Google to Launch First AI Chips Into Orbit Next Week

Google is preparing to launch its first dedicated artificial intelligence chips into space next week, marking a significant milestone for Project Suncatcher. The mission involves sending a prototype satellite equipped with Tensor Processing Units (TPUs) aboard SpaceX’s Transporter-18 rideshare vehicle. This flight serves as the initial test of Google’s vision to move AI computation out of terrestrial data centers and into low Earth orbit.

The satellite was constructed in collaboration with Planet, a prominent Earth-imaging firm. The primary objective of this inaugural journey is to evaluate how these specialized processors withstand the harsh conditions of spaceflight. Engineers need to determine if the hardware can survive the intense vibrations of launch, the relentless bombardment of cosmic radiation, and the extreme temperature fluctuations between direct sunlight and shadow.

Announced last November, Project Suncatcher investigates the feasibility of powering large-scale AI operations using fleets of satellites. Google highlights that solar panels positioned in low Earth orbit can harvest up to eight times more energy than those installed on the ground. This abundance of power could theoretically support the massive energy demands of modern AI models without relying on terrestrial infrastructure.

Before attempting the orbital journey, the team subjected the hardware to rigorous ground-based simulations. A typical launch profile subjects components to forces ranging from 50 to 100 times standard gravity. To replicate these stresses, engineers shook the satellite violently along all three axes. The hardware remained intact and functional throughout the testing phase, demonstrating structural resilience against launch-induced trauma.

Radiation resistance was verified at UC Davis’s Crocker Nuclear Laboratory. Researchers placed the Trillium TPUs inside a proton beam to simulate years of exposure. According to Google, the chips successfully endured a total radiation dose exceeding what they would accumulate over a five-year mission lifespan.

Thermal management presents another unique challenge in the vacuum of space, where air cannot facilitate cooling. Heat must dissipate entirely through radiators. Google is currently testing a hybrid system combining heat pipes and radiators. Early tests conducted in Earth-based vacuum chambers have shown promising results, suggesting the thermal design is viable for orbital deployment.

Looking ahead, the project will expand in 2027. During this phase, two satellites will test laser communication links, which are essential for connecting clusters of TPU-equipped spacecraft. Google has indicated that this subsequent test will be a joint effort with Planet, further solidifying their partnership in developing space-based computing architectures.

The race to establish an in-orbit AI economy is intensifying. Nvidia-backed startup Starcloud previously launched an H100 chip into space last November and has already utilized it to run Google’s Gemma model. However, economic viability remains a critical hurdle. Google’s internal research suggests that launch costs must drop to approximately $200 per kilogram for space-based computing to become cost-competitive with traditional ground-based data centers.