Google’s “Project Suncatcher” is not just sending any computer to space; it’s sending its own custom-designed AI brains. The new research “envisions compact constellations… carrying Google TPUs,” a move that is central to the entire “moonshot” strategy.
TPUs, or Tensor Processing Units, are Google’s proprietary processors, “optimised for training and the day-to-day use of AI models.” By using its own chips, Google can create a highly efficient, vertically integrated system, just as it does in its terrestrial datacenters.
This is a strategic choice. While Nvidia’s chips are being launched by a partner (Starcloud), Google is betting on its own hardware. This allows it to fine-tune the entire system, from the 8x-efficient solar panels to the optical-link networking, all around the specific power and processing needs of its TPUs.
This hardware-specific plan, however, also raises the stakes. These TPUs have never been tested in the harsh, high-radiation environment of space. Solving the “thermal management” for these specific high-performance chips will be a “significant engineering challenge.”
The 2027 prototypes will be the first-ever test of Google’s TPUs in orbit. Their survival and performance will be a critical “first milestone” in proving that Google’s AI-optimized hardware has a future beyond the confines of Earth.