Google Builds Solar-Powered Space AI Data Center

Google has revealed plans to deploy an artificial intelligence (AI) data center in space to harness massive energy through solar power.

Google announced on the 4th, local time, its ‘Project Suncatcher’ to build a data center that directly uses solar energy in space. As AI demand surges, the power consumption of data centers performing computations is also growing exponentially. Google stated, “The sun emits energy exceeding 100 trillion times the entire human power production,” and added, “In the future, space could become a suitable place to expand AI computing.”

The basic plan involves deploying multiple small satellites equipped with large solar panels and Google’s own AI chip, the Tensor Processing Unit, TPU, into low Earth orbit. Google said, “Solar panels in the appropriate orbit can produce energy up to 8 times more efficiently than on Earth.” The satellites will form an AI computing cluster through optical communication.

The key challenge for this is increasing the communication speed between satellites. This is because hundreds to thousands of TPUs need to exchange their computational results in real time. Google has adopted a method to increase communication speed by flying the satellites in extremely close formation, with distances between them measured in hundreds to thousands of meters. Google’s research team announced that they achieved a bidirectional speed of 1.6 Tbps (1.6 terabits per second) in ground experiments. To realize a space cluster, they need to achieve at least 10 Tbps. Additionally, Google’s developed Trillium TPU maintained operation without critical damage even in the space radiation environment corresponding to a 5-year mission.

Google announced plans to collaborate with private space company Planet Labs to launch two prototype satellites by early 2027.