Google is preparing to put its AI hardware through an orbital test with Project Suncatcher, a prototype set to launch to low Earth orbit on SpaceX’s Transporter-18 mission. The immediate goal is not an operational space-based data center, but to establish whether Google’s tensor processing units, or TPUs, can remain functional in the conditions needed to build one.
The project is part of Google’s longer-term concept for solar-powered AI infrastructure in space. Google outlined the initiative in November 2025, proposing constellations of satellites connected through optical links. It had previously targeted two orbital prototypes by early 2027; this first test satellite brings a hardware validation step forward.
- Launch resilience: the roughly 10-minute trip to low Earth orbit can expose a satellite to acceleration of up to 10 g, while individual components may face loads of 50 g to 100 g. Google says it has conducted three-axis vibration testing in the lab.
- Radiation tolerance: Google exposed Trillium TPUs running AI workloads to protons at the UC Davis Crocker Nuclear Laboratory. The company says the chips tolerated a total ionizing-radiation dose beyond that expected over a five-year orbital mission.
- Thermal control: without air, spacecraft cannot use convection to remove heat. Google is testing heat pipes and radiators intended to move heat away from TPUs and radiate it into space.
Solar power is a central attraction of the concept. Google estimates that a solar panel in a suitable orbit could generate up to eight times more energy than an equivalent panel on Earth. But continuous power alone does not solve the engineering problem: cooling high-performance processors in a vacuum is among the key questions the flight is expected to address.
Google’s proposed end-state architecture would place dozens of TPUs on each satellite and combine multiple spacecraft into a compute cluster. Its next planned milestone, in 2027, is to launch two satellites designed to communicate through high-speed laser links. Maintaining those links between rapidly moving spacecraft will require extremely precise pointing, another major hurdle before an orbital AI cluster can become practical.
Transporter-18 will therefore test a foundational premise rather than deploy a conventional data center: whether powerful AI processors can operate reliably in orbit for long enough to justify the far more complex satellite network that would follow.
