Google Launches First AI Chips to Orbit for Space Data Centers

Four Trillium Tensor Processing Units were deployed into a sun-synchronous path on October 1 to test whether the hardware can maintain AI workloads without terrestrial cooling systems. Launched by SpaceX from Vandenberg, California, the refrigerator-sized prototype is part of 'Project Suncatcher' and runs Gemini inference in 15 minute bursts to prevent overheating in the vacuum of space. Google is partnering with Planet Labs on the Transporter-18 rideshare mission to evaluate the impact of cosmic radiation and physical stress on its custom AI accelerators.

The company aims to build scalable machine learning infrastructure in space to bypass land and electricity constraints, leveraging orbital solar panels that provide 8x more energy than ground based ones. Next year, Google plans to launch two additional satellites to test laser links capable of 800 Gbps for potential clusters of 81 satellites. Google estimates economic viability for a 1 GW orbital cluster may depend on launch costs falling to $200/kg by 2035 to approach the cost of terrestrial data centers.