Google的“Suncatcher（捕日者）”原型太空数据中心卫星已正式投入运行，并成功与地面建立通信。这颗卫星于本月早些时候搭乘SpaceX猎鹰9号火箭升空，与100余颗航天器一同进入轨道，标志着Google将人工智能计算能力扩展至太空环境的重要一步。
这颗卫星搭载GoogleTensor Processing Unit（TPU）人工智能加速芯片，并可产生最高约1千瓦的电力。项目目标是为未来由多颗近距离协同飞行卫星组成的“轨道数据中心网络”铺路，使大型AI推理任务能够直接在太空中完成。
不过，目前的技术仍面临诸多挑战。由于TPU功耗较高，卫星每次只能启动芯片约20分钟，这使本就复杂的天地通信工作变得更加困难。经过约三周时间的在轨测试与系统校准后，卫星终于进入正式运行阶段。
为了验证系统功能，Google工程师本周首次向卫星发送提示指令，要求预装于卫星上的Google DeepMind Gemma 4模型创作一首关于太空的俳句。模型给出的作品大意为：“蓝色星球在下方静静闪耀，数据流安然穿梭，星光滋养我的核心。”
除了让AI在太空写诗之外，Google还利用该模型收集和汇报卫星周边轨道环境信息。Google发言人表示，目前卫星运行状态正常，各项系统均按预期工作。
接下来，Google将重点评估TPU芯片在太空极端环境下的可靠性，包括温度变化带来的热应力以及宇宙辐射对硬件的影响。虽然Google此前已在地面粒子加速器中对TPU进行过大量测试，但真实太空环境中的长期表现仍需要进一步验证。
其中最受关注的问题之一是辐射导致的“比特翻转”现象，即计算机存储器中的数据位在高能粒子作用下发生意外变化。这种情况可能造成数据损坏，而常规监测系统不一定能够及时发现。
此前地面实验显示，辐射对AI推理任务的影响相对有限，但在模型训练过程中可能带来更明显的问题。如今，Google终于能够通过实际在轨运行数据验证这些推测，而不仅仅依赖实验室模拟结果。
如果此次任务证明TPU能够在太空中稳定运行，Google计划于2027年再发射两颗采用更先进芯片的后续卫星。新一代卫星将拥有更强的发电能力，并展示卫星之间的光学通信链路技术。
Google设想中的太空数据中心方案与当前行业其他项目有所不同。其核心思路是在轨卫星之间建立高速互联网络，使多颗卫星能够协同处理同一工作负载，从而形成具备分布式计算能力的轨道计算平台。相比之下，虽然部分竞争方案同样计划采用激光通信，但卫星间距离较远，难以实现高效协同计算。
随着人工智能对算力和能源需求持续增长，Google正尝试将数据中心从地面扩展至太空。此次Suncatcher卫星成功运行，不仅验证了相关概念的可行性，也为未来建设真正的轨道AI计算基础设施迈出了关键一步。
本文由 AI宝库自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Google's Suncacher (Suncacher) prototype space data centre satellite is officially operational and has successfully established communications with the ground. The launch earlier this month of the SpaceX Falcon 9 rocket into orbit with more than 100 spacecraft marked an important step by Google in extending artificial intelligence computing capabilities to the space environment.
The satellite carries the GoogleTensor Production Unit (TPU) artificial intelligence acceleration chip and can generate electricity up to about 1,000 watts. The objective of the project is to pave the way for a future “orbit data centre network” consisting of several close-range coordinated flight satellites, so that the large AI reasoning mission can be carried out directly in space.
However, current technology still faces many challenges. Owing to the high cost of TPU, satellites can only start chips for about 20 minutes at a time, making it more difficult to communicate on a complex basis. After approximately three weeks of in-orbit testing and system calibration, the satellite finally entered the formal operational phase.
In order to verify the functioning of the system, Google Engineers sent the first alert to the satellite this week, requiring the Google DeepMind Gemma 4 model pre-loaded on the satellite to create a sentence on space. The model says, "The blue planet is still and shining below, data flows are fine, and stars feed my core."
In addition to having AI write poetry in space, Google uses the model to collect and report information on the orbit environment around the satellite. The Google spokesperson said that the satellite was currently in a normal state of operation and that the systems were working as expected.
Next, Google will focus on assessing the reliability of the TPU chips in extreme space environments, including thermal stress from temperature changes and the effects of cosmic radiation on hardware. Although Google had previously conducted extensive tests of TPU in the ground particle accelerator, the long-term performance of the real space environment required further validation.
One of the most significant concerns is the phenomenon of “bit flips” caused by radiation, i.e., accidental changes in the data slots in the computer memory, which are active in high-energy particles. This may result in data damage, which is not necessarily detected in a timely manner by the regular monitoring system.
Previous ground experiments have shown that radiation has a relatively limited impact on the AI reasoning mission, but may pose more obvious problems in modelling training. Today, Google is finally able to validate these predictions through actual in-orbiting data rather than relying solely on laboratory simulations.
If the mission proves that TPU is capable of operating steadily in space, Google plans to launch two more follow-on satellites with more advanced chips in 2027. The new generation of satellites will have greater power generation capacity and will demonstrate optical communication links between satellites.
The space data centre programme envisaged by Google differs from other projects in the current industry. The core idea is to establish a high-speed Internet network between orbital satellites, enabling multiple satellites to work together on the same task load, thus forming an orbit computing platform with distributed computing capabilities. In contrast, while some competition programmes also plan to use laser communications, the distance between satellites makes it difficult to achieve efficient and synergistic calculations.
Google is trying to expand the data centre from the ground to space, as the demand for artificial intelligence for algorithms and energy continues to grow. The successful operation of the Suncacher satellite had not only validated the feasibility of the concept but had also taken a crucial step towards the future establishment of a real orbital AI computing infrastructure.
Auto-fetched and translated by AI Gem Box. All rights belong to the original author.