Google's Ambitious Plan: Building an AI Data Center in Space

Описание: Google is taking a giant leap in technology by planning to establish an AI-based data center in space, independent of Earth's power resources. Through the 'Project Suncatcher' research initiative, Google aims to place its AI chips on solar-powered satellites, revolutionizing energy constraints for AI data centers on Earth.
This groundbreaking move, as reported by detikInet, addresses the energy limitations of terrestrial AI data centers. By positioning AI chips in orbit, Google hopes to harness the unlimited solar energy available for 24 hours a day. Travis Beals, Google's Senior Director, envisions space as the ideal future location for scaling AI computing, as quoted in his blog post.
In the right orbit, solar panels can be eight times more productive than on Earth, generating nearly continuous electricity and reducing battery requirements. Google's Project Suncatcher will send Tensor Processing Units (TPUs), its internally developed AI chips, to satellites equipped with solar panels for operation in low Earth orbit. The goal is to create an orbiting AI data center that operates 24/7 with clean energy from sunlight.
Google has already tested its chips for radiation tolerance, finding that they can withstand a five-year mission without permanent failure. Another challenge is ensuring all satellites can communicate with each other, requiring high-speed data connections of tens of terabits per second with low latency, which is difficult to achieve in space due to the increased power needed for long-distance data transmission.
To address this, Google may need to position the satellites in a tighter formation, possibly within a few kilometers of each other, which would also reduce the required satellite positioning maneuvers. Launching a data center into space will be costly, but Google's analysis shows that the project is comparable to Earth-based data centers in terms of power efficiency by the mid-2030s.
Google plans to launch a joint mission with Planet in 2027, experimenting with satellite prototypes to test how AI models and TPUs operate in space.