Google has launched a refrigerator-sized satellite equipped with AI processing chips as part of Project Suncatcher, marking the first in-orbit test of its Tensor Processing Units (TPUs) designed for machine learning. The satellite, deployed via a SpaceX Falcon 9 rocket, will operate for 15-minute intervals to assess how the chips handle the radiation, thermal extremes, and physical stress of spaceflight.
The mission aims to evaluate whether solar-powered AI computing infrastructure can function reliably in orbit, where solar panels will remain exposed to sunlight continuously in a sun-synchronous orbit. This eliminates the need for heavy batteries or backup power systems. The prototype carries a version of Google’s Gemma AI model, running on open-weight architecture to answer simple queries.
Google’s Sundar Pichai previously stated that demand for AI services exceeds supply, with capital expenditures projected to reach $180 billion to $190 billion in 2026—more than six times the 2022 figure—as the company expands computing infrastructure. The project reflects growing industry interest in space-based data centers as an alternative to Earth-based facilities, which face opposition over energy consumption and environmental impact.
Project Suncatcher’s lead, Travis Beals, emphasized the potential of solar power in space, noting that the sun provides the vast majority of energy in the solar system. The satellite’s design leverages this by ensuring near-constant exposure to sunlight, reducing reliance on stored power.
The launch, part of SpaceX’s Transporter-18 mission, also carried Planet Labs satellites, including a solar-powered prototype. Google and SpaceX have a close partnership, with Alphabet holding a $82 billion stake in SpaceX following its record IPO in June 2026. While their AI divisions compete, the companies collaborate on infrastructure development.
The test follows Google’s November 2025 announcement of Project Suncatcher, positioning it as a “moonshot” initiative to explore scalable machine learning infrastructure in orbit. If successful, the project could redefine AI computing by tapping into unlimited solar energy and avoiding terrestrial constraints.