Google's audacious plan: AI data centers in space
Google has unveiled a groundbreaking concept to address the resource constraints of energy-intensive AI data centers on Earth: launching AI chips into space on solar-powered satellites. This ambitious project, dubbed Project Suncatcher, aims to create space-based data centers that can harness solar power 24/7. By tapping into an almost limitless source of clean energy, Google envisions a future where its AI ambitions can soar without the environmental and financial concerns associated with terrestrial data centers.
In a blog post, Travis Beals, a Google senior director, suggests that space might be the ideal location to scale AI computing. The company has also released a preprint paper detailing its progress, though it awaits academic peer review. Google's vision involves Tensor Processing Units (TPUs) orbiting Earth on satellites equipped with solar panels, generating electricity nearly continuously and eight times more efficiently than those on Earth. This setup could revolutionize AI infrastructure.
However, Google acknowledges significant challenges. Effective communication between satellites is crucial, requiring links that support tens of terabits per second. Achieving this involves maneuvering satellite constellations into tight formations, flying satellites within kilometers of each other. This approach, while innovative, faces the growing risk of space junk from collisions. Additionally, Google must ensure its TPUs can withstand higher radiation levels in space, a concern addressed through radiation tolerance testing.
The cost of launching TPUs into space is currently high, but Google's analysis suggests that by the mid-2030s, the costs of space-based data centers could become comparable to those on Earth on a per-kilowatt/year basis. To test this, Google plans a joint mission with Planet to launch prototype satellites by 2027, marking a significant step towards this futuristic vision.
