Google's AI Chips Blast Off into Space in Ambitious Project

Google has taken a significant step into space by launching its Tensor Processing Units (TPUs) aboard SpaceX's Transporter-18 mission, marking their first trip into orbit. The project, code-named Suncatcher, aims to test whether Google's custom chips can survive the harsh space environment and eventually be used for low-Earth orbit applications.

The TPUs were designed to withstand launch forces, radiation, and extreme heat, and the prototype is primarily intended to determine their feasibility in space. The satellite was built in partnership with Planet Labs PBC, which Google enlisted to develop the spacecraft platform.

Radiation, cooling, and networking are key challenges that Google must overcome for its space-based AI ambitions. Tests have shown that the TPUs can withstand radiation beyond what they would face during a five-year mission, but the attached high-bandwidth memory was more sensitive.

Google's ultimate vision is to create satellite clusters linked by high-bandwidth optical connections, which could potentially address one of the biggest constraints facing AI data centers: power. The company estimates that if launch costs fall to around $200 per kilogram, the economics of powering computing infrastructure in space could approach those of terrestrial data centers.