Google launches orbital chip prototype to test space-based data center feasibility

Google has launched a prototype satellite carrying an advanced Tensor Processing Unit (TPU) into orbit aboard a SpaceX rocket. This test marks the beginning of Project Suncatcher, a long-term research initiative exploring the potential for deploying large-scale AI compute clusters in space.

Key details

• •According to TechCrunch, the prototype satellite was built by Planet Labs to test if Google's TPU can maintain cooling and continuous power in an orbital environment.
• •Project Suncatcher lead Travis Beals stated that while ground testing is rigorous, the current mission is necessary to observe how the hardware performs under real-world orbital conditions.
• •Google envisions a future network of 81 satellites flying in close formation to handle parallel processing for advanced AI workloads, as reported by TechCrunch.
• •The companies plan a follow-up demonstration next year featuring two satellites capable of communicating via laser links to run more substantial workloads, according to TechCrunch.
• •NPR Topics reports that the current satellite is approximately the size of a refrigerator and aims to harness solar energy alongside processing AI tasks.

Background

Google recently released a peer-reviewed white paper in the journal Joule detailing the technical challenges of moving compute infrastructure to orbit. The company notes that current orbital data center designs are limited by the availability of cost-effective launch technology.

Why it matters

Google executive Travis Beals stated that achieving the necessary scale for orbital data centers would require approximately 1,600 launches by SpaceX’s Starship. The project focuses on preparing for future AI workload demands, with Google identifying high-speed bandwidth and low latency between chips as critical hurdles for multi-rack operations in space.