Google is preparing to launch a prototype satellite as part of its research project, Project Suncatcher, with the aim of collecting direct data on the performance of Tensor Processing Units (TPUs) in space. The first test focuses on the hardware’s ability to withstand the severe vibrations during launch, radiation, and the thermal challenges imposed by the low Earth orbit environment.
Google is carrying out the mission aboard SpaceX’s Transporter-18 rideshare flight, in partnership with Planet. The satellite is not a fully operational orbital data center, but rather a test platform that will help the team determine what works and what might fail before moving to broader stages.
Testing Hardware Under Launch and Space Conditions
The journey to low Earth orbit takes approximately 10 minutes, but it imposes vibrations and accelerations that may reach 10 times the force of gravity, while individual components, including TPU chips, may be subjected to forces ranging from 50 to 100 times gravity. For this reason, Google tested the satellite by shaking it along all three axes to simulate launch frequencies, and said that the hardware withstood the test.
The team also exposed TPU chips to a proton beam at the University of California, Davis’s Crocker Nuclear Laboratory, while running AI workloads and monitoring errors such as bit flips. Initial results indicate that Trillium TPU chips withstood a total ionizing radiation dose exceeding what Google expects them to encounter during a five-year space mission, although the orbital test will remain necessary to verify performance in the actual environment.
Cooling Is the Biggest Engineering Challenge
TPU chips produce a large amount of heat within a small area. On Earth, airflow helps dissipate it, but the vacuum provides no convective thermal load, requiring reliance on thermal radiators and other heat-transfer methods. Google is currently testing a design that combines heat pipes and radiators inside a chamber simulating vacuum and thermal conditions in space. The first mission will be used to evaluate and improve this system.
From an Experimental Satellite to Processing Clusters
In the long term, Project Suncatcher is exploring the possibility of building scalable infrastructure for machine-learning processing in orbit. Google says that satellites in low Earth orbit could receive nearly continuous sunlight, enabling solar power generation of up to eight times what is available on Earth, according to the comparison presented in the material.
Future concepts include clusters carrying dozens of TPU chips on each satellite, with satellites linked via laser to provide the bandwidth required for AI workloads. However, this requires extremely high precision in directing communications because the satellites are constantly moving, and because the proposed system needs high bandwidth over short distances. Google plans to test this aspect when it places two satellites in orbit in 2027.
Why Does This Test Matter?
The actual change here is not the launch of a commercial product, but the shift of the question of operating AI chips in space from theoretical consideration to field measurement. The mission’s results will determine whether resistance to radiation and vibration is sufficient, and whether passive cooling can protect the chips, while also revealing the constraints that could prevent scaling. The practical feasibility of building orbital processing infrastructure remains open, as Google did not provide estimates for cost or capacity, or a timeline for operating a commercial service, in the material.