Google Boots TPU-Equipped AI Satellite to Test In-Orbit Inference
Google’s "Suncatcher" prototype satellite is now operational in low Earth orbit, testing whether Tensor Processing Units can handle inference tasks like poetry generation and environmental reporting despite strict power and radiation constraints.
Google has brought its Suncatcher prototype satellite online, marking the company's first foray into space-based data center operations. According to TechCrunch, the spacecraft, which launched earlier this month aboard a SpaceX Falcon 9 rocket, has successfully completed a three-week commissioning process. The satellite is currently executing inference tasks using Google DeepMind’s Gemma 4 model. While the initial demonstrations involved generating poetry and reporting on orbital conditions, the primary engineering objective is to determine if custom Tensor Processing Units can reliably function in the harsh environment of low Earth orbit.
The mission highlights the significant power limitations inherent in orbital computing. Although the satellite is equipped with four TPUs, it generates only 1 kilowatt of electricity, forcing engineers to operate only one chip at a time. TechCrunch reports that these operational windows are restricted to roughly 20 minutes, a constraint that complicates the already challenging task of space-to-ground communication. These power budgets are a critical hurdle for the project, which aims to eventually establish networks of close-flying satellites capable of performing large-scale inference workloads.
A central goal of the Suncatcher campaign is to gather real-world data on how radiation affects processor reliability. While ground-based testing in particle accelerators suggested that radiation would have a minimal impact on inference workloads, the team is now verifying these findings in orbit. Engineers are specifically monitoring for "bit flips"—errors where stored values in computer memory are altered by radiation, potentially leading to data corruption without triggering standard system alerts. This real-world validation is essential before the project can scale to more complex, high-stakes processing tasks.
Looking ahead, the Suncatcher project has set its sights on 2027 for the next phase of development. Google plans to launch two additional satellites that will feature improved power generation capabilities and demonstrate inter-satellite optical links. This architecture is intended to support coherent, distributed workloads, distinguishing Suncatcher from other industry visions that rely on individual satellites performing isolated inference. The success of these future deployments remains tied to broader infrastructure goals, including the development of SpaceX’s Starship rocket, which is necessary for the economical deployment of such orbital assets.
The project arrives during a period of financial scrutiny for Alphabet, Google’s parent company. TechCrunch reports that Alphabet’s stock recently declined following disclosures that efforts to integrate AI into search products are negatively impacting ad revenue growth. Suncatcher is framed as a long-term "moonshot" investment, with infrastructure designs optimized for economic viability in the 2030s. Whether the project can transition from a proof-of-concept to a functional distributed data center network will depend on both the technical performance of its orbital hardware and the company's ability to navigate current financial pressures.