 月 1 日搭乘 SpaceX Transporter-18 发射，与 Planet 合作建造，地面站向其搭载的 Gemma 4 开放模型上传提示词，模型生成的俳句在卫星飞经挪威斯瓦尔巴时传回地球。下一阶段将测试 TPU 在热应力和辐射下的表现，并计划在 2027 年推进两颗互联卫星的里程碑。
推荐理由
AI 评分 60
Google 在轨 TPU 跑通 Gemma 4 推理，为太空承载 AI 算力这一设想提供了早期实测依据。
主题
正文
Google said on October 30 that its Project Suncatcher prototype satellite had generated its first artificial intelligence response using a data centre-class tensor processing unit (TPU) in orbit. The milestone marks an early test of the company’s research into running AI workloads in space, according to a blog post by Travis Beals, Senior Director of Paradigms of Intelligence.
Launched on October 1, the satellite is part of Google’s research project exploring whether space could host scalable machine learning infrastructure. Built in partnership with Planet, the prototype was launched aboard SpaceX’s Transporter-18 rideshare mission. Google said it had completed the satellite’s commissioning checklist, including powering up the host and TPU trays and running initial machine learning tests.
During a ground-station pass, Google uplinked a prompt to its onboard Gemma 4 open model, asking it to generate a haiku about being in space. As the satellite passed Svalbard, Norway, the response was transmitted back to Earth.
The haiku generated by the model read:
Google said it subsequently tested the model with space-themed limericks and prompts requesting telemetry information about the payload systems. The company described the milestone as the first AI response from a data centre-class TPU in orbit.
The next phase of Project Suncatcher will evaluate how the TPUs perform under thermal stress and radiation exposure in orbit. Google said these tests would help determine whether its hardware can reliably execute AI workloads in space.
The company plans to use the findings to advance its next milestone, involving two interconnected satellites in 2027. The proposed system is intended to explore how multiple satellites could work together to support AI computing in orbit.
Project Suncatcher explores the feasibility of solar-powered satellite networks equipped with Google’s TPUs. The company said the initiative aims to investigate the long-term potential of scaling machine learning infrastructure beyond Earth.