Google has launched a satellite carrying a prototype designed to test the viability of operating artificial intelligence (AI) chips in orbit, marking an early step toward establishing space-based data centers. The satellite was deployed by SpaceX on Thursday from California at approximately 11:30 a.m. local time as part of a year-long experiment to evaluate the performance of Google’s Tensor Processing Units (TPUs) in the harsh conditions of space.

The project, known as Suncatcher, involves a refrigerator-sized satellite developed in collaboration with Planet Labs. It is equipped with solar panels and placed into a sun-synchronous orbit to ensure near-continuous exposure to sunlight. This orbit allows the solar panels to potentially generate up to eight times more energy than those on Earth, where atmospheric conditions, weather, and the day-night cycle limit solar power availability. The satellite’s solar arrays will energize four TPUs, which will be used to run large language models and respond to simple queries.

Travis Beals, senior director of Paradigms of Intelligence at Google’s AI research unit, emphasized the importance of real-world testing after the chips underwent ground-based simulations designed to mimic space conditions. “No amount of testing is as good as the real thing,” he said. Google aims to keep the TPUs operational in orbit for about a year to gather data on their durability and function.

Google is among the first companies to deploy AI processing hardware in space. Other players, including SpaceX and Blue Origin, have announced plans for similar ventures. Industry experts see space-based data centers as particularly valuable for specialized applications, such as enabling governments to store sovereign data and allowing companies to process satellite imagery in real time without delays inherent to ground-based systems.

Proponents of orbital data centers also point to potential advantages, including avoiding terrestrial constraints like land availability, local opposition, and limited energy sources. However, significant technical and economic challenges remain. The equipment must withstand extreme forces during launch—up to 100 times Earth’s gravity—as well as continuous exposure to cosmic rays and solar radiation once in orbit. Thermal management is another critical issue because traditional Earth-based cooling techniques are ineffective in space. Instead, heat generated by the chips must be radiated as infrared light into deep space, requiring radiators potentially much larger than the chips themselves.

Val Elbert, global leader of Boston Consulting Group’s technology, media, and telecommunications practice, highlighted that future satellites could require radiators comparable in size to a Boeing 747. He also estimated that approximately 10,000 interconnected satellites would be needed to assemble a data center with 1-gigawatt capacity in space, a configuration that could cost in the vicinity of $30 billion based on current launch costs.

While technological advances in launch services may reduce expenses over time, Elbert cautioned that Earth-based data centers will likely retain a cost advantage for the foreseeable future. Google remains optimistic about narrowing this gap, projecting that space-based facilities could eventually become as economical as their terrestrial counterparts, although such progress may take years.

For now, Google is focused on monitoring the prototype satellite’s performance as it orbits Earth, with initial assessment expected within weeks. The company plans to expand its experiment in 2027 by deploying TPUs on two interconnected satellites, continuing its exploration of how AI workloads might be managed beyond the planet.