Google is taking Project Suncatcher out of the lab and into the real world with an orbital test satellite prototype launching next week aboard SpaceX’s Transporter-18 rideshare mission.
Developed with space company Planet, the satellite will take Google Tensor Processing Units (TPUs) into low Earth orbit. The first mission will test the reliability of the AI hardware for the launch environment and space conditions.
The test will gather information on how the chips withstand radiation, extreme temperatures and the physical stresses of getting to orbit, Google said. The company is also exploring how its cooling system can maintain safe operating temperatures for high-performance AI hardware in a vacuum.
One of the key engineering challenges for space based computing is cooling. Google’s solution is to use a combination of heat pipes and radiators to remove heat from the TPUs, a combination that won’t work in a terrestrial data center with conventional airflow.
Another problem is radiation. High-energy particles can disrupt electronic systems and lead to errors called bit-flips. Google has already tested radiation and AI workloads on the ground, but claims an orbital mission is required to observe performance of the hardware in the actual space environment.
This is not the only satellite in the Suncatcher concept. Google’s vision is to have arrays of solar-powered spacecraft with TPUs and high-bandwidth optical connections between them. According to its research, satellites in the right orbit can be in the sun for significantly longer than solar systems on Earth, meaning that they could be a source of energy for future AI infrastructure.
Google and Planet earlier announced that they will launch two prototype satellites in early 2027. The next phase will be high-bandwidth laser communications between spacecraft, a crucial feature if multiple satellites are to be employed in the future to perform larger AI workloads.
However, the first mission is not intended to be a proof of a space-based data center. Google says it’s a learning experience to find technical issues and collect data that can inform future missions.
Google is also playing in a new space-computing arena that already has others. In November 2025, Starcloud says it deployed its Starcloud-1 satellite with an NVIDIA H100 GPU on board, which was then used to showcase AI workloads in space.
Part of the promise of bringing AI infrastructure into space is that it will be long-term. AI systems need lots of computing power and electricity, and LEO can offer a lot of access to sunlight. Google’s work thus focuses on whether satellite networks can be used as a new means to scale machine-learning infrastructure in the future.
Currently, Suncatcher is a research project and not a commercial data-center service. Google’s first orbital test will be the first real-world data it has on the performance of its TPU technology outside the planet.
