Google is preparing to deploy its artificial intelligence processors beyond Earth's atmosphere for the first time, marking a crucial milestone for Project Suncatcher. The experimental satellite, developed in partnership with Planet, will ride SpaceX's Transporter-18 rideshare vehicle into low Earth orbit to evaluate whether the company's Tensor Processing Units can endure the extreme conditions of spaceflight.

The mission will subject Google's custom chips to real-world testing across three critical challenges: the violent forces of launch, the radiation environment of space, and temperature swings between extreme heat and cold. Project Suncatcher, unveiled in November of last year, investigates the feasibility of running AI workloads on satellite constellations. According to Google's analysis, solar panels positioned in low Earth orbit can harvest roughly eight times more energy than equivalent installations on the ground.

Ground Testing and Preparations

Before reaching orbit, the hardware underwent rigorous laboratory validation. Launch forces subject spacecraft to accelerations between 50 and 100 times Earth's gravity over approximately ten minutes. Google's team replicated these conditions by vibrating the satellite across all three axes, and the equipment performed successfully.

Radiation exposure presented another testing challenge. Google ran AI computations on its Trillium TPUs inside a proton beam facility at UC Davis's Crocker Nuclear Laboratory. The chips absorbed a radiation dose exceeding what they would experience across a full five-year orbital mission, demonstrating sufficient resilience.

Thermal management in the vacuum of space requires innovative solutions, since conventional air-based cooling becomes impossible. Google designed a system combining heat pipes and radiators, which demonstrated effectiveness during vacuum chamber testing on Earth.

Next Steps and Competition

The roadmap extends to 2027, when two additional satellites will validate laser communication links connecting clusters of TPU-equipped spacecraft. Google has indicated this subsequent test will involve Planet as a collaborative partner.

Google faces competition in this emerging field. Starcloud, a startup backed by Nvidia, successfully launched an H100 processor into orbit in November and has already executed inference runs using Google's Gemma model in space.

Economics remain the primary constraint on widespread adoption. Google's own analysis indicates that launch costs would need to decline to approximately $200 per kilogram for space-based computing infrastructure to achieve cost parity with terrestrial alternatives.

Source: The Next Web