- Project Suncatcher envisions solar-powered satellites hosting AI data centres in space to reduce Earth’s energy and water consumption used for cooling terrestrial facilities.
- These orbiting data centres will use free-space optical links to transfer data at speeds of tens of terabits per second, forming a distributed network similar to satellite internet constellations like Starlink.
- Google plans to launch two prototype satellites by early 2027 to test its space-based TPU (Tensor Processing Unit) hardware.
- Rationale for Space-Based AI Centres:
- AI data centres consume enormous electricity and water for cooling — a growing environmental concern as AI adoption increases globally.
- In space, solar panels are up to 8 times more efficient than on Earth, providing a continuous and clean energy supply.
- How It Works:
- The satellites will rely on solar energy and optical data transmission to function as a cohesive high-performance computing network.
- TPUs (Trillium v6e) are being tested for radiation resistance and performance in extreme conditions.
Key features:
- Solar-Powered Satellite Constellation: Uses solar panels up to 8 times more efficient in orbit than on Earth.
- Orbiting TPUs: AI accelerators (Trillium v6e) tested under radiation for space durability.
- High-Speed Optical Links: Free-space optical communication capable of tens of terabits per second, connecting satellite nodes.
- Prototype Launch: Two test satellites planned for early 2027 to validate hardware and communication systems.
- Scalability: Analytical models suggest satellites can operate just hundreds of meters apart, allowing clustered space-based data hubs.
- Future Cost Efficiency: By mid-2030s, falling launch costs (as low as $200/kg) could make orbital data centres economically viable.